OCTOBER 2018 AJNR VOLUME 39 • PP 1775–1962 OCTOBER 2018 THE JOURNAL OF DIAGNOSTIC AND VOLUME 39 INTERVENTIONAL NEURORADIOLOGY NUMBER 10 WWW.AJNR.ORG Deep learning in neuroradiology Resources and payment methods: Canada vs US Reperfusion failure causes in stent-retriever thrombectomy

Official Journal ASNR • ASFNR • ASHNR • ASPNR • ASSR ® Aneurysm Therapy LVIS Solutions Intraluminal Support Device

The first and only stent with Premarket Approval (PMA) for use in stent-assisted coil embolization

Stent Deployment. Refined.

Braided Coil Assist Stents with High Neck Coverage, Excellent Visibility and Improved Conformability*

Low-profile Visualized Intraluminal Support

INDICATIONS FOR USE:

The LVIS® and LVIS® Jr. devices are indicated for use with neurovascular embolization coils in patients ≥ 18 years of age for the treatment of wide-neck (neck width ≥ 4 mm or dome to neck ratio < 2) saccular intracranial aneurysms arising from a parent vessel with a diameter ≥ 2.0 mm and ≤ 4.5 mm. Rx Only: Federal (USA) law restricts this device to sale by or on the order of a physician.

The HydroCoil® Embolic System (HES) and MicroPlex® Coil System (MCS) are intended for the endovascular embolization of intracranial aneurysms and other neurovascular abnormalities such as arteriovenous malformations and arteriovenous fistulae. The HES and MCS are also intended for vascular occlusion of blood vessels within the neurovascular system to permanently obstruct blood flow to an aneurysm or other vascular malformation and for arterial and venous embolizations in the peripheral vasculature. The device should only be used by physicians who have undergone pre-clinical training in all aspects of HES/MCS procedures as prescribed by MicroVention. A Complete Coil Portfolio Hydrogel Embolic Coils & Platinum Coils

MicroVention’s comprehensive portfolio features clinically proven Hydrogel coils, which can be used exclusively or in combination with our trusted Platinum coils to treat a wide range of aneurysms and neurovascular lesions.

Breakthrough Hydrogel Technology

• Less Recurrence • Less Retreatment • More Progressive Occlusion Compared to platinum coils with comparable safety1

RER FEERENCESE : 1.. TTaschner eet al.. Seconco d-Generattion Hydrooggel Coils for the Endovascular Treaeatmemenenntto of Intracrananial AAneurryysmm; A RaR nndod mizizi edd CControllelledTd Tririaial.i l 2018;49:000-0-000. DOII:100.116161// STROKOKEAHAHA.117.0187077

For more information or a product demonstration, MicroVention Worldwide contact your local MicroVention representative: Innovation Center PH +1.714.247.8000 35 Enterprise Aliso Viejo, CA 92656 USA MicroVention UK Limited PH +44 (0) 191 258 6777 MicroVention Europe, S.A.R.L. PH +33 (1) 39 21 77 46 MicroVention Deutschland GmbH PH +49 211 210 798-0 microvention.com

MICROVENTION, MicroPlex and HydroCoil are registered trademarks of MicroVention, Inc. • Refer to Instructions for Use, contraindications and warnings for additional information. Federal (USA) law restricts this device for sale by or on the order of a physician. ©2018 MicroVention, Inc. Jan. 2018 FLOW-DEPENDENT MICROCATHETER

MAGIC catheters are designed for general intravascular use. They may be used for the controlled, selective regional infusion of therapeutic agents or embolic materials into vessels.1 1. Magic Catheters IFU – Ind 19 MKTG-068 Rev. A 18 Technology Drive #169, Irvine Ca 92618 P 949.788.1443 | F 949.788.1444 A Company Now you have 24 hours to make a lifetime of difference in stroke patients like Nora

The Trevo Retriever is the only device cleared to reduce disability in stroke patients up to 24 hours from time last seen well.

For more information, visit strykerneurovascular.com/trevo24hours

Copyright © 2018 Stryker AP002078 v1.0 MAY 18-23, 2019 | BOSTON

ASNR 57th Annual Meeting & The Foundation of the ASNR Symposium 2019 CALL FOR ABSTRACTS

You are invited! Join us next year in Boston to present the best scientific research in neuroradiology. Submit your abstract online at 2019.asnr.org Deadline: 5:00pm (PST), Wednesday, November 7, 2018

Submit an abstract for the ASNR 57th Annual Meeting (May 18-23, 2019, Boston, MA) in one of the following presentation categories: • Scientific Paper ...... Oral • Electronic Educational Exhibit ...... Electronic • Scientific Poster ...... Print • Educational Exhibit Poster ...... Print • Electronic Scientific Poster ...... Electronic • Excerpta ...... Oral This activity has been approved for AMA PRA Category 1 Credits™.

Abstract Submissions Information and Criteria

1. Abstracts should describe the learning objectives 6. Submit each abstract in one category only. of the presentation. 7. Submission topic areas include: Adult Brain, 2. The American Journal of Neuroradiology (AJNR) Spine, Head and Neck, Pediatrics, Functional encourages presenters to submit manuscripts Imaging, Interventional, Health Policy, and AI/ based on their work to AJNR before considering Informatics. other journals. 8. Maximum length: 300 words including graphics. 3. Presenters of accepted abstracts must register Charts, graphs and tables may be included as an for the ASNR 57th Annual Meeting. image, for all categories. 4. Published or previously presented works should 9. Submission site allows uploading of files into NOT be submitted. the system. 5. Format abstract text using headings required for 10. Changes can be made to the abstract until submission category. the deadline.

Complete instructions for submissions are available online: 2019.asnr.org CALL FOR AJNR EDITORIAL

AJNRAMERICAN JOURNAL OF NEURORADIOLOGY FELLOWSHIP CANDIDATES

2019 Candidate Information and Requirements

ASNR and AJNR are pleased GOALS once again to join efforts • Increase interest in editorial and publication-related activities in younger individuals. • Increase understanding and participation in the AJNR review process. with other imaging-related • Incorporate into AJNR’s Editorial Board younger individuals who have previous experience journals that have training in the review and publication process. programs on editorial • Fill a specific need in neuroradiology not offered by other similar fellowships. • Increase the relationship between “new” generation of neuroradiologists and more aspects of publishing for established individuals. trainees or junior staff • Increase visibility of AJNR among younger neuroradiologists. (3–5 years after training), ACTIVITIES OF THE FELLOWSHIP including Radiology • Serve as Editorial Fellow for one year. This individual will be listed on the masthead as such. (Olmsted fellowship), • Review at least one manuscript per month for 12 months. Evaluate all review articles submitted to AJNR. AJR (Figley and Rogers • Learn how electronic manuscript review systems work. • Be involved in the final decision of selected manuscripts together with the Editor-in-Chief. fellowships), JACR (Bruce J. • Participate in all monthly Senior Editor telephone conference calls. Hillman fellowship), • Participate in all meetings of the Editors during the annual meetings of ASNR and RSNA and the Radiology Editors Forum as per candidate’s availability. The Foundation of the ASNR will provide and Radiologia. $2000 funding for this activity. • Evaluate progress and adjust program to specific needs in annual meeting or telephone conference with the Editor-in-Chief. • Embark on an editorial scientific or bibliometric project that will lead to the submission of an article to AJNR or another appropriate journal as determined by the Editor-in-Chief. This project will be presented by the Editorial Fellow at the ASNR annual meeting. • Serve as liaison between AJNR and ASNR’s Young Professionals Network. Participate in meetings and telephone calls with this group. Design one electronic survey/year, polling the group regarding readership attitudes and wishes. • Recruit trainees as reviewers as determined by the Editor-in-Chief. • Organize and host a Fellows’ Journal Club podcast. • Serve as Guest Editor for an issue of AJNR's News Digest with a timely topic.

QUALIFICATIONS • Be a fellow in neuroradiology from North America, including Canada (this may be extended to include other countries). • Be a junior faculty neuroradiology member (< 3 years) in either an academic or private environment. • Be an “in-training” or member of ASNR in any other category.

APPLICATION • Include a short letter of intent with statement of goals and desired research project. CV must be included. • Include a letter of recommendation from the Division Chief or fellowship program director. A statement of protected time to perform the functions outlined is desirable. • Applications will be evaluated by AJNR’s Senior Editors prior to the ASNR meeting. The name of the selected individual will be announced at the meeting. • Applications should be received by March 1, 2019 and sent to Ms. Karen Halm, AJNR Managing Editor, electronically at [email protected]. AXS Infinity LS™ Plus Long Sheath RX ONLY WARNINGS DEVICE DESCRIPTION Contents supplied STERILE using an ethylene oxide (EO) process. Do not use if sterile barrier is See package insert for complete indications, contraindications, warnings damaged. If damage is found, call your Stryker representative. and instructions for use. The AXS Vecta Aspiration System consists of the AXS Vecta 71 Aspiration Catheter, the Aspiration Tubing Set, and the VC 701 Cliq Aspirator Pump. For single use only. Do not reuse, reprocess or resterilize. Reuse, reprocessing or resterilization may INDICATIONS FOR USE compromise the structural integrity of the device and/or lead to device failure which, in turn, may The AXS Vecta 71 Aspiration Catheter is a single lumen, flexible, variable stiffness catheter. It has result in patient injury, illness or death. Reuse, reprocessing or resterilization may also create a risk The AXS Infinity LS Plus Long Sheath is indicated for the introduction of interventional devices into a radiopaque marker band on the distal end and a Luer hub at the proximal end. The AXS Vecta 71 of contamination of the device and/or cause patient infection or cross-infection, including, but not the peripheral, coronary, and neuro vasculature. Aspiration Catheter shaft has a lubricious coating at the distal end to reduce friction during use. limited to, the transmission of infectious disease(s) from one patient to another. Contamination of RX ONLY The Scout Introducer may be used in conjunction with the AXS Vecta 71 Aspiration Catheter to the device may lead to injury, illness or death of the patient. facilitate in the introduction of the AXS Vecta 71 Aspiration Catheter into distal vasculature and aid in navigation to distal anatomy. The Scout Introducer has a lubricious coating at the distal end to After use, dispose of product and packaging in accordance with hospital, administrative and/or CONTRAINDICATIONS local government policy. There are no known contraindications. reduce friction during use. The inner lumen of the AXS Vecta 71 Aspiration Catheters is compatible with the Scout Introducer, guide wires and micro catheters. The inner lumen of the Scout Introducer 1. The AXS Vecta 71 Aspiration Catheter has not been evaluated for more than one (1) clot POTENTIAL ADVERSE EVENTS is compatible with guide wires and micro catheters of an outer diameter of less than 0.044in. retrieval attempt. • Acute vessel occlusion Each package includes one AXS Vecta 71 Aspiration Catheter, one Scout Introducer, one 2. The AXS Vecta 71 Aspiration Catheter was evaluated for an average duration of direct • Air embolism hemostasis valve, and two peel-away introducers. Dimensions of the AXS Vecta 71 Aspiration aspiration of 4 minutes. • Death Catheter and Scout Introducer are included on the individual device label. The AXS Vecta 71 3. This product is intended for single use only, do not re-sterilize or reuse. Re-sterilization and/or Aspiration Catheters are available in 3 different lengths, the device configurations including the reuse may result in cross contamination and/or reduced performance. • Distal embolization length of the Scout packaged with each catheter and the recommended microcatheter length is 4. When the catheter is exposed to the vascular system, it should be manipulated while under • Emboli presented in the table below. high-quality fluoroscopic observation. Do not advance or retract the catheter if resistance is • False aneurysm formation met during manipulation; determine the cause of the resistance before proceeding. • Hematoma or hemorrhage at the puncture site INC-11129- INC-11129- INC-11129- 5. Operators should take all necessary precautions to limit x-radiation doses to patients and Catheter part number • Infection 115 125 132 themselves by using sufficient shielding, reducing fluoroscopy times, and modifying x-ray • Intracranial hemorrhage technical factors where possible. Catheter inner diameter (in) 0.071 0.071 0.071 • Ischemia Distal catheter outer diameter (in) 0.082 0.082 0.082 PRECAUTIONS • Neurological deficit including stroke Catheter working length (cm) 115 125 132 1. Store in a cool, dry, dark place. • Vessel spasm, thrombosis, dissection or perforation Scout Introducer length (cm) 133 143 150 2. Do not use kinked, damaged, or opened devices. Recommended compatible 3. Use the device prior to the “Use By” date specified on the package. WARNINGS 150 160 160 Contents supplied STERILE using an ethylene oxide (EO) process. Do not use if sterile barrier is microcatheter length (cm) 4. Exposure to temperatures above 54°C (130°F) may damage device. Do not autoclave. Recommended compatible damaged. If damage is found, call your Stryker Neurovascular representative. 0.044 max 0.044 max 0.044 max 5. Torqueing or moving the device against resistance may result in damage to the vessel or For single use only. Do not reuse, reprocess or resterilize. Reuse, reprocessing or resterilization may microcatheter outer diameter (in) device. Recommended compatible guidewire compromise the structural integrity of the device and/or lead to device failure which, in turn, may 0.038 max 0.038 max 0.038 max 6. Maintain a constant infusion of appropriate flush solution. result in patient injury, illness or death. Reuse, reprocessing or resterilization may also create a risk outer diameter (in) 7. If flow through the device becomes restricted, do not attempt to clear the lumen by infusion. of contamination of the device and/or cause patient infection or cross-infection, including, but not Remove and replace the device. limited to, the transmission of infectious disease(s) from one patient to another. Contamination of The AXS Vecta Aspiration System is recommended for use in the following vessel size ranges based on non-clinical testing: 8. Examine the device to verify functionality and to ensure that its size and shape are suitable for the device may lead to injury, illness or death of the patient. the specific procedure for which it is to be used. After use, dispose of product and packaging in accordance with hospital, administrative and/or AXS Vecta 71 Aspiration Catheter Vessel size (mm) 9. The AXS Vecta Aspiration System should be used only by physicians trained in percutaneous local government policy. procedures and/or interventional techniques. INC-11129-115 2-4 1. Do not re-sterilize or reuse, intended for single use only. Re-sterilization and/or reuse may 10. The Scout Introducer should be used with a guidewire and microcatheter inserted when in result in cross contamination and/or reduced performance. INC-11129-125 2-4 vasculature. 2. When the long sheath is exposed to the vascular system, it should be manipulated while under INC-11129-132 2-4 11. If using the AXS Vecta Aspiration System for thrombectomy, monitor the canister fluid level high-quality fluoroscopic observation. Do not advance or retract the long sheath if resistance is and replace the canister if the fill level reaches 75% of the canister volume. met during manipulation; determine the cause of the resistance before proceeding. CONTRAINDICATIONS 12. Administration of anticoagulants and antiplatelets should be suspended until 24 hours The AXS Vecta 71 Aspiration Catheter has not been evaluated for use in the coronary vasculature. post-treatment. Medical management and acute post stroke care should follow the American PRECAUTIONS Do not use automated high-pressure contrast injection equipment with the AXS Vecta 71 Stroke Association (ASA) guidelines. 1. Store in a cool, dry, dark place. Aspiration Catheter because it may damage the device. 2. Do not use kinked, damaged, or opened devices. 13. Any neurological determination should be evaluated by urgent CT scan and other evaluations POTENTIAL ADVERSE EVENTS 3. Use the device prior to the “Use By” date specified on the package. as indicated according to investigator/hospital best practice. • Acute vessel occlusion 4. Exposure to temperatures above 54°C (130°F) may damage device. Do not autoclave. 14. As in all surgical interventions, monitoring of intra-procedural blood loss is recommended so • Air embolism that appropriate management may be instituted. 5. Torquing or moving the device against resistance may result in damage to the vessel or device. • Allergic reaction and anaphylaxis from contrast media 15. Limit the usage of the AXS Vecta 71 Aspiration Catheter to arteries greater than the catheter’s 6. Maintain a constant infusion of appropriate flush solution. • Arteriovenous fistula outer diameter. 7. If flow through the device becomes restricted, do not attempt to clear the lumen by infusion. • Death 16. Excessive aspiration with the distal tip of the AXS Vecta 71 Aspiration Catheter covered by Remove and replace the device. the vessel wall may cause vessel injury. Carefully investigate location of the distal tip under • Device malfunction 8. Examine the device to verify functionality and to ensure that its size and shape are suitable for fluoroscopy prior to aspiration. • Distal embolization the specific procedure for which it is to be used. 17. There is an inherent risk with the use of angiography and fluoroscopy. • Emboli 9. The AXS Infinity LS Plus Long Sheath should be used only by physicians trained in 18. When transporting the VC-701 Cliq pump, utilize the pump handle. percutaneous procedures and/or interventional techniques. • False aneurysm formation 19. Do not use if labeling is incomplete or illegible. 10. Do not use if labeling is incomplete or illegible. • Hematoma or hemorrhage at the puncture site • Inability to completely remove thrombus AXS Vecta™ 71 Aspiration Catheter • Infection • Intracranial hemorrhage See package insert for complete indications, contraindications, warnings • Ischemia and instructions for use. • Kidney damage from contrast media INDICATIONS FOR USE • Neurological deficit including stroke The AXS Vecta™ Aspiration System, including the AXS Vecta 71 Aspiration Catheter, Aspiration • Risks associated with angiographic and fluoroscopic radiation including but not limited to: Tubing Set, and VC-701 Cliq Aspirator Pump, is indicated in the revascularization of patients with alopecia, burns ranging in severity from skin reddening to ulcers, cataracts, and delayed acute ischemic stroke secondary to intracranial large vessel occlusive disease (within the internal neoplasia carotid, middle cerebral – M1 and M2 segments, basilar, and vertebral arteries) within 8 hours of • Vessel spasm, thrombosis, dissection or perforation Stryker Neurovascular symptom onset. Patients who are ineligible for intravenous tissue plasminogen activator (IV t-PA) or who failed IV t-PA therapy are candidates for treatment. 47900 Bayside Parkway Fremont, CA 94538 Stryker Corporation or its divisions or other corporate affiliated entities own, use or have applied for the following trademarks or service marks: AXS Infinity LS, AXS Vecta, Stryker. All other trademarks are trademarks of their respective owners or holders. strykerneurovascular.com Scout is a trademark of InNeuroCo, Inc. Date of Release: SEP/2018 Copyright © 2018 Stryker AP002264 v1.0 | Page 2 of 2 EX_EN_US

Trevo® XP ProVue Retrievers Retrievers are compatible with the Abbott Vascular DOC® Guide Wire Extension (REF 22260). • If Retriever is difficult to withdraw from the vessel, do not torque Retriever. Advance Retrievers are compatible with Boston Scientific Rotating Hemostatic Valve (Ref 421242). Microcatheter distally, gently pull Retriever back into Microcatheter, and remove Retriever See package insert for complete indications, complications, warnings, and and Microcatheter as a unit. If undue resistance is met when withdrawing the Retriever into instructions for use. SPECIFIC WARNINGS FOR INDICATION 1 the Microcatheter, consider extending the Retriever using the Abbott Vascular DOC guidewire • The safety and effectiveness of the Trevo Retrievers in reducing disability has not been extension (REF 22260) so that the Microcatheter can be exchanged for a larger diameter INDICATIONS FOR USE established in patients with large core infarcts (i.e. ASPECTS ≤ 7). There may be increased risks, catheter such as a DAC® Catheter. Gently withdraw the Retriever into the larger diameter 1. The Trevo Retriever is indicated for use to restore blood flow in the neurovasculature by such as intracerebral hemorrhage, in these patients. catheter. removing thrombus for the treatment of acute ischemic stroke to reduce disability in patients • The safety and effectiveness of the Trevo Retrievers in reducing disability has not been • Administer anti-coagulation and anti-platelet medications per standard institutional guidelines. with a persistent, proximal anterior circulation, large vessel occlusion, and smaller core infarcts established or evaluated in patients with occlusions in the posterior circulation (e.g., basilar or • Users should take all necessary precautions to limit X-radiation doses to patients and who have first received intravenous tissue plasminogen activator (IV t-PA). Endovascular vertebral arteries) or for more distal occlusions in the anterior circulation. themselves by using sufficient shielding, reducing fluoroscopy times, and modifying X-ray therapy with the device should start within 6 hours of symptom onset. technical factors where possible. 2. The Trevo Retriever is intended to restore blood flow in the neurovasculature by removing SPECIFIC WARNINGS FOR INDICATION 2 thrombus in patients experiencing ischemic stroke within 8 hours of symptom onset. Patients • To reduce risk of vessel damage, take care to appropriately size Retriever to vessel diameter at PRECAUTIONS who are ineligible for intravenous tissue plasminogen activator (IV t-PA) or who fail IV t-PA intended site of deployment. • Prescription only – device restricted to use by or on order of a physician. therapy are candidates for treatment. SPECIFIC WARNINGS FOR INDICATION 3 • Store in cool, dry, dark place. 3. The Trevo Retriever is indicated for use to restore blood flow in the neurovasculature by • The safety and effectiveness of the Trevo Retrievers in reducing disability has not been • Do not use open or damaged packages. removing thrombus for the treatment of acute ischemic stroke to reduce disability in patients • Use by “Use By” date. with a persistent, proximal anterior circulation, large vessel occlusion of the internal carotid established in patients with large core infarcts (i.e., ASPECTS ≤ 7). There may be increased artery (ICA) or middle cerebral artery (MCA)-M1 segments with smaller core infarcts (0-50cc for risks, such as intracerebral hemorhage, in these patients. • Exposure to temperatures above 54°C (130°F) may damage device and accessories. Do not age < 80 years, 0-20cc for age ≥ 80 years). Endovascular therapy with the device should start • The safety and effectiveness of the Trevo Retrievers in reducing disabillity has not been autoclave. within 6-24 hours of time last seen well in patients who are ineligible for intravenous tissue established or evaluated in patients with occlusions in the posterior circulation (e.g., basilar or • Do not expose Retriever to solvents. plasminogen activator (IV t-PA) or who fail IV t-PA therapy. vertebral arteries) or for more distal occlusions in the anterior circulation. • Use Retriever in conjunction with fluoroscopic visualization and proper anti-coagulation agents. • Users should validate their imaging software analysis techniques to ensure robust and COMPLICATIONS • To prevent thrombus formation and contrast media crystal formation, maintain a constant consistent results for assessing core infarct size. infusion of appropriate flush solution between guide catheter and Microcatheter and between Procedures requiring percutaneous catheter introduction should not be attempted by physicians Microcatheter and Retriever or guidewire. unfamiliar with possible complications which may occur during or after the procedure. WARNINGS APPLIED TO ALL INDICATIONS • Do not attach a torque device to the shaped proximal end of DOC® Compatible Retriever. Possible complications include, but are not limited to, the following: air embolism; hematoma • Administration of IV t-PA should be within the FDA-approved window (within 3 hours of stroke ® or hemorrhage at puncture site; infection; distal embolization; pain/headache; vessel spasm, symptom onset). Damage may occur, preventing ability to attach DOC Guide Wire Extension. thrombosis, dissection, or perforation; emboli; acute occlusion; ischemia; intracranial hemorrhage; • To reduce risk of vessel damage, adhere to the following recommendations: DOC is a trademark of Abbott Laboratories. false aneurysm formation; neurological deficits including stroke; and death. – Do not perform more than six (6) retrieval attempts in same vessel using Retriever devices. COMPATIBILITY – Maintain Retriever position in vessel when removing or exchanging Microcatheter. 3x20mm retrievers are compatible with Trevo® Pro 14 Microcatheters (REF 90231) and Trevo® • To reduce risk of kinking/fracture, adhere to the following recommendations: Pro 18 Microcatheters (REF 90238). 4x20mm retrievers are compatible with Trevo® Pro 18 – Immediately after unsheathing Retriever, position Microcatheter tip marker just proximal Microcatheters (REF 90238). 4x30mm retrievers are compatible with Excelsior® XT-27® to shaped section. Maintain Microcatheter tip marker just proximal to shaped section of Stryker Neurovascular Microcatheters (150cm x 6cm straight REF 275081) and Trevo® Pro 18 Microcatheters (REF 90238). Retriever during manipulation and withdrawal. ® ® 6x25mm Retrievers are compatible with Excelsior XT-27 Microcatheters (150cm x 6cm straight – Do not rotate or torque Retriever. 47900 Bayside Parkway REF 275081). Recommended minimum vessel ID for all Retriever sizes is 2.5mm. Compatibility of – Use caution when passing Retriever through stented arteries. the Retriever with other microcatheters has not been established. Performance of the Retriever Fremont, CA 94538 device may be impacted if a different microcatheter is used. • The Retriever is a delicate instrument and should be handled carefully. Before use and when possible during procedure, inspect device carefully for damage. Do not use a device that shows Balloon Guide Catheters (such as Merci® Balloon Guide Catheter and FlowGate® Balloon Guide signs of damage. Damage may prevent device from functioning and may cause complications. strykerneurovascular.com Catheter) are recommended for use during thrombus removal procedures. • Do not advance or withdraw Retriever against resistance or significant vasospasm. Moving or torquing device against resistance or significant vasospasm may result in damage to vessel Date of Release: APR/2018 Copyright © 2018 Stryker or device. Assess cause of resistance using fluoroscopy and if needed resheath the device AP002078 v1.0 | Page 2 of 2 to withdraw. EX_EN_US SAVE THE DATE

1 st Annual Meeting of the AMERICAN SOCIETY OF Pediatric Neuroradiology January 18-20, 2019 Royal Sonesta Hotel - New Orleans • New Orleans, Louisiana

Please contact Educational Symposia, at 813-806-1000 or [email protected] or visit www.ASPNR.org for additional information. IMPORTANT SAFETY INFORMATION1

WARNING: NEPHROGENIC SYSTEMIC FIBROSIS (NSF)

Gadolinium-based contrast agents (GBCAs) increase the risk for NSF among patients with impaired elimination of the drugs. Avoid use of GBCAs in these patients unless the diagnostic information is essential and not available with non-contrasted MRI or other modalities. NSF may result in fatal or debilitating fibrosis affecting the skin, muscle and internal organs. • The risk for NSF appears highest among patients with: – Chronic, severe kidney disease (GFR < 30 mL/min/1.73m2), or – Acute kidney injury. • Screen patients for acute kidney injury and other conditions that may reduce renal function. For patients at risk for chronically reduced renal function (e.g. age > 60 years, hypertension, diabetes), estimate the glomerular filtration rate (GFR) through laboratory testing. • For patients at highest risk for NSF, do not exceed the recommended DOTAREM dose and allow a sufficient period of time for elimination of the drug from the body prior to any re-administration.

INDICATIONS AND USAGE USE IN SPECIFIC POPULATIONS DOTAREM® (gadoterate meglumine) injection is a prescription gadolinium- • Pregnancy: GBCAs cross the human placenta and result in fetal based contrast agent indicated for intravenous use with magnetic resonance exposure and gadolinium retention. Use only if imaging is essential imaging (MRI) in brain (intracranial), spine and associated tissues in adult and during pregnancy and cannot be delayed. pediatric patients (including term neonates) to detect and visualize areas with • Lactation: There are no data on the presence of gadoterate in human disruption of the blood brain barrier (BBB) and/or abnormal vascularity. milk, the effects on the breastfed infant, or the effects on milk production. However, published lactation data on other GBCAs indicate that 0.01 to CONTRAINDICATIONS 0.04% of the maternal gadolinium dose is present in breast milk. History of clinically important hypersensitivity reactions to DOTAREM. • Pediatric Use: The safety and efficacy of DOTAREM at a single dose WARNINGS AND PRECAUTIONS of 0.1 mmol/kg has been established in pediatric patients from birth UFSN
OFPOBUFT
ˈ

XFFLT
HFTUBUJPOBM
BHF
UP

ZFBST
PG
BHF
CBTFE
• Hypersensitivity Reactions: Anaphylactic and anaphylactoid reactions have on clinical data. The safety of DOTAREM has not been established in been reported with DOTAREM, involving cardiovascular, respiratory, and/or preterm neonates. No cases of NSF associated with DOTAREM or any cutaneous manifestations. Some patients experienced circulatory collapse and other GBCA have been identified in pediatric patients age 6 years died. In most cases, initial symptoms occurred within minutes of DOTAREM and younger. administration and resolved with prompt emergency treatment. • Before DOTAREM administration, assess all patients for any history of a You are encouraged to report negative side effects of prescription drugs to reaction to contrast media, bronchial asthma and/or allergic disorders. the FDA. Visit www.fda.gov/medwatch or call 1-800-FDA-1088. These patients may have an increased risk for a hypersensitivity reaction to DOTAREM. Please see the full Prescribing Information, including the patient • Administer DOTAREM only in situations where trained personnel and Medication Guide, for additional important safety information. therapies are promptly available for the treatment of hypersensitivity reactions, including personnel trained in resuscitation. References: 1. Dotarem [package insert]. Princeton, NJ: Guerbet LLC; Aug 2018. 2. Maravilla K et al. Comparison of Gadoterate Meglumine and • Gadolinium Retention: Gadolinium is retained for months or years in several Gadobutrol in the Diagnosis of Primary Brain Tumors: A Double-Blind organs. The highest concentrations have been identified in the bone, followed Randomized Controlled Intraindividual Crossover Study (the REMIND by brain, skin, kidney, liver and spleen. The duration of retention also varies 4UVEZ 

+VOF

EPJ
BKOS"
<&QVC
BIFBE
PG
QSJOU>

by tissue, and is longest in bone. Linear GBCAs cause more retention than 3. de Kerviler E et al. Adverse reactions to gadoterate meglumine: macrocyclic GBCAs. SFWJFX
PG
PW‡PFS

ZFBST
PG
DMJOJDBM
VTF
BOE
NPSF
UIBO

NJMMJPO
• Consequences of gadolinium retention in the brain have not been EPTFT
*OWFTU
3BEJPM
r
$SPTT3FG
.FEMJOF

4. Endrikat established. Adverse events involving multiple organ systems have been J et al. Safety of gadobutrol: results from 42 clinical phase II to IV reported in patients with normal renal function without an established causal TUVEJFT
BOE
QPTUNBSLFUJOH
TVSWFJMMBODF
BGUFS

NJMMJPO
BQQMJDBUJPOT
link to gadolinium retention. *OWFTU
3BEJPM

r
$SPTT3FG
.FEMJOF

5. Port M et al. • Acute Kidney Injury: In patients with chronically reduced renal function, acute Efficiency,thermodynamic and kinetic stability of marketed gadolinium kidney injury requiring dialysis has occurred with the use of GBCAs. The risk chelates and their possible clinical consequences: a critical review. of acute kidney injury may increase with increasing dose of the contrast agent; #JPNFUBMT


6. Frenzel T et al. Stability of gadolinium- administer the lowest dose necessary for adequate imaging. based magnetic resonance imaging contrast agents in human serum BU
‹$
*OWFTU
3BEJPM


• Extravasation and Injection Site Reactions: Ensure catheter and venous patency before the injection of DOTAREM. Extravasation into tissues during DOTAREM® is a registered trademark of Guerbet LLC, and is available by DOTAREM administration may result in tissue irritation. prescription only.

ADVERSE REACTIONS Gadavist® is a registered trademark of the Bayer group of companies, and • The most common adverse reactions associated with DOTAREM in is available by prescription only. clinical trials were nausea, headache, injection site pain, injection site coldness and rash. • Serious adverse reactions in the Postmarketing experience have been reported with DOTAREM. These serious adverse reactions include but are not (6 limited to: arrhythmia, cardiac arrest, respiratory arrest, pharyngeal edema, laryngospasm, bronchospasm, coma and convulsion. IMPORTANT SAFETY INFORMATION1

WARNING: NEPHROGENIC SYSTEMIC FIBROSIS (NSF)

Gadolinium-based contrast agents (GBCAs) increase the risk for NSF among patients with impaired elimination of the drugs. Avoid use of GBCAs in these patients unless the diagnostic information is essential and not available with non-contrasted MRI or other modalities. NSF may result in fatal or debilitating fibrosis affecting the skin, muscle and internal organs. Comparison of Dotarem® to Gadavist® for Overall Visualization and Characterization in the MRI Diagnosis of Primary Brain Tumors • The risk for NSF appears highest among patients with: – Chronic, severe kidney disease (GFR < 30 mL/min/1.73m2), or – Acute kidney injury. • Screen patients for acute kidney injury and other conditions that may reduce renal function. For patients at risk for chronically reduced renal function (e.g. age > 60 years, hypertension, diabetes), estimate the glomerular filtration rate (GFR) through laboratory testing. • For patients at highest risk for NSF, do not exceed the recommended DOTAREM dose and allow a sufficient period of time for elimination of the drug from the body prior to any re-administration.

INDICATIONS AND USAGE USE IN SPECIFIC POPULATIONS DOTAREM® (gadoterate meglumine) injection is a prescription gadolinium- • Pregnancy: GBCAs cross the human placenta and result in fetal based contrast agent indicated for intravenous use with magnetic resonance exposure and gadolinium retention. Use only if imaging is essential imaging (MRI) in brain (intracranial), spine and associated tissues in adult and during pregnancy and cannot be delayed. pediatric patients (including term neonates) to detect and visualize areas with • Lactation: There are no data on the presence of gadoterate in human disruption of the blood brain barrier (BBB) and/or abnormal vascularity. milk, the effects on the breastfed infant, or the effects on milk production. %FTQJUF
UIF
EJșFSFODF
JO
SFMBYJWJUZ
CFUXFFO
UIFTF

(#$"T However, published lactation data on other GBCAs indicate that 0.01 to CONTRAINDICATIONS 0.04% of the maternal gadolinium dose is present in breast milk. History of clinically important hypersensitivity reactions to DOTAREM. • Pediatric Use: The safety and efficacy of DOTAREM at a single dose 5IFSF
*T
/P
.FBTVSBCMF
%JșFSFODF
JO
$MJOJDBM
#FOFȚU
WARNINGS AND PRECAUTIONS of 0.1 mmol/kg has been established in pediatric patients from birth UFSN
OFPOBUFT
ˈ

XFFLT
HFTUBUJPOBM
BHF
UP

ZFBST
PG
BHF
CBTFE
® ®2 • Hypersensitivity Reactions: Anaphylactic and anaphylactoid reactions have on clinical data. The safety of DOTAREM has not been established in Observed Between Dotarem and Gadavist been reported with DOTAREM, involving cardiovascular, respiratory, and/or preterm neonates. No cases of NSF associated with DOTAREM or any cutaneous manifestations. Some patients experienced circulatory collapse and other GBCA have been identified in pediatric patients age 6 years died. In most cases, initial symptoms occurred within minutes of DOTAREM and younger. as demonstrated by the REMIND Study, a multicenter, double-blind, administration and resolved with prompt emergency treatment. randomized, controlled intraindividual crossover study.2 • Before DOTAREM administration, assess all patients for any history of a You are encouraged to report negative side effects of prescription drugs to reaction to contrast media, bronchial asthma and/or allergic disorders. the FDA. Visit www.fda.gov/medwatch or call 1-800-FDA-1088. These patients may have an increased risk for a hypersensitivity reaction This study “demonstrates the non-inferiority of gadoterate meglumine [Dotarem®] versus gadobutrol [Gadavist®] for overall to DOTAREM. Please see the full Prescribing Information, including the patient visualization and characterization of primary brain tumors”.2 Additionally, there was no preference of the readers for either • Administer DOTAREM only in situations where trained personnel and Medication Guide, for additional important safety information. contrast agent, in most cases, regarding border delineation, internal morphology, and the qualitative degree of contrast therapies are promptly available for the treatment of hypersensitivity 2 reactions, including personnel trained in resuscitation. References: 1. Dotarem [package insert]. Princeton, NJ: Guerbet LLC; enhancement, despite quantitative mean lesion percentage enhancement being higher with gadoburtol. Aug 2018. 2. Maravilla K et al. Comparison of Gadoterate Meglumine and • Gadolinium Retention: Gadolinium is retained for months or years in several Gadobutrol in the Diagnosis of Primary Brain Tumors: A Double-Blind • 'PS
BMM
SFBEFST
JO
UIF
3&.*/%
4UVEZ
NPSF
UIBO

PG
QBUJFOUT
QSFTFOUFE
XJUI
HPPE
PS
FYDFMMFOU
PWFSBMM
MFTJPO
organs. The highest concentrations have been identified in the bone, followed Randomized Controlled Intraindividual Crossover Study (the REMIND visualization and characterization with either Dotarem® or Gadavist®.2 by brain, skin, kidney, liver and spleen. The duration of retention also varies 4UVEZ 

+VOF

EPJ
BKOS"
<&QVC
BIFBE
PG
QSJOU>

by tissue, and is longest in bone. Linear GBCAs cause more retention than 3. de Kerviler E et al. Adverse reactions to gadoterate meglumine: • The REMIND Study also demonstrated a low incidence of immediate reported AEs with Dotarem® and with Gadavist®, macrocyclic GBCAs. SFWJFX
PG
PW‡PFS

ZFBST
PG
DMJOJDBM
VTF
BOE
NPSF
UIBO

NJMMJPO
as shown in multiple previous studies.   • Consequences of gadolinium retention in the brain have not been EPTFT
*OWFTU
3BEJPM
r
$SPTT3FG
.FEMJOF

4. Endrikat established. Adverse events involving multiple organ systems have been J et al. Safety of gadobutrol: results from 42 clinical phase II to IV • Dotarem® is the only imaging contrast with macrocyclic and ionic structure for high thermodynamic and kinetic stability.  reported in patients with normal renal function without an established causal TUVEJFT
BOE
QPTUNBSLFUJOH
TVSWFJMMBODF
BGUFS

NJMMJPO
BQQMJDBUJPOT
  link to gadolinium retention. *OWFTU
3BEJPM

r
$SPTT3FG
.FEMJOF

5. Port M et al. • Dotarem® is not only trusted for high molecular stability 
UIF
3&.*/%
4UVEZ
EFNPOTUSBUFT
UIBU
JU
JT
BT
FˋFDUJWF
2 • Acute Kidney Injury: In patients with chronically reduced renal function, acute Efficiency,thermodynamic and kinetic stability of marketed gadolinium as Gadavist® for MRI diagnosis of primary brain tumors. kidney injury requiring dialysis has occurred with the use of GBCAs. The risk chelates and their possible clinical consequences: a critical review. of acute kidney injury may increase with increasing dose of the contrast agent; #JPNFUBMT


6. Frenzel T et al. Stability of gadolinium- administer the lowest dose necessary for adequate imaging. based magnetic resonance imaging contrast agents in human serum Please see Important Safety information on opposite page. For more information on Dotarem®, BU
‹$
*OWFTU
3BEJPM


• Extravasation and Injection Site Reactions: Ensure catheter and venous please see Full Prescribing Information including Boxed Warning and Medication Guide. patency before the injection of DOTAREM. Extravasation into tissues during DOTAREM® is a registered trademark of Guerbet LLC, and is available by DOTAREM administration may result in tissue irritation. prescription only.

ADVERSE REACTIONS Gadavist® is a registered trademark of the Bayer group of companies, and • The most common adverse reactions associated with DOTAREM in is available by prescription only. Learnearn more ata clinical trials were nausea, headache, injection site pain, injection site coldness and rash. RemindStudy.commindStudy.c • Serious adverse reactions in the Postmarketing experience have been reported with DOTAREM. These serious adverse reactions include but are not (6 limited to: arrhythmia, cardiac arrest, respiratory arrest, pharyngeal edema, laryngospasm, bronchospasm, coma and convulsion. Redefi ne aspiration.

AXS Vecta 71 Aspiration Catheter

Big 0.071in ID aspiration lumen to ingest more clot

Deliver through a 0.088in ID long sheath or the new 0.091in AXS Infi nity LS™ Plus Long Sheath

Packaged with the Scout Introducer, a 0.044in lumen nitinol cross coil catheter that replaces the need for a 3MAX or other delivery catheter

Copyright © 2018 Stryker AP002264 v1.0 | Page 1 of 2 AMERICAN JOURNAL OF NEURORADIOLOGY OCTOBER 2018 Publication Preview at www.ajnr.org features articles released in advance of print. AJNRVOLUME 39 Visit www.ajnrblog.org to comment on AJNR content and chat with colleagues NUMBER 10 and AJNR’s News Digest at http://ajnrdigest.org to read the stories behind the WWW.AJNR.ORG latest research in neuroimaging. 1775 PERSPECTIVES B. Thomas

REVIEW ARTICLE 1776 Deep Learning in Neuroradiology G. Zaharchuk, et al. ADULT BRAIN

PRACTICE PERSPECTIVES 1785 Comparison of Advanced Imaging Resources, Radiology Workforce, and Payment Methodologies between the United States and Canada H.A. Valand, et al. GENERAL CONTENTS 1791 Clinical Value of Hybrid TOF-PET/MR Imaging–Based Multiparametric ADULT BRAIN Imaging in Localizing Seizure Focus in Patients with MRI-Negative Temporal Lobe Epilepsy K. Shang, et al. 1799 Longitudinal Persistence of Meningeal Enhancement on Postcontrast 7T ADULT BRAIN 3D-FLAIR MRI in Multiple Sclerosis S.N. Jonas, et al. 1806 Automated Integration of Multimodal MRI for the Probabilistic Detection ADULT BRAIN of the Central Vein Sign in White Matter Lesions J.D. Dworkin, et al. 1814 Predicting Genotype and Survival in Glioma Using Standard Clinical MR ADULT BRAIN Imaging Apparent Diffusion Coefficient Images: A Pilot Study from The Cancer Genome Atlas C.-C. Wu, et al. 1821 Near-Term Decrease in Brain Volume following Mild Traumatic Injury Is ADULT BRAIN Detectible in the Context of Preinjury Volumetric Stability: Neurobiologic Insights from Analysis of Historical Imaging Examinations A.E. Goldman-Yassen, et al. 1827 3D Black-Blood Luminal Angiography Derived from High-Resolution MR ADULT BRAIN Vessel Wall Imaging in Detecting MCA Stenosis: A Preliminary Study X. Bai, et al. 1833 Clinical Evaluation of Highly Accelerated Compressed Sensing ADULT BRAIN Time-of-Flight MR Angiography for Intracranial Arterial Stenosis S.s. Lu, et al. 1839 Cerebrovascular Reactivity during Prolonged Breath-Hold in ADULT BRAIN Experienced Freedivers V.C. Keil, et al.

AJNR (Am J Neuroradiol ISSN 0195–6108) is a journal published monthly, owned and published by the American Society of Neuroradiology (ASNR), 800 Enterprise Drive, Suite 205, Oak Brook, IL 60523. Annual dues for the ASNR include $170.00 for journal subscription. The journal is printed by Cadmus Journal Services, 5457 Twin Knolls Road, Suite 200, Columbia, MD 21045; Periodicals postage paid at Oak Brook, IL and additional mailing offices. Printed in the U.S.A. POSTMASTER: Please send address changes to American Journal of Neuroradiology, P.O. Box 3000, Denville, NJ 07834, U.S.A. Subscription rates: nonmember $400 ($470 foreign) print and online, $320 online only; institutions $460 ($530 foreign) print and basic online, $915 ($980 foreign) print and extended online, $380 online only (basic), extended online $825; single copies are $35 each ($40 foreign). Indexed by PubMed/Medline, BIOSIS Previews, Current Contents (Clinical Medicine and Life Sciences), EMBASE, Google Scholar, HighWire Press, Q-Sensei, RefSeek, Science Citation Index, SCI Expanded, Meta/CZI and ReadCube. Copyright © American Society of Neuroradiology. 1848 Reasons for Reperfusion Failures in Stent-Retriever-Based INTERVENTIONAL Thrombectomy: Registry Analysis and Proposal of a Classification System J. Kaesmacher, et al. 1854 Thrombus Permeability on Dynamic CTA Predicts Good Outcome after INTERVENTIONAL Reperfusion Therapy Z. Chen, et al. 1860 Identification of Hostile Hemodynamics and Geometries of Cerebral INTERVENTIONAL Aneurysms: A Case-Control Study B.J. Chung, et al. 1867 Integrating 3D Rotational Angiography into Gamma Knife Planning INTERVENTIONAL H. Hasegawa, et al. 1871 Quantification of Blood Velocity with 4D Digital Subtraction INTERVENTIONAL Angiography Using the Shifted Least-Squares Method Y. Wu, et al. 1878 Correlation between Human Papillomavirus Status and Quantitative HEAD & NECK MR Imaging Parameters including Diffusion-Weighted Imaging and Texture Features in Oropharyngeal Carcinoma M. Ravanelli, et al. 1884 Negative Predictive Value of NI-RADS Category 2 in the First HEAD & NECK Posttreatment FDG-PET/CT in Head and Neck Squamous Cell Carcinoma P. Wangaryattawanich, et al. 1889 The Diagnostic Value of Diffusion-Weighted Imaging in Differentiating HEAD & NECK Metastatic Lymph Nodes of Head and Neck Squamous Cell Carcinoma: A Systematic Review and Meta-Analysis C.H. Suh, et al. 1896 Radiologic-Pathologic Correlation of Tumor Thickness and Its HEAD & NECK Prognostic Importance in Squamous Cell Carcinoma of the Oral Cavity: Implications for the Eighth Edition Tumor, Node, Metastasis Classification E.A.M. Weimar, et al. 1903 MR Imaging of the Facial Nerve through the Temporal Bone at 3T with a HEAD & NECK Noncontrast Ultrashort Echo Time Sequence J.P. Guenette, et al. 1907 CT versus MR Imaging in Estimating Cochlear Radiation Dose during HEAD & NECK Gamma Knife Surgery for Vestibular Schwannomas A.M. Faramand, et al. PATIENT SAFETY 1912 Multidelay Arterial Spin-Labeling MRI in Neonates and Infants: Cerebral PEDIATRICS Perfusion Changes during Brain Maturation H.G. Kim, et al. 1919 The Impact of Persistent Leukoencephalopathy on Brain White Matter PEDIATRICS Microstructure in Long-Term Survivors of Acute Lymphoblastic Leukemia Treated with Chemotherapy Only N.D. Sabin, et al. 1926 Apparent Diffusion Coefficient Levels and Neurodevelopmental PEDIATRICS Outcome in Fetuses with Brain MR Imaging White Matter Hyperintense Signal E. Katorza, et al. 1932 Longitudinal Findings of MRI and PET in West Syndrome with Subtle PEDIATRICS Focal Cortical Dysplasia Y. Sakaguchi, et al. 1938 Imaging of Clival Hypoplasia in CHARGE Syndrome and Hypothesis for PEDIATRICS Development: A Case-Control Study C.M. de Geus, et al. 1943 Multiple Brain Developmental Venous Anomalies as a Marker for PEDIATRICS Constitutional Mismatch Repair Deficiency Syndrome S.I. Shiran, et al. 1947 Spinal Imaging Findings of Open Spinal Dysraphisms on Fetal and PEDIATRICS Postnatal MRI U.D. Nagaraj, et al. SPINE 1953 Dilated Vein of the Filum Terminale on MRI: A Marker for Deep Lumbar SPINE and Sacral Dural and Epidural Arteriovenous Fistulas W. Brinjikji, et al. 1957 Spinal Instrumentation Rescue with Cement Augmentation A. Cianfoni, et al. SPINE ONLINE FEATURES LETTERS

E105 Comment on “Common Origin of Brachiocephalic and Left Common Carotid Arteries: Proposal of New Terminology” R. Rajagopal, et al. E106 Reply A.M. Tawfik, et al. E107 Quality-Control Assessment to Improve the Accuracy of Dynamic Contrast-Enhanced MR Imaging Perfusion A. Lecler, et al. E108 Reply S. Karimi E109 Vertebroplasty: Expectation or Evidence-Based Interventional Radiology? A. Braillon, et al. E110 Reply R.V. Chandra, et al.

BOOK REVIEWS R.M. Quencer, Section Editor Please visit www.ajnrblog.org to read and comment on Book Reviews.

The goal is to classify MR images into 4 specific diagnoses. Multiple different images form the training set. For each new case, the image is broken down into its constituent voxels, each one of which acts as an input into the network. This example has 3 hidden layers with 7 neurons in each layer. Final output is the probabilities of the 4 classification states. At the bottom is a zoomed-in view of an individual neuron in the second hidden layer, which receives input from the previous layer, performs a standard matrix multiplication, passes this through a nonlinear function, and outputs Indicates Editor’s Indicates Fellows’ Indicates open access to non- Indicates article with a single value to all Choices selection Journal Club selection subscribers at www.ajnr.org supplemental on-line table the neurons of the Indicates article with Indicates article with Evidence-Based Evidence-Based next layer. supplemental on-line photo supplemental on-line video Medicine Level 1 Medicine Level 2 AMERICAN JOURNAL OF NEURORADIOLOGY Publication Preview at www.ajnr.org features articles released in advance of print. Visit www.ajnrblog.org to comment on AJNR content and chat with colleagues AJNR and AJNR’s News Digest at http://ajnrdigest.org to read the stories behind the latest research in neuroimaging.

OCTOBER 2018 • VOLUME 39 • NUMBER 10 • WWW.AJNR.ORG

Official Journal: EDITORIAL BOARD A´lex Rovira-Can˜ellas, Barcelona, Spain American Society of Neuroradiology Ashley H. Aiken, Atlanta, GA Paul M. Ruggieri, Cleveland, OH American Society of Functional Neuroradiology Lea M. Alhilali, Phoenix, AZ Amit M. Saindane, Atlanta, GA Kubilay Aydin, Istanbul, Turkey Erin Simon Schwartz, Philadelphia, PA American Society of Head and Neck Radiology John D. Barr, Dallas, TX Lubdha M. Shah, Salt Lake City, UT American Society of Pediatric Neuroradiology Ari Blitz, Baltimore, MD Maksim Shapiro, New York, NY American Society of Spine Radiology Barton F. Branstetter IV, Pittsburgh, PA Timothy Shepherd, New York, NY EDITOR-IN-CHIEF Jonathan L. Brisman, Lake Success, NY Mark S. Shiroishi, Los Angeles, CA Keith Cauley, Danville, PA Jeffrey S. Ross, MD Bruno P. Soares, Baltimore, MD James Y. Chen, San Diego, CA Maria Vittoria Spampinato, Charleston, SC Professor of Radiology, Department of Radiology, Asim F. Choudhri, Memphis, TN Khin Khin Tha, Sapporo, Hokkaido, Japan Mayo Clinic College of Medicine, Phoenix, AZ Daniel Chow, Irvine, CA Krishnamoorthy Thamburaj, Hershey, PA J. Matthew Debnam, Houston, TX SENIOR EDITORS Cheng Hong Toh, Taipei, Taiwan Seena Dehkharghani, New York, NY Harry J. Cloft, MD, PhD Yonghong Ding, Rochester, MN Aquilla S. Turk, Charleston, SC Professor of Radiology and Neurosurgery, Clifford J. Eskey, Hanover, NH Anja G. van der Kolk, Madison, WI Department of Radiology, Mayo Clinic College of Saeed Fakhran, Phoenix, AZ Willem Jan van Rooij, Tilburg, Netherlands Medicine, Rochester, MN Massimo Filippi, Milan, Italy Arastoo Vossough, Philadelphia, PA Thierry A.G.M. Huisman, MD Reza Forghani, Montreal, Quebec, Canada Elysa Widjaja, Toronto, Ontario, Canada Nils D. Forkert, Calgary, Alberta, Canada Max Wintermark, Stanford, CA Radiologist-in-Chief, Texas Children’s Hospital, Wende N. Gibbs, Los Angeles, CA Ronald L. Wolf, Philadelphia, PA Houston, TX Christine M. Glastonbury, San Francisco, CA Kei Yamada, Kyoto, Japan Yvonne W. Lui, MD John L. Go, Los Angeles, CA Carlos Zamora, Chapel Hill, NC Associate Professor of Radiology, Philipp Gölitz, Erlangen, Germany Vahe M. Zohrabian, New Haven, CT Chief of Neuroradiology, Allison Grayev, Madison, WI New York University School of Medicine, Brent Griffith, Detroit, MI EDITORIAL FELLOW New York, NY Ajay Gupta, New York, NY Alireza Radmanesh, New York, NY Rakesh Kumar Gupta, Haryana, India C.D. Phillips, MD, FACR Lotfi Hacein-Bey, Sacramento, CA SPECIAL CONSULTANTS TO THE EDITOR Professor of Radiology, Weill Cornell Medical Christopher P. Hess, San Francisco, CA AJNR Blog Editor College, Director of Head and Neck Imaging, Andrei Holodny, New York, NY Neil Lall, Denver, CO New York-Presbyterian Hospital, New York, NY Benjamin Huang, Chapel Hill, NC Case of the Month Editor Mahesh V. Jayaraman, Providence, RI Pamela W. Schaefer, MD Nicholas Stence, Aurora, CO Valerie Jewells, Chapel Hill, NC Clinical Director of MRI and Associate Director of Case of the Week Editors Christof Karmonik, Houston, TX Neuroradiology, Massachusetts General Hospital, Juan Pablo Cruz, Santiago, Chile Timothy J. Kaufmann, Rochester, MN Boston, Massachusetts, Associate Professor, Sapna Rawal, Toronto, Ontario, Canada Radiology, Harvard Medical School, Cambridge, MA Hillary R. Kelly, Boston, MA Toshibumi Kinoshita, Akita, Japan Classic Case Editor Charles M. Strother, MD Kennith F. Layton, Dallas, TX Sandy Cheng-Yu Chen, Taipei, Taiwan Professor of Radiology, Emeritus, University of Alexander Lerner, Los Angeles, CA Health Care and Socioeconomics Editor Wisconsin, Madison, WI Michael Lev, Boston, MA Pina C. Sanelli, New York, NY STATISTICAL SENIOR EDITOR Karl-Olof Lovblad, Geneva, Switzerland Physics Editor Franklin A. Marden, Chicago, IL Greg Zaharchuk, Stanford, CA Bryan A. Comstock, MS Joseph C. McGowan, Merion Station, PA Podcast Editor Senior Biostatistician, Stephan Meckel, Freiburg, Germany Wende N. Gibbs, Los Angeles, CA Department of Biostatistics, Christopher J. Moran, St. Louis, MO Twitter Editor University of Washington, Seattle, WA Takahisa Mori, Kamakura City, Japan Jennifer McCarty, Houston, TX ARTIFICIAL INTELLIGENCE DEPUTY EDITOR Suresh Mukherji, Ann Arbor, MI Alexander J. Nemeth, Chicago, IL Christopher G. Filippi, MD Renato Hoffmann Nunes, Sao Paulo, Brazil Professor and Vice Chair of Biomedical and Sasan Partovi, Cleveland, OH Translational Science, Laurent Pierot, Reims, France Donald and Barbara Zucker School of Medicine at Jay J. Pillai, Baltimore, MD Hofstra/Northwell, Whitney B. Pope, Los Angeles, CA Lenox Hill Hospital and Greenwich Village Joana Ramalho, Lisbon, Portugal Healthplex, New York, NY Otto Rapalino, Boston, MA

Founding Editor Managing Editor Juan M. Taveras Karen Halm Editors Emeriti Assistant Managing Editor Mauricio Castillo, Robert I. Grossman, Laura Wilhelm Michael S. Huckman, Robert M. Quencer Communications Coordinator Rebecca Artz Marketing and Social Media Manager Kylie Mason Executive Director, ASNR Mary Beth Hepp PERSPECTIVES

Title: Purple Toronto. Oil on canvas 22Љϫ15Љ. This is a view of the harborfront from the central island, recollections from my fellowship days. Bejoy Thomas, MD, DNB, PDCC, Trivandrum, Kerala, India

AJNR Am J Neuroradiol 39:1775 Oct 2018 www.ajnr.org 1775 REVIEW ARTICLE

Deep Learning in Neuroradiology

X G. Zaharchuk, X E. Gong, X M. Wintermark, X D. Rubin, and X C.P. Langlotz

ABSTRACT SUMMARY: Deep learning is a form of machine learning using a convolutional neural network architecture that shows tremendous promise for imaging applications. It is increasingly being adapted from its original demonstration in computer vision applications to medical imaging. Because of the high volume and wealth of multimodal imaging information acquired in typical studies, neuroradiology is poised to be an early adopter of deep learning. Compelling deep learning research applications have been demonstrated, and their use is likely to grow rapidly. This review article describes the reasons, outlines the basic methods used to train and test deep learning models, and presents a brief overview of current and potential clinical applications with an emphasis on how they are likely to change future neuroradiology practice. Facility with these methods among neuroimaging researchers and clinicians will be important to channel and harness the vast potential of this new method.

ABBREVIATIONS: AD ϭ Alzheimer disease; ADNI ϭ Alzheimer’s Disease Neuroimaging Initiative; ASL ϭ arterial spin-labeling; CNN ϭ convolutional neural network; MCI ϭ mild cognitive impairment; NC ϭ normal control

eep learning is a form of artificial intelligence, roughly mod- Deep learning has the potential to revolutionize entire indus- Deled on the structure of neurons in the brain, which has tries, including medical imaging. Given the centrality of neuro- shown tremendous promise in solving many problems in com- imaging in the diagnosis and treatment of neurologic diseases, puter vision, natural language processing, and robotics.1 It has deep learning will likely affect neuroradiologists first and most recently become the dominant form of machine learning, due to a profoundly. This article will introduce deep learning methods, convergence of theoretic advances, openly available computer overview their current successes, and speculate on the future software, and hardware with sufficient computational power. The evolution of these methods, focusing on their application to current excitement in the field of deep learning stems from neuroradiology. new data suggesting its excellent performance in a wide variety of tasks. One benchmark of machine learning performance is What is Deep Learning? the ImageNet Challenge. In this annual competition, teams It is useful to consider where deep learning fits into the broader compete to classify millions of images into discrete categories context of artificial intelligence (Fig 1). One definition suggested (tens of different kinds of dogs, fish, cars, and so forth). A for artificial intelligence is any computer method that performs watershed year was 2012, when the first neural network–based tasks normally requiring human intelligence. Machine learning is entry bested the competition and prior years’ results by a wide one type of artificial intelligence that develops algorithms to 2 margin. Since then, every winning entry has used a deep learn- enable computers to learn from existing data without explicit pro- ing framework, with performance now exceeding that of gramming. Examples are classification algorithms such as cluster- humans. ing, logistic regression, and support vector machines. Machine learning methods can be further divided into super- vised and unsupervised learning. In supervised learning, some From the Departments of Radiology (G.Z., M.W., D.R., C.P.L.) and Electrical Engi- neering (E.G.), Stanford University and Stanford University Medical Center, “ground truth” exists, which is used to train the algorithms. One Stanford, California. example is a collection of brain CT scans that a neuroradiologist Drs Rubin and Langlotz are co-senior authors. has classified into different groups (ie, hemorrhage versus no Please address correspondence to Greg Zaharchuk, MD, PhD, Department of hemorrhage). In contrast, for unsupervised learning, no criterion Radiology, Stanford University, 1201 Welch Rd, mail code 5488, Stanford, CA 94305-5488; e-mail: [email protected]; @GregZ_MD standard images or classifications are used—the computer itself Indicates open access to non-subscribers at www.ajnr.org must determine the classes. One example is clustering, in which http://dx.doi.org/10.3174/ajnr.A5543 images are placed in multiple groups based on similarity metrics

1776 Zaharchuk Oct 2018 www.ajnr.org FIG 1. Artificial intelligence methods. Within the subset of machine learning methods, deep learning is usually implemented as a form of supervised learning. without knowing a priori what is driving the separation.3 While this is known as the network architecture (Fig 3). Each neuron unsupervised learning holds great promise for medical imaging, stores a numeric value, and each connection between neurons this review focuses on supervised learning. represents a weight. Weights connect the neurons in different Viewed in this context, deep learning is a supervised machine layers and represent the strength of connections between the neu- learning method that uses a specific architecture, namely some rons. A “fully connected” layer in which all neurons in one layer form of neural network. The power of these techniques is in their are connected to all neurons in the next can be interpreted and scalability, which is largely based on their ability to automatically implemented as a matrix multiplication. Finally, it is customary to extract relevant features. In the past, constructing an image-clas- include a nonlinear “activation function” at the output of the sification algorithm took years of effort on the part of domain neuron. This introduces nonlinearity into the equations so that experts and experienced artificial intelligence researchers. Deep complex functions can be represented that would not otherwise learning allows such a classifier to be created automatically from a be possible. Historically, sigmoids and hyperbolic tangents have labeled dataset in days. These neural networks are loosely inspired been used on the basis of insights from neuroscience; however, from how the brain is structured, with hidden layers representing researchers have since found that the rectified linear unit is both interneurons (Fig 2). While modern neural networks share these simpler to implement and more effective. The rectified linear unit similarities to the brain, whether more fidelity to known brain function outputs the value of the neuron for positive values and structures would improve performance is an actively debated zero for negative values. question.4 For example, in computer vision applications, many of The choice of network architecture for a specific application is the features to which hidden layers are sensitive (such as edges in not always obvious, though some typical configurations and as- different orientations) have correlates in the mammalian visual sumptions exist. The number of neurons in the hidden layers cortex.5 tends to be larger than that in either the input or the output layers. For neuroimaging, a simple deep learning model may accept The final layer encodes the desired outcomes or labeled states. For image data as a vector composed of voxel intensities, with each example, if one wishes to classify an image as “hemorrhage” or voxel serving as an input “neuron.” While the examples below “no hemorrhage,” 2 final layer neurons are appropriate. Com- assume the use of individual images, more generally, the input can monly, the value stored by each final neuron is interpreted as the consist of entire imaging series, multiple series, or even multiple probability that the training example corresponds to a specific modalities. Next, one must determine how many layers (how class. The goal of training is to optimize the network weights deep) and how many neurons per layer (how wide) to include; so that when a new sample image is input, the probabilities AJNR Am J Neuroradiol 39:1776–84 Oct 2018 www.ajnr.org 1777 FIG 2. Parallels between artificial and biologic neural networks. Hidden layers of artificial neural networks can be thought to be analogous to brain interneurons.

FIG 3. Example of a simple deep network architecture. The goal of this network is to classify MR images into 4 specific diagnoses (normal, tumor, stroke, hemorrhage). Multiple different images form the training set. For each new case, the image is broken down into its constituent voxels, each one of which acts as an input into the network. This example has 3 hidden layers with 7 neurons in each layer, and the final output is the probabilities of the 4 classification states. All layers are fully connected. At the bottom is a zoomed-in view of an individual neuron in the second hidden layer, which receives input from the previous layer, performs a standard matrix multiplication (including a bias term), passes this through a nonlinear function (the rectified linear unit function in this example), and outputs a single value to all the neurons of the next layer.

1778 Zaharchuk Oct 2018 www.ajnr.org measured at the output are heavily skewed to the correct class. For of the final trained model to new data is usually computationally example, if we input an image with hemorrhage, we would like the fast. model to output a high probability for hemorrhage and low prob- Choosing the right cost function is important. For classifica- ability for other classes. How is this accomplished? tion, the value of the cost function should be low when the model predicts the correct class and high when its predictions are off. A Training Simple Neural Network Deep Learning Models popular cost function for classification is “cross-entropy loss,” an Neural networks are ideally trained using large numbers of cases extension of logistic regression to multiple classes that can be that are divided into several groups. Usually the largest fraction is implemented using the softmax function. For image prediction, used for model training (50%–60%), with another 10%–20% for common cost functions include the root-mean-square error be- validation and 20%–40% for testing. The training cases are used tween the predicted and reference images and measures of simi- to set the model parameters; large training sets are important larity, such as the structural similarity index metric.6,7 One prom- because even relatively shallow networks may have 100,000s of ising approach is replacing the cost function itself with a network free parameters (weights). The training dataset is looped through whose goal is to make it optimally difficult to distinguish refer- multiple times (epochs) until the accuracy of the model con- ence images from predicted images, an approach known as the verges. At first, predictions will be poor. However, the beauty of “generative adversarial network” approach.8 Generative adver- this setup is that you can compare the output of the model with sarial networks strive to eliminate systematic differences between the ground truth via the use of a “cost function,” a single number the predicted and reference images, which is highly desirable in that quantifies how far off the model is. Back-propagation, a tech- the radiology setting. nique whereby the strength of connections between neurons (weights) can be adjusted on the basis of the value of the cost From Simple to Convolutional Neural Networks function, is then used to reinforce correct predictions and penal- Fully connected neural networks are computationally expensive ize incorrect predictions. This procedure is repeated using sepa- because the number of weights is very large, especially with images rate training examples and multiple iterations, thereby optimiz- of typical matrix sizes (256 ϫ 256 ϭ 65,536 voxels). With even just ing the weights and effectively training the model. Once the model 1 slice, Ͼ4 billion weights are required to implement a fully con- is trained, there are several “hyperparameters” to optimize, in- nected layer. Thus, much research in image-based deep learning cluding the learning rate and number of epochs. Finally, the test- has moved to using more computationally efficient structures, ing set is used to assess the model accuracy on data that has not specifically convolutional neural networks (CNNs). been used for training. This assessment will yield an error rate CNNs are well-suited for imaging. Instead of full connections, similar to or higher than that for the training set, and this helps to a small “kernel” of weights is applied at each image position to gauge how well the final model will perform on real world data. determine the value of the neuron of the next layer (Fig 4). The While training the model is often time-intensive, the application approach mimics the mathematic operation of convolution. Be-

FIG 4. Example of training and deployment of deep convolutional neural networks. During training, each image is analyzed separately, and at each layer, a small set of weights (convolution kernel) is moved across the image to provide input to the next layer. Each layer can have multiple channels. By pooling adjacent voxels or using larger stride distances between application of the kernel, deeper layers often have smaller spatial dimensions but more channels, each of which can be thought of as representing an abstract feature. In this example, 5 convolutional layers are followed by 3 fully connected layers, which then output a probability of the image belonging to each class. These probabilities are compared with the known class (stroke in the training example) and can be used to measure of how far off the prediction was (cost function), which can then be used to update the weights of the different kernels and fully connected parameters using back-propagation. When the model training is complete and deployed on new images, the process will produce a similar output of probabilities, in which it is hoped that the true diagnosis will have the highest likelihood. AJNR Am J Neuroradiol 39:1776–84 Oct 2018 www.ajnr.org 1779 tween 2 layers, the only weights required are those for the kernel, sis arise. Each link in this chain can potentially benefit from a deep which is then rastered across the image to obtain the next layer. learning approach. This method has several advantages. First, it markedly reduces the number of weights. Second, it allows spatial invariance: Image Imaging Logistics features may occur in different locations, and a CNN allows their After a study is ordered, it needs to be triaged to a specific neuro- identification independent of their precise location. Often, CNNs imaging protocol. This process often involves the precious time of pool adjacent voxels or slide the kernel across these images at radiologists and relies on their knowledge of imaging protocols spaced intervals (a hyperparameter known as the “stride length”), and attention to clinicians’ specific requests, which are often en- so that the dimensions in each subsequent layer are smaller than capsulated in the order history as free text. Deep learning methods those in the last one. For each layer, multiple different kernels can to interpret natural language are already mature, making automa- be trained, creating multiple “channels” in each layer; such a tion of the protocoling process quite feasible. In theory, this prob- structure allows the network to learn many location-invariant lem is just about classification, with the different protocols being features, such as edges, textures, and other nonlinear representa- the classes to predict and the input being the order itself and tions of the data. With pooling or increased stride lengths, it is patient metadata. The protocolling application is ideal for deep possible to incorporate ever larger features into the hidden layers learning because of the immense amount of training data that of the network. Indeed, the ability of CNNs to extract relevant already exists; all prior studies that have been protocolled by hu- imaging features in a location-invariant way parallels the struc- mans can be used for training. ture of the visual system of the brain; Hubel and Wiesel5 showed Another promising application is to triage image review in the in the 1960s that different regions of the cat brain responded order of suspected acuity. For example, if models can be trained to strongly to features such as edges oriented in different directions. identify critical findings on images, it is possible to prioritize ra- For classification, Ն1 fully connected layer is typically added diologic review of these studies, even if they were not initially to reach the final output layer. For image prediction, upsampling ordered as “stat” studies. For large organizations, such triage of- layers are used to “re-form” the smaller dimension hidden layers fers the potential to reduce the time between acquisition and in- back into the original size of the input image. Such an architecture terpretation for critical cases, with presumably positive effects on is called an “encoder-decoder” because it represents the image in patient outcome. terms of increasing abstraction (encoding) in the hidden layers and then uses them to recreate (decode) the image. Image Acquisition and Improvement Deep learning methods can be used to perform image reconstruc- Overfitting and Data Augmentation tion and improve image quality. Deep learning frameworks are As described above, typical deep learning models have millions of capable of “learning” standard MR imaging reconstruction weights. Analogous to the idea that you need more equations than techniques, such as Cartesian and non-Cartesian acquisition variables to solve algebraic equations, if a deep network is trained schemes.10 Combining deep learning to k-space undersampling on too few examples, it is possible to perfectly represent the trans- with model-based/compressed sensing reconstruction schemes form between the input and output states. However, this ap- holds the potential to revolutionize imaging science by optimiz- proach will not generalize to new cases, a problem known as ing how image data are collected.11-13 “overfitting.” The best solution to overfitting is collecting more Also, one could apply deep learning methods to improve im- training examples, though other solutions such as regularization age quality. If images at low-resolution and high-resolution are and drop-out can also be used.7,9 Another potential solution is available, it is possible to use a deep network for super-resolu- data augmentation. tion.14 If paired image sets of low and high quality are available, Data augmentation is a method of increasing the amount of learning the optimal nonlinear transformation between them can training data. Because most image data should be recognizable be considered. This has already been applied to CT imaging and whether offset in the x-y plane, rotated, flipped, or slightly has been demonstrated to be of value on a dataset consisting of stretched or skewed, it is conventional to perform such image normal-dose and simulated low-dose CT.15 Another approach manipulations to augment the training data. While these image uses paired MR images of the same anatomy, which are acquired alterations do not add more data, they have been shown to im- at different field strengths. A study using 3T input data and 7T prove the robustness of the models, possibly by preventing the output data showed that a deep network can be trained to create model from learning features that occur only in a specific simulated “7T-like” images from 3T data.16 Often obtaining a cer- orientation. tain imaging sequence can be very time-consuming; an example is DTI, in which the need for multiple angular directions lengthens the Broad Classes of Applications examination beyond what many patients can tolerate. A deep learn- Deep learning can address many aspects of neuroradiology. The ing approach can reduce imaging duration 12-fold by predicting final overall flow of work in neuroradiology is a useful framework in parameter maps (fractional anisotropy, mean diffusivity, and so which to consider these applications. This starts with referring forth) from relatively few angular directions.17 By acquiring paired clinicians ordering studies and then moves to image acquisition. arterial spin-labeling (ASL) CBF images with 2 and 30 minutes of Next, the images are put before radiologists, and tasks surround- acquisition time, our group has trained a deep network to boost the ing detection and segmentation of lesions and differential diagno- SNR of ASL significantly (Fig 5).18 1780 Zaharchuk Oct 2018 www.ajnr.org FIG 5. An example of improving the SNR of arterial spin-labeling MR imaging using deep learning. The model is trained using low-SNR ASL images acquired with only a single repetition, while the reference image is a high-SNR ASL image acquired with multiple repetitions (in this case, 6 repetitions). Proton-density-weighted images (acquired routinely as part of the ASL scans for quantitation) and T2-weighted images are also used as inputs to the model to improve performance. The results of passing the low-SNR ASL image through the model are shown on the right, a synthetic image with improved SNR. In this example, the root-mean-squared error (RSME) between the reference image and the synthetic image compared with the original image is reduced nearly 3-fold, from 29.3% to 10.8%.

Image Transformation and marks potential abnormalities. The goal of the latter is to An extension of this is to create images with different contrast or circumscribe the regions encompassing the abnormal structures. with features of different modalities. For example, using the Identifying and delineating the margins of a lesion is important National Alliance for Medical Imaging Computing data base because neuroradiologists are often tasked with monitoring the (http://www.insight-journal.org/midas/community/view/17), Ve- change in size or activity of known lesions across time or in re- 19 mulapalli et al used a deep network to predict T1 images from T2 sponse to treatment. Deep learning also has advantages for seg- images, and vice versa. Another application is to PET/MR imag- menting normal brain structures because existing methods are ing; unlike PET/CT, in which CT is used to calculate an attenua- time-consuming and may not generalize to younger or older sub- tion map, MR images do not directly yield attenuation images. jects.23-25 Furthermore, many research projects rely on manual However, if there is information about soft tissue, air, and bone in delineation of image lesions. One can train a deep network to take the MR images, these sequences can be used as input to a deep images as input and hand-drawn manually segmented masks as network. Crucially here, the image to predict is no longer another output. Indeed, such an approach has shown great early success. MR image, but rather a coregistered CT scan of the same subject. While there are many examples in different neurologic disease Proof of principle was recently demonstrated for brain MR imag- conditions, we will reference 3 representative areas: detecting mi- ing attenuation correction, with performance superior to that of crohemorrhages, identifying infarcts and predicting final infarct competing techniques.20 Another study demonstrated a similar volumes in patients with stroke, and segmenting brain tumors. use of MR imaging to create synthetic CT for radiation therapy.21 Dou et al26 described a process to detect brain microhemor- In clinical trials, situations arise in which patients may not be rhages by training a CNN on an annotated dataset of susceptibil- able to undergo a certain diagnostic technique, such as patients ity-weighted images. They proposed a cascaded, 2-step approach, with MR imaging–incompatible implants. Alternatively, they in which candidate lesions are first identified by the CNN and may lack images at specific time points. While statistical tech- then only these lesions are input to a discriminatory CNN (ie, true niques can be used to account for such missing data, if enough microhemorrhage or mimic). With this approach, they achieved a patients drawn from the same population have completed all sensitivity of Ͼ93%, with an average of about 3 false-positive imaging examinations, it is possible to train a deep learning network to recreate these data. Li et al22 demonstrated this identifications per subject. using the Alzheimer’s Disease Neuroimaging Initiative (ADNI; Automatic identification and outlining of infarcted brain tis- 27 http://www.adni-info.org/). They trained a CNN on patients with sue would be useful in the acute stroke setting. Chen et al used both FDG-PET and T1-weighted MR imaging and then used this DWI as input to a 2-stage deep learning algorithm and were able network on a test set to predict the expected PET images from to detect 94% of all acute infarcts. Using the Dice coefficient as a patients’ MR imaging studies,22 showing that the CNN method marker of accuracy, they showed a mean score of 0.67 in a large outperformed more traditional methods. cohort of patients 2 days after stroke. Another study using a 3- layer-deep CNN followed by 2 fully connected layers showed sim- Lesion Detection and Segmentation ilar results and outperformed several other machine learning Detecting and segmenting lesions is an onerous task for humans methods.28 Another intriguing application is to predict final in- but is well-suited to machine learning. While related, they are farct volume from early DWI and/or PWI in patients with acute really 2 different tasks. The former starts with an unlabeled image stroke. Currently, the diffusion-perfusion mismatch approach is AJNR Am J Neuroradiol 39:1776–84 Oct 2018 www.ajnr.org 1781 FIG 6. An example of the predicted risk of final infarct for 2 patients with acute ischemic stroke using 2 neural networks trained, respectively, on patients with and without rtPA administration. Patient A is a 76-year-old woman with an admission NIHSS score of 10 scanned 1.5 hours after symptom onset. The ϩrtPA network estimates a negligible permanent lesion, consistent with the acute DWI suggesting little permanent tissue damage at follow-up. The ϪrtPA network indicates that without treatment, a considerable volume of the acute ischemic region will progress to a permanent lesion. Patient B is a 72-year-old man also with an admission NIHSS score of 10, scanned 2 hours after onset. In this case, the 2 networks indicate little expected impact of treatment, likely due to the progression at the time of imaging of the ischemic event as seen on the DWI. CMRO2 indicates cerebral metabolic rate of oxygen; Tmax, time-to-maximum. Figure courtesy of Kim Mouridsen and Anne Nielsen/Aarhus University, Combat Stroke. the dominant paradigm, which states that DWI lesions represent which they had 3 expert segmentations and measured a Dice co- irreversibly damaged tissue, while PWI identifies tissue at risk of efficient of 0.88 based on a method that incorporated the individ- infarction. Rather than using such hand-crafted features, a deep ual tracings of the 3 readers. They did note significant variability neural network can be trained using initial DWI and PWI maps as among their readers’ segmentations, pointing out the importance input and final infarct size measured several days later as the out- of how the segmentation criterion standard is implemented. put. Using such a framework, Nielsen et al29 demonstrated that a deep learning architecture outperforms traditional state-of-the- Deep Learning for Image-Based Diagnosis art lesion prediction methods in acute stroke. They also showed The “Holy Grail” of machine learning in radiology is the so-called that a 37-layer architecture outperformed a shallower 3-layer ar- “end-to-end” solution, in which images are used as input and the chitecture, highlighting the importance of the depth feature of the output is a draft radiology report encompassing all the salient network. One exciting application of this approach is to train features of the image that an expert radiologist would include. As separate networks based on different treatments. In stroke, one improbable as this might seem, progress is being made on this could train networks in patients who received stroke treatment front with deep learning. Such approaches require tremendous and those who did not. New predictions using these 2 different amounts of annotated training data, which exist currently in models could give insight into whether treatment would lead to a many heterogeneous forms. Structured reporting, use of standard reduced infarct volume (Fig 6). lexicons (such as RadLex; http://www.rsna.org/RadLex.aspx), Automated segmentation of brain tumors, not just their en- and the standardization of electronic medical record platforms hancing margins, but also other features such as regions of en- are all steps that are enabling the formation of the required large hancement and necrosis, would be useful for a wide range of imaging/diagnosis datasets. This section will deal with some early indications, including diagnosis, presurgical planning, and fol- applications of deep learning to neuroradiologic diagnosis. low-up. The Brain Tumor Image Segmentation dataset is a pub- Gao et al33 classified 285 noncontrast brain CT examinations licly available dataset of brain tumor images with expert manual with a deep network into 1 of 3 categories (normal aging, lesion segmentations that has been a useful proving ground for new [such as tumor], or Alzheimer disease [AD]). With an average segmentation algorithms.30 The highest performance on the classification accuracy of 88%, the approach performed margin- Brain Tumor Image Segmentation dataset in 2016 was achieved ally better than other approaches that relied on hand-crafted fea- using a fully convolutional residual neural network, built on the tures. Another study34 showed that a multimodal stacked deep structure that won the 2015 ImageNet Challenge, with Dice coef- polynomial network using the ADNI dataset could classify pa- ficients for complete tumor, core tumor, and enhancing tumor tients into different binary groups (ie, AD versus healthy control between 0.72 and 0.87.31 In a separate study, Korfiatis et al32 [NC], or mild cognitive impairment converters [MCI-C] versus trained a deep autoencoder-decoder to segment T2-FLAIR le- nonconverters [MCI-nonconverters]). Instead of using images as sions on the Brain Tumor Image Segmentation dataset, which input, they used the volumes of 93 structures segmented from included manually drawn outlines in 186 patients. They then ap- T1-weighted images along with the [18F] FDG-PET signal inten- plied their model to a separate group of 135 patients with tumor in sity in these same regions. For distinguishing patients with AD 1782 Zaharchuk Oct 2018 www.ajnr.org from NCs, they showed an impressive area under the curve of how and why deep networks perform so well is an active area of 0.97. For the more challenging task of predicting MCI converters research in the artificial intelligence community. from nonconverters, they still showed areas under the curve of Also, the application of deep learning is still limited by require- Ͼ0.80. Suk et al35 showed that similar features could be combined ments for large amounts of annotated training datasets and the with CSF data in the ADNI dataset using a deep-weighted sparse challenge of keeping models current as source data and practice multitask learning framework to improve classification, showing patterns change. Among applications that are amenable to dis- 95% accuracy to distinguish patients with AD from NCs. How- ruption, it is still unclear which applications are supported by ever, when trying to predict among 3 classes (AD, NC, and MCI), sufficient clinical need to drive widespread adoption. Thus, it is the accuracy dropped to 63%, which further dropped to 54% important that radiologists remain engaged with artificial intelli- when trying to classify among 4 groups (AD, NC, MCI-C, and gence scientists to both understand the capabilities of existing MCI-nonconverter). This latter point speaks to the challenges of methods and direct future research in an intelligent way. Al- moving beyond simple binary classifications into the kinds of though this technology is still developing, state-of-the art deep tasks in which many diagnostic groups are possible, a situation learning models show little evidence that they could replace all familiar to neuroradiologists. functions of radiologists, though this issue is contentious. For the Another application is to use deep networks to identify the foreseeable future, they will likely serve as powerful image-pro- presence of hemorrhage on noncontrast brain CT. While general cessing and decision-support tools that will augment the accuracy machine learning techniques have been applied to these cases with and efficiency of radiologists. great success,36,37 only recently have deep learning methods been evaluated. Phong et al38 demonstrated a particularly interesting Outlook approach, using several “pretrained” deep networks as starting The wealth of applications of deep learning will likely lead to points for their training. Specifically, they used the optimal increased use of this technology. How fast this will happen de- weights from the GoogLeNet (http://deeplearning.net/tag/ pends on a few key factors: Deep learning can learn more from googlenet/) or Inception-ResNet (https://keras.rstudio.com/ larger datasets, usually by adding additional hidden layers. As reference/application_inception_resnet_v2.html) that were larger labeled datasets become available, the power of deep learn- trained on nonmedical images and then used their data to train a ing approaches will increase. The importance of data-sharing ini- final fully connected layer. This approach (called “transfer learn- tiatives such as ADNI and the Cancer Imaging Archive (https:// ing”) is attractive, given that it allows networks to be trained with public.cancerimagingarchive.net/ncia/legalRules.jsf) cannot be 39 overstated. Second, the computer hardware required to run these less data than if they were trained from scratch. Trained on 80 methods continues to improve and become less expensive. The cases and tested on 20 cases, they showed classification accuracies availability of open-source software frameworks, such as Caffe, of Ͼ98%. Tensorflow, PyTorch, and Keras, is greatly facilitating progress. Plis et al40 examined both structural and functional MR imag- However, before these methods become a routine part of clinical ing as input to deep networks for predicting various neurologic practice, vendors will need to provide “turn-key” systems that diseases. For structural imaging, they showed that they could dis- integrate well into current workflow patterns. As more neurora- tinguish patients with schizophrenia and Huntington disease diology researchers and practitioners become comfortable with from healthy subjects. For fMRI, they showed that deep networks these methods, through exposure at medical imaging conferences performed similar to independent component analysis for iden- and conversations with colleagues, applications will branch out tifying functional networks but tended to preserve edge details from the current “low-hanging fruit” to address more complex better. Similar results using a temporal-autoencoding neural net- and specialized questions. Therefore, we can expect further ad- work to predict the next time point in a resting-state fMRI time- vances in this field with accompanying benefits in many areas of series were applied to the Human Connectome Project data and radiology. showed a similar ability to identify task-specific networks.41

Disclosures: Greg Zaharchuk—UNRELATED: Grants/Grants Pending: GE Healthcare, Impact of Deep Learning on Neuroradiology Practice National Institutes of Health*; OTHER RELATIONSHIPS: Subtle Medical Inc, co- One concern is that if these approaches are successful, some work founder and equity relationship. Enhao Gong—UNRELATED: Board Membership: that radiologists have traditionally performed may become obso- GE Healthcare, Comments: research grant*; Patents (Planned, Pending or Issued): Subtle Medical Inc; Stock/Stock Options: Subtle Medical Inc. Max Wintermark— lete. Recently, a framework for inputting medical images (in this UNRELATED: Board Membership: GE-NFL Advisory Board. Daniel Rubin— case, pathology slides) and outputting diagnostic text-based re- UNRELATED: Grants/Grants Pending: National Institutes of Health*. Curtis P. ports has been reported.42 While this technology is still very rudi- Langlotz—OTHER RELATIONSHIPS: Montage Healthcare Solutions, Comments: founder, shareholder, board member; received consulting fees and travel reimburse- mentary, it is not difficult to imagine training a similar network ment. *Money paid to the institution. with CT scans and their reports. While deep learning does hold much promise to automate tasks that radiologists find unpleas- ant, ways of checking and verifying results will be needed. Another REFERENCES concern with deep learning is that we have little insight into the 1. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature 2015;521: inner workings of the models; they work well for prediction, but 436–44CrossRef Medline 2. Krizhevsky A, Sutskever I, Hinton G. ImageNet classification with precisely how they accomplish this is unclear. This contrasts with deep convolutional neural networks. http://www.cs.toronto.edu/ much prior radiology research, which relied heavily on domain ϳfritz/absps/imagenet.pdf. Accessed January 23, 2018 knowledge and the building of realistic models. Understanding 3. Mwangi B, Soares JC, Hasan KM. Visualization and unsupervised AJNR Am J Neuroradiol 39:1776–84 Oct 2018 www.ajnr.org 1783 predictive clustering of high-dimensional multimodal neuroimag- 24. Mehta R, Majumdar A, Sivaswamy J. BrainSegNet: a convolutional ing data. J Neurosci Methods 2014;236:19–25 CrossRef Medline neural network architecture for automated segmentation of 4. Hassabis D, Kumaran D, Summerfield C, et al. Neuroscience- human brain structures. J Med Imaging (Bellingham) 2017;4:024003 inspired artificial intelligence. Neuron 2017;95:245–58 CrossRef CrossRef Medline Medline 25. Chen H, Dou Q, Yu L, et al. VoxResNet: deep voxelwise residual 5. Hubel D, Wiesel T. Receptive fields, binocular interaction and func- networks for brain segmentation from 3D MR images. Neuroimage tional architecture of the cat’s visual cortex. J Physiol (London) 1962; 2018;170:446–55 CrossRef Medline 160:106–54 CrossRef Medline 26. Dou Q, Chen H, Yu L, et al. Automatic detection of cerebral microb- 6. Wang Z, Bovik A, Sheikh H, et al. Image quality assessment: from leeds from MR images via 3D convolutional neural networks. IEEE error visibility to structural similarity. IEEE Trans Image Process Trans Med Imaging 2016;35:1182–95 CrossRef Medline 2004;13:600–12 CrossRef Medline 27. Chen L, Bentley P, Rueckert D. Fully automatic acute ischemic le- 7. Zhou H, Gallo O, Frosio I, et al. Loss functions for image restoration sion segmentation in DWI using convolutional neural networks. with neural networks. IEEE Transactions on Computational Imaging Neuroimage Clin 2017;15:633–43 CrossRef Medline 2017;3:47–57 CrossRef 28. Maier O, Schro¨der C, Forkert ND, et al. Classifiers for ischemic 8. Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial stroke lesion segmentation: a comparison study. PloS One 2015;10: nets. https://papers.nips.cc/paper/5423-generative-adversarial-nets.pdf. e0145118 CrossRef Medline Accessed January 23, 2018 29. Nielsen A, Mouridsen K, Hansen M, et al. Deep learning: utilizing 9. Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to the potential in data bases to predict individual outcome in acute prevent neural networks from overfitting. J Machine Learning Re- stroke. In: Proceedings of the Annual Meeting of the International So- search 2014;15:1929–58 ciety for Magnetic Resonance in Medicine, Honolulu, Hawaii. April 10. Zhu B, Liu JZ, Rosen BR, et al. Image reconstruction by domain tra- 22–27, 2017;5665 nsform manifold learning. https://arxiv.org/pdf/1704.08841.pdf. Ac- 30. Menze BH, Jakab A, Bauer S, et al. The multimodal brain tumor cessed January 23, 2018 image segmentation benchmark (BRATS). IEEE Trans Med Imaging 11. Schlemper J, Caballero J, Hajnal JV, et al. A deep cascade of convolu- 2015;34:1993–2024 CrossRef Medline tional neural networks for dynamic MR image reconstruction. IEEE 31. Chang P. Fully convolutional deep residual neural networks for Trans Med Imaging 2018;37:491–503 CrossRef Medline brain tumor segmentation. In: Proceedings of the International Work- 12. Yang Y, Sun J, Li H, et al. Deep admm-net for compressive sensing shop on Brain Lesion: Glioma, Multiple Sclerosis, Stroke and Traumatic MRI. In: Proceedings of the 30th Conference on Neural Information Brain Injuries, Athens, Greece. October 17, 2016;108–18 Processing Systems, Barcelona, Spain. December 6–10, 2016 32. Korfiatis P, Kline TL, Erickson BJ. Automated segmentation of hy- 13. Wang S, Su Z, Ying L, et al. Accelerating magnetic resonance imaging perintense regions in FLAIR MRI using deep learning. Tomography via deep learning. In: Proceedings of the IEEE 13th International Sympo- 2016;2:334–40 CrossRef Medline sium on Biomedical Imaging, Prague, Czech Republic. April 9–16, 2016 33. Gao XW, Hui R, Tian Z. Classification of CT brain images based on 14. Dong C, Loy C, He K, et al. Learning a deep convolutional network deep learning networks. Comput Methods Programs Biomed 2017; for image super-resolution. In: Proceedings of the European Confer- 138:49–56 CrossRef Medline ence on Computer Vision, Zurich, Switzerland. September 6–12, 2014 34. Shi J, Zheng X, Li Y, et al. Multimodal neuroimaging feature learn- 15. Chen H, Zhang Y, Kalra MK, et al. Low-dose CT with a residual ing with multimodal stacked deep polynomial networks for diag- encoder-decoder convolutional neural network (RED-CNN). https:// nosis of Alzheimer’s disease. IEEE J Biomed Health Inform 2018;22: arxiv.org/pdf/1702.00288.pdf. Accessed January 23, 2018 173–83 CrossRef Medline 16. Bahrami K, Shi F, Zong X, et al. Reconstruction of 7T-like images from 35. Suk HI, Lee SW, Shen D; Alzheimer’s Disease Neuroimaging Initia- 3T MRI. IEEE Trans Med Imaging 2016;35:2085–97 CrossRef Medline tive. Deep sparse multi-task learning for feature selection in Alzhei- 17. Golkov V, Dosovitskiy A, Sperl JI, et al. Q-space deep learning: mer’s disease diagnosis. Brain Struct Funct 2016;221:2569–87 twelve-fold shorter and model-free diffusion MRI scans. IEEE Trans CrossRef Medline Med Imaging 2016;35:1344–51 CrossRef Medline 36. Scherer M, Cordes J, Younsi A, et al. Development and validation of 18. Gong E, Pauly J, Zaharchuk G. Boosting SNR and/or resolution of an automatic segmentation algorithm for quantification of intrace- arterial spin label (ASL) imaging using multi-contrast approaches rebral hemorrhage. Stroke 2016;47:2776–82 CrossRef Medline with multi-lateral guided filter and deep networks. In: Proceedings of 37. Al-Ayyoub M, Alawad D, Al-Darabsah K, et al. Automatic detection the Annual Meeting of the International Society for Magnetic Resonance and classification of brain hemorrhages. WSEAS Transactions on in Medicine, Honolulu, Hawaii. April 22–27, 2017 Computers 2013;10:395–405 19. Vemulapalli R, Nguyen H, Zhou S. Deep networks and mutual in- 38. Phong T, Duong H, Nguyen H, et al. Brain hemorrhage diagnosis by formation maximization for cross-modal medical image synthesis. using deep learning. In: Proceedings of the International Conference on In: Zhou S, Greenspan H, Shen D, eds. Deep Learning for Medical Machine Learning and Soft Computing, Ho Chi Minh City, Vietnam. Image Analysis. London: Elsevier; 2017 January 13–16, 2017 20. Liu F, Jang H, Kijowski R, et al. Deep learning MR-based attenu- 39. Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classifica- ation correction for PET/MRI. Radiology 2018;286:676–84 CrossRef tion of skin cancer with deep neural networks. Nature 2017;542: Medline 115–18 CrossRef Medline 21. Han X. MR-based synthetic CT generation using a deep convolu- 40. Plis SM, Hjelm DR, Salakhutdinov R, et al. Deep learning for tional neural network method. Med Phys 2017;44:1408–19 CrossRef neuroimaging: a validation study. Front Neurosci 2014;8:229 Medline CrossRef Medline 22. Li R, Zhang W, Suk HI, et al. Deep learning based imaging data 41. Lee JH, Wong E, Bandettini P. Temporal-autoencoding neural net- completion for improved brain disease diagnosis. In: Proceedings of work revealed the underlying functional dynamics of fMRI data: the International Conference on Medical Image Computing and Com- evaluation using the Human Connectome Project data. In: Proceed- puter-Assisted Intervention, Boston, Massachusetts. September 14–18, ings of the Annual Meeting of the International Society for Magnetic 2014;17:305–12 Resonance in Medicine, Honolulu, Hawaii. April 22–27, 2017 23. Wachinger C, Reuter M, Klein T. DeepNAT: deep convolutional 42. Zhang Z, Xie Y, Xing F, et al. MDNet: a semantically and visually neural network for segmenting neuroanatomy. Neuroimage 2018; interpretable medical image diagnosis network. http://arxiv.org/pdf/ 170:434–45 CrossRef Medline 1707.02485.pdf. Accessed January 23, 2018

1784 Zaharchuk Oct 2018 www.ajnr.org PRACTICE PERSPECTIVES

Comparison of Advanced Imaging Resources, Radiology Workforce, and Payment Methodologies between the United States and Canada

X H.A. Valand, X S. Chu, X R. Bhala, X R. Foley, X J.A. Hirsch, and X R.K. Tu

ABSTRACT SUMMARY: The purpose of this Practice Perspectives was to review the United States and Canadian approaches to health care access and payment for advanced imaging. The historical background, governmental role, workforce, coding, payment, radiologic challenges, cost, resource intensity, and overall outcomes in longevity are reviewed.

ABBREVIATION: BC ϭ British Columbia

lthough Canada and the United States share a 5425-mile bor- plan that covers all residents living in each jurisdiction for a min- Ader, common heritage, culture, and language,1 the health imum of 6–8 months annually, depending on the province or care system of each country, payment methodology, and radiolo- territory.6 The resident health card permits access to a full range of gist workforce differ and are detailed below. medical services, including diagnostic imaging at no or minimal out-of-pocket cost. Charging additional fees for medically neces- Health Systems sary services is strictly forbidden by the Canada Health Act. Por- tability of health care services exists for Canadians who travel or Canada. In the post–World War II era, Canada and the United work in other provinces, whereby their provincial health system States faced similar hospital and bed shortages. During the 1950s, compensates the delivering province at an agreed interprovincial Premier Tommy Douglas designed a hospitalization plan to ad- rate.7 dress these shortages in rural and low-income regions in the Ca- Six of 10 Canadian provinces prohibit private insurance from nadian province of Saskatchewan.2 The 1984 Canada Health Act covering services that are provided under the provincial health mandated that “medically necessary” comprehensive services be care plan, and 3 of the remaining 4 provinces allow private insur- provided in the provinces, largely at no cost.3 Today, Canadians in ance coverage of these services but at a considerable disadvantage all 10 provinces and 3 territories enjoy “reasonable access to med- to physicians who decide to opt out from public payers. For ex- ically necessary hospital and physician services without paying out of pocket.”4 Administration of health services, including ample, it is illegal for a physician in Nova Scotia to charge fees that imaging, remains the responsibility of the provinces and terri- exceed the provincial public rate. Saskatchewan and New Bruns- tories under the 5 principles of the Canada Health Act: public wick physicians are not reimbursed if they opt-out of the provin- administration, comprehensiveness, universality, portability, cial plan. Only Newfoundland and Labrador authorize private and accessibility.5 insurance coverage of medical services reimbursable by the public All provinces and territories have a publicly funded health plan without any economic disincentives to opt-out of the pro- vincial health plan.8 Many Canadians have supplemental insur- ance for uncovered services, such as prescription drugs, ambu- 9 Received May 16, 2018; accepted after revision June 14. lance services, vision, and dental care. From the American University of Integrative Sciences (H.A.V.), Brampton, Ontario, Unlike the United States, payment of Canadian physicians is Canada; Vancouver Coastal Health Authority (S.C.), Washington State Radiological generally restricted to specialty, as defined by the Royal College of Society, Vancouver, British Columbia, Canada; American Society of Neuroradiol- ogy (R.B.), Oak Brook, Illinois; Ontario Association of Radiologists (R.F.), Oakville, Physicians and Surgeons of Canada and recorded by the provin- Ontario, Canada; Massachusetts General Hospital (J.A.H.), Harvard Medical School, cial medical licensing authorities—that is, in Canada, only a radi- Boston, Massachusetts; and Progressive Radiology (R.K.T.), George Washington 10,11 University, United Medical Center, Falls Church, Virginia. ologist can provide professional radiology services. Profes- Please address correspondence to Raymond K. Tu, MD, MS, FACR, Progressive Ra- sional fees of physicians are negotiated between the province and diology, George Washington University, United Medical Center, 7799 Leesburg 12-18 Pike, Suite 1000N, Falls Church, VA 22043; e-mail: provincial medical association. Canadian CT and MR imag- [email protected] ing services are largely hospital-based performed in a hospital http://dx.doi.org/10.3174/ajnr.A5755 setting. Professional fees of Canadian radiologists vary between

AJNR Am J Neuroradiol 39:1785–90 Oct 2018 www.ajnr.org 1785 Table 1: Comparing vital and economic statistics of Canada versus Table 2: Examples of differing procedure codes and professional US in 2015a reimbursement for an unenhanced CT of the head in the 10 12–18,59–61 Canada US provinces of Canada Population30 35.8 million 320.9 million Professional Uninsured (% of total population)56 ϳ0%b 10% Province Fee Code Fee (CaD$) Average life expectancy (yr)26 82.1 78.7 British Columbia 08690 44.88 a Per capita health care spending25,27 $4734 $9994 Alberta b Total health care25,27 $170 billion $3.2 trillion Saskatchewan GDP25,57 $1.6 trillion $18.1 trillion Manitoba 7113 49.15 Total health care spending 11.4 17.7 Ontario X400 43.25 (% of GDP)25,57 Quebec 08259 33.40 New Brunswick 3166 67.00 Note:—GDP indicates gross domestic product. a Currency is in US dollars. Nova Scotia 1105 42.33 b According to the Wellesley Institute report, approximately 200,000–500,000 unin- Prince Edward Island 8925 83.16 sured individuals reside in Canada. They include students, workers from overseas, Newfoundland and Labrador 73800 55.53 58 undocumented refugees, and newly landed immigrants. a Benefits for noninvasive diagnostic procedures performed in hospitals and urgent care clinics are payable through hospitals or health authorities.12 the Provinces and Territories of Canada as they are independently b Not listed in the Payment Schedule for Insured Services Provided by a Physician, negotiated with the medical associations, Ministry of Health and October 1, 2017, published by the Government of Saskatchewan. providers.19 More than 90% of CT and MR imaging services are delivered In 2013, the radiology workforce in the United States was 8.14 in publicly funded hospitals in Canada. CT and MR imaging fees radiologists per 100,000 with 25,730 radiologists serving a total in Canadian provinces are either tied to the provincial fee sched- United States population of 320.9 million.28 In 2013, the Cana- ule or negotiated on a regional basis. The remaining CT and MR dian per capita radiology workforce was 6.82 with 2396 radiolo- imaging services are provided at private imaging clinics. Accord- gists, serving a Canadian population of 35.16 million.29,30 ing to the Canadian Agency for Drugs and Technologies in Health The Canadian procedure codes, including imaging, are unique 2017, there are approximately 561 CT and 366 MR imaging scan- and individualized to each province with updates as needed. Ca- ners in Canada.20 Additionally, there are approximately 59 true nadian fee codes for CT and MR imaging are organized along private payer MR imaging and CT facilities across Canada, located body systems and contrast (without contrast, with contrast, and in a subset of provinces in urban areas: British Columbia (BC) pre and post contrast). With the exceptions of cardiac CT and CT (n ϭ 17), Alberta (n ϭ 10), Quebec (n ϭ 31), and Nova Scotia colonography, Canadian radiologists are paid on a fee-for-service (n ϭ 1).21 In Ontario, there are approximately 184 CT scanners basis. (Table 2). and 120 MR imaging scanners, all in hospitals.20 In the United States, the imaging procedures are standardized by the 5-digit Current Procedural Terminology code, a method- United States. The United States has a multipayer system depen- ology published and owned by the American Medical Association dent on the private marketplace or government subsidy. Nongov- with standing meetings by the Current Procedural Terminology ernment plans include fee for service and prepaid plans. There are Editorial Panel 3 times a year.31,32 Medicare payments are based 6 government-based plans: Medicare, Medicaid, Children’s on a relative value unit formula weighted by the work of physi- Health Insurance Program, Tricare, Indian Health Service, and cians, professional liability insurance, and practice expense with a the Veterans Health Administration. Medicare plans are region- geographic variation modifier as defined by the resource-based alized into 7 Medicare Administrative Contractors, which stan- Relative Value Scale Update Committee, which also has standing dardize covered benefits and rates. There are 274 different man- meetings 3 times a year to evaluate updates and revisions, and, aged Medicaid plans, each with distinct coverage policies and when approved by the United States Congress, with a conversion reimbursement.22 In 1965, President Lyndon B. Johnson signed factor converting the relative value units into dollars.33,34 Medic- the bill that led to Medicare and Medicaid. Medicare is federally aid reimbursement and private insurance payments use the Medi- funded by tax revenue benefiting largely the elderly (65 years or care rates as a reference, but each plan individually determines the older) and individuals with disabilities. Medicaid is a federal and rate.35 state cofunded program for those with limited financial re- sources.23 There are 3 basic types of private health insurance plans Radiology Services in Canada in the United States: Health Maintenance Organizations, Pre- ferred Provider Organizations, and Point of Service Plans, which CT and MR Imaging (Advanced Imaging) Wait Times. More CT combine Health Maintenance and Preferred Provider Organiza- and MR imaging scanners have been approved in the past decade, tion features.24 but an undersupply continues to exist in every province. The combination of an inadequate number of scanners and under- Spending, Work Force, Coding, and Payment funding for the operation of these scanners has led to prolonged In 2015, the United States spent $3.2 trillion on total health care, wait times for both modalities, but especially MR imaging. This roughly $9994 per capita. Canada spent $170 billion (US dollars), issue is occurring with an increasing demand to meet longstand- approximately $4734 (in US dollars) per capita, a little more than ing indications as well as new clinical applications (eg, breast MR half that of the United States. Despite these expenditures, the av- imaging and CT colonography) driven by a growing and aging erage life expectancy in the United States was 78.7 years, whereas population. The Fraser Institute survey, based on physician rec- it was 82.1 years in Canada in 2015 (Table 1).25-27 ollection and self-reported methodology for “medically necessary 1786 Valand Oct 2018 www.ajnr.org Table 3: Comparison of CT scanner use in Canada versus US (2016)62 imaging and surgical centers. BC physi- Scans Canada US cians are not allowed to balance bill for Total CT use (hospital ϩ ambulatory ϩ other) services provided in a public hospital or Total No. (MM) 5.68 82 community care facility or charge extra Per 1000 people 156.6 253.8 technical/facility fees for “medically Per scanner 10,561.9 6062.8 44 CT use necessary services.” Enrolled doctors, CT in hospital setting however, may operate clinics for non- Total No. (MM) 5.61 65.7 medically necessary services such as Per 1000 people 154.4 203.3 cosmetic surgery. Per scanner 8973.8a 7373.7 CT in ambulatory care setting The Uninsured. All Canadians have Total No. 79,261 16.3 MM provincial government health insur- Per 1000 people 2.2 50.4 ance. There are approximately 28.2 mil- Per scanner 2735.3a 3532 lion uninsured United States citizens Note:—MM indicates million. 45 a Data were obtained in 2012. Remainder are from 2015. (Table 1), 9% of the United States population according to the 2016 Na- 62 Table 4: Comparison of MRI scanner use in Canada versus US (2016) tional Center for Health Statistics re- Scans Canada US port.46 In Canada, the Aboriginal popu- Total MRI use (hospital ϩ ambulatory ϩ other) lation has access difficulties; “medically Total No. (MM) 2.03 39 necessary” services for Aboriginals are Per 1000 people 55.9 120.7 Per scanner 5946.9 3287 the direct responsibility of the prov- MRI use inces, though the federal government MRI in hospital setting does provide some services for mental Total No. (MM) 1.8 16.5 health, chronic conditions, and pre- Per 1000 people 49.4 51.1 scription drugs.47 The confusing and Per scanner 6485.6a 3064.1 MRI in ambulatory care setting bureaucratic “patchwork” of coverage Total No. 234,308 22.5 MM and the often rural location of Aborigi- Per 1000 people 6.5 69.6 nals have led to challenging care for this a Per scanner 3170.9 3471.2 community. a Data were obtained in 2012. Remainder are from 2015. Professional Liability Insurance Cost. elective treatment,” reported average Canadian CT wait times Professional liability insurance costs differ between Canada and ranging from 3 to 6 weeks in 2016 among the provinces and av- the United States. California legislated a $250,000 cap on noneco- erage MR imaging wait times from 4 to 24 weeks.36 In Ontario nomic damages and fostered the low professional liability insur- (the most populous Canadian province with 38.5% of the popu- ance rate for radiology. State-filed malpractice premiums of diag- lation of the country), average wait times were 3 weeks for CT and nostic radiologists in California range from $5729 to $21,095, 6 weeks for MR imaging. The Canadian Institute for Health In- with an average of $10,934; low-risk physicians receive a discount formation, an agency of Health Canada that collects and reports up to 50% on state-filed rates.48 Greater than 90% of Canadian “essential information on Canada’s health systems and the health physicians are members of the Canadian Medical Protective As- of Canadians,”37 reported a median CT wait time ranging from 6 sociation, a mutual legal defense association that aims to decrease to 34 days and MR imaging wait times from 33 to 84 days. Median medical-legal risk for its members and improve the safety of CT wait times in Ontario were 6 days and 33 days for MR imag- health care. In BC, the annual fees (premiums) for diagnostic ing.38,39 Outpatients requiring nonurgent advanced imaging ser- radiologists are near the median with other Canadian physicians, vices in hospital radiology departments face significant backlogs; CaD$5280 in 2016 (ϳ$3830 US dollars).49 The actual fees paid however, urgent and emergent priority procedures are completed may be less because each provincial medical association negoti- within acceptable standards.40 The average wait time in Ontario is ates with the provincial government to subsidize a portion of the 94 days. At 5 days a week, that is ϳ19 weeks.41 Because of the high fees paid. In BC, the government rebated nearly half of the annual demand and an inadequate supply of available imagining ap- fee, CaD$2597 (ϳ$1884 US dollars), to radiologists. In the United pointments, the system is biased toward patients completing rou- States, the malpractice claim rate of radiologists is approximately tine imaging protocols versus patients requiring more time-in- 7%,50 compared with a Canadian malpractice claim rate of 1.9% tensive, advanced studies such as breast MR imaging or cardiac in 2013 per the Canadian Medical Protective Association report of CT, which have longer procedural times and longer wait lists.42 46 claims against radiologists.51 Nearly all 2396 radiologists in Canada were covered by the Canadian Medical Protective BC Government Initiative to Close Private Imaging and Surgical Association. Centers. The legality of private, self-pay surgical facilities, such as the BC Cambie Surgery Centre, is under current constitutional Advanced Imaging Equipment Funding and Use challenge. The government of BC has recently announced a desire to close private imaging centers in the province.43 That contrasts Funding of Radiology Equipment. In Canada, each provincial with some other provinces outside BC that permit private-pay ministry of health funds most hospital operating budgets, either AJNR Am J Neuroradiol 39:1785–90 Oct 2018 www.ajnr.org 1787 Table 5: Neuro CT procedures in 2016 in the US 63 average of 5900 examinations were per- CT Procedure Total CT % of All CT % of CT Sites formed each year in Canada, while this Categories Procedure (MM) Procedures Performing number was 3300 in the United States.55 Spine 6.4 8% 87% According to the Canadian Agency Brain 15.3 19% 94% for Drugs and Technologies in Health Head and neck 7.2 9% 90% Total neuro CT procedure 28.9 36% 2017 report, there were approximately Total CT procedures 82 100% 1 million neurologic CT procedures, Note:—Neuro indicates neurologic; MM, million. 19.8% of 5.68 million total CT proce- dures.20 The United States performs Table 6: All versus neuro CT and MR scans/1000 people 29.5 million neurologic CT procedures, Scans/1000 People Neuro CT All CT Neuro MRI All MRI 36% of 82 million (Table 5). Correcting a c Canada 31 156.6 14.2 55.9 for the population difference between US 96.4b 253.8 65.2d 120.7 the United States and Canada, the CT Note:—Neuro indicates neurologic. a According to the Canadian Agency for Drugs and Technologies in Health, total neurologic CT examinations were 19.8% use/1000 Canadians is 31 (19.8% of of all CT examinations in 2017. 156.6 all CT procedures/1000 Canadi- b Based on the Medical Information Division benchmark report, total neurologic (spine, brain, and head and neck) CT ans) versus 96.4 neurologic CT proce- examinations were 36% of all CT examinations in 2016. c According to the Canadian Agency for Drugs and Technologies in Health, total neurologic MRI examinations were dures/1000 Americans (Table 6). In 25.4% of all MR imaging examinations in 2017. comparison with Canada, the United d Based on the Medical Information Division benchmark report, total neurologic (spine, brain, and head and neck) CT States performs ϳ3 times as many neuro- examinations were 54% of all MR imaging examinations in 2016. logic CT examinations/1000 individuals. Table 7: Neuro MRI procedures in 2016 in the US64 Similarly, the neurologic MR imaging use/ MRI Procedure Total MRI % of All MRI % of MRI Sites 1000 individuals in Canada is 14.2 neuro- Categories Procedures (MM) Procedures Performing logic MR imaging/1000 Canadians versus Spine 9.5 24% 99% 65.2 neurologic MR imaging/1000 Amer- Brain (nonvascular) 8.8 22% 97% icans (4.6 times higher) (Tables 6 and 7). Head and neck 3.0 8% 84% Total neuro MRI procedure 21.3 54% Total MRI procedures 39 100% CONCLUSIONS Note:—MM indicates million. The American and Canadian systems both strive to provide health care to their directly or indirectly through Regional Health Authorities. Hos- citizens. The system in United States uses standardized coding pital imaging equipment purchases involve some combination of within a dual system of government and private payer health in- an approval of the health authority, ministry of health capital surance. The Canadian universal coverage is largely funded from equipment funding, or funds raised through the charitable foun- tax revenue collected by the federal legalization payments and dation of the hospital. The ministry/health authority controls op- provincial government but is independently administered by erating budgets and approval of capital purchases. In the provinces and territories using different coding and payment pol- United States, except for the Veterans Administration and icies. The Canadian health system, including advanced imaging, is large government-owned hospitals, manufacturers and pro- highly regulated and accounts for the per capita expenditure for viders negotiate imaging equipment costs directly with the health care in Canada being more than half that of the United hospitals and facilities.52 States. As the United States struggles with increasing health care expenditures, Canada is struggling with the demand for greater Advanced Imaging Use and Units. The variation in the volume of access. The United States population is 10 times that of Canada, advanced imaging procedures in Canada and the United States is with health care costs nearly twice as high per beneficiary. The marked (Tables 3 and 4). The Organization for Economic Co- explanation of less expensive care and longer outpatient proce- operation and Development 2016 reported 157 CT procedures dure wait times is complex, though perhaps due to a less ad- per 1000 individuals in Canada and 254 CT procedures per 1000 verse malpractice climate in Canada compared with the United individuals in the United States. The number of MR imaging States. Delivery in a nonprofit Canadian Government–con- scans in Canada was 56 per 1000, while in the United States, the trolled system may be an oversimplification; perhaps differing number was 121 per 1000 individuals. As of 2015, the 9.48 MR populations, poverty rates, and medical conditions served de- imaging units per million Canadian inhabitants was dwarfed by spite similar culture and origins may account for the apparent the United States ratio of 38.96 MR imaging units per million. discordance of health care cost and access. During the same year, Canada had 15.01 CT units per million individuals, compared with 40.98 units per million in the United 53,54 Disclosures: Joshua A. Hirsch—UNRELATED: Consultancy: Medtronic; Grants/ States The proportionally lower number of CT and MR im- Grants Pending: Harvey L. Neiman Health Policy Institute. aging units in Canada is accompanied by the need for greater numbers of procedures per unit per day. For CT, 10,600 exami- REFERENCES nations were performed on average in 2015 on each CT unit in 1. Canada FAQ. What countries does Canada have borders with? http:// Canada, while an average of 6100 examinations were performed www.canadafaq.ca/whatϩcountriesϩcanadaϩbordersϩwith/. Ac- on each CT unit in the United States. For MR imaging in 2015, an cessed April 10, 2018 1788 Valand Oct 2018 www.ajnr.org 2. Brown L, Tayolor D. The Birth of Medicare. Canadian Dimension. USE REPORT. http://www.cadth.ca/sites/default/files/pdf/canadian_ July 3, 2012. https://canadiandimension.com/articles/view/the-birth- medical_imaging_inventory_2017.pdf. Accessed April 3, 2018 of-medicare. Accessed March 27, 2018 21. Canadian Association for Health Services and Policy Research. The 3. Henderson JW. Health Economics and Policy. Cincinnati: South- Private Provision and Insurance of Diagnostic Imaging Services in Western College Pub; 1999:395 Canada. https://www.cahspr.ca/en/presentation/52413982f44a 4. Government of Canada. Canada’s health care system. About Medi- 7df3179aa9c3. Accessed May 6, 2018 care. https://www.canada.ca/en/health-canada/services/canada- 22. Kaiser Family Foundation. Total Medicaid MCOs. https://www. health-care-system.html. Accessed April 20, 2017 kff.org/medicaid/state-indicator/total-medicaid-mcos/?current 5. Justice Law’s Website-Government of Canada. Canada Health Act, Timeframeϭ0&sortModelϭ%7B%22colId%22:%22Location%22, RSC 1985, c. C-6. Updated January 2018. http://laws-lois.justice. %22sort%22:%22asc%22%7D. Accessed April 10, 2018 gc.ca/eng/acts/C-6/page-1.html#h-3. Accessed June 28, 2018 23. Medicare.gov. How is Medicare funded? Medicare Trust Funds. https:// 6. Government of Canada. A Canadian’s guide to working, studying, www.medicare.gov/about-us/how-medicare-is-funded/medicare- volunteering or retiring in a foreign country. https://travel.gc.ca/ funding.html. Accessed July 25, 2017 travelling/publications/living-abroad. Accessed April 30 2018 24. Snyder L, Rudowitz R. c May 20, 2015. http://www.kff.org/medicaid/ 7. Senate of Canada. The Health Canada Act. 17.4 Portability. https:// issue-brief/medicaid-financing-how-does-it-work-and-what-are-the- sencanada.ca/content/sen/committee/372/soci/rep/repoct02vol6 implications/. Accessed July 20, 2017 part7-e.htm. Accessed May 4, 2018 25. Canadian Institute for Health Information. National Health Expen- 8. Parliament of Canada. Madore O, Tiedemann M. Private Health Care diture Trends, 1975 to 2016. https://secure.cihi.ca/free_products/ Funding and Delivery Under the Canada Health Act. Updated De- NHEX-Trends-Narrative-Report_2016_EN.pdf. Accessed January cember 28, 2005. http://www.res.parl.gc.ca/Content/LOP/Research 27, 2018 Publications/prb0552-e.htm. Accessed January 31, 2018 26. The World Bank. Life Expectancy at Birth (years). https://data.world- 9. Government of Canada. Canada’s health care system. Updated Au- bank.org/indicator/SP.DYN.LE00.IN?locationsϭCA-US. Accessed gust 22, 2016. https://www.canada.ca/en/health-canada/services/ January 30, 2018 canada-health-care-system.html. Accessed March 30, 2018 27. Centers for Medicare & Medicaid Services Historical. Highlights. Na- 10. College of Physicians & Surgeons of British Columbia. Understand- tional Health Expenditure Data. Last Updated January 8, 2018. https:// ing Your Physician’s Credentials. https://www.cpsbc.ca/physician_ www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends- search/credentials. Accessed April 2, 2018 and-Reports/NationalHealthExpendData/NationalHealthAccounts 11. The College of Physicians and Surgeons of Ontario. What’s Public Historical.html. Accessed January 27, 2018 (and What’s Not) About Doctors. http://www.cpso.on.ca/Public- 28. Center for Disease Control. Table 84. Doctors of medicine, by place of Information-Services/Find-a-Doctor/What%E2%80%99s-Public- medical education and activity: United States and outlying U.S. areas, (and-What%E2%80%99s-Not)-About-Doctors. Accessed April 2, selected years 1975–2013. https://www.cdc.gov/nchs/data/hus/2015/ 2018 084.pdf. Accessed January 27, 2018 12. British Columbia Government. British Columbia Ministry of Health. 29. Canadian Medical Association. Number of Physicians by Province/Ter- Medical Services Commission Payment Schedule, July 1, 2017. ritory and Specialty, Canada, 2013. https://www.cma.ca/Assets/assets- https://www2.gov.bc.ca/assets/gov/health/practitioner-pro/medical- library/document/en/advocacy/policy-research/physician- services-plan/msc-payment-schedule-july-2017.pdf. Accessed Janu- historical-data/2013-01-spec-prov.pdf. Accessed June 12, 2017 ary 27, 2018 30. The World Bank. Population, Total. https://data.worldbank.org/ 13. Alberta Government. Updated Alberta Health Care Insurance Plan. indicator/SP.POP.TOTL?endϭ2016&locationsϭUS-CA&startϩϭ Medical Governing Rules List. April 1, 2017. https://open.alberta.ca/ 1960&viewϭchart&year_high_descϭfalse. Accessed December 7, 2017 dataset/8d90a766-e1d3-4b2c-a21f-4178cca65112/resource/1edd1e62- 31. Leslie Mazwi TM, Bello JA, Tu R, et al. Current procedural e8f7-43b4-acca-498a9b2874b7/download/somb-medical-rules-2017- terminology: history, structure, and relationship to valuation for 01.pdf. Accessed January 27, 2018 the neuroradiologist. AJNR Am J Neuroradiol 2016 Jul 14. [Epub 14. Government of Saskatchewan. Payment Schedule for Insured Services ahead of print] CrossRef Medline Provided by a Physician. Published October 1, 2017. https://www. 32. Hirsch JA, Leslie-Mazwi TM, Nicola GN, et al. Current procedural ehealthsask.ca/services/resources/Resources/physician-payment- terminology: a primer. J Neurointerv Surg 2015;7:309–12 CrossRef schedule-oct-17.pdf. Accessed November 28, 2017 Medline 15. Manitoba Physician’s Manual. The Minister of Health (Manitoba). 33. UCSF Medical Group. RVU & RBRVS Basics. http://medgroup. Published April 1, 2018. https://www.gov.mb.ca/health/documents/ ucsf.edu/data-warehouse/rvu-rbrvs-basics. Accessed February 18, physmanual.pdf. Accessed September 17, 2017 2018 16. Ontario Ministry of Health and Long-Term Care. Published Schedule 34. Hirsch JA, Silva E 3rd, Nicola GN, et al. The RUC: a primer for neu- of Benefits: Physician Services Under the Health Insurance Act, De- rointerventionalists. J Neurointerv Surg 2014;6:61–64 CrossRef cember 22, 2015 (effective March 1, 2016). http://www.health.gov. Medline on.ca/en/pro/programs/ohip/sob/physserv/sob_master20160401. 35. Centers for Medicare & Medicaid Services. Final Policy, Payment, and pdf. Accessed May 1, 2018 Quality Provisions in the Medicare Physician Fee Schedule for 17. Re´gie de l’assurance maladie (Que´bec). Manuel des Me´decins Spe´- Calendar Year 2018. https://www.cms.gov/Newsroom/MediaRelease cialistes-Re´mune´ration a` l’acte. http://www.ramq.gouv.qc.ca/Site Database/Fact-sheets/2017-Fact-Sheet-items/2017-11-02.html. Ac- CollectionDocuments/professionnels/manuels/syra/medecins- cessed April 22, 2018 specialistes/150-facturation-specialistes/manuel-specialistes- 36. Barua B, Ren F. Waiting Your Turn: Wait Times for Health Care in remuneration-acte-RFP.pdf. Accessed November 23, 2017 Canada, 2016 Report. Fraser Institute. https://www.fraserinstitute. 18. Nova Scotia Medical Services Insurance. Physician’s Manual 2014. org/studies/waiting-your-turn-wait-times-for-health-care-in-canada- http://msi.medavie.bluecross.ca/wp-content/uploads/sites/3/2015/ 2016. Accessed December 5, 2017 07/PhysicianManual.pdf. Accessed November 27, 2017 37. Canadian Institute for Health Information. CIHI Overview Board of 19. Canadian Medical Association. Module 8: Physician Remuneration Directors Governance Handbook, 2017. https://www.cihi.ca/sites/ Options. https://www.cma.ca/Assets/assets-library/document/ default/files/document/governance-handbook-update-aug-2017- en/practice-management-and-wellness/module-8-physician- en-web.pdf. Accessed December 5, 2017 remuneration-options-e.pdf. Accessed on May 6, 2018 38. Canadian Institute for Health Information. Wait Times for CT Scan 20. Ottawa: Canadian Agency for Drugs and Technologies in Health. The (2017). http://waittimes.cihi.ca/procedure/ctscan?showϭ5090#year. Canadian Medical Imaging Inventory, 2017; CADTH OPTIMAL Accessed January 23, 2018 AJNR Am J Neuroradiol 39:1785–90 Oct 2018 www.ajnr.org 1789 39. Canadian Institute for Health Information. Wait Times for MRI Scan perspective/com_17_perspective_march-e.pdf. Accessed October 25, (2017). http://waittimes.cihi.ca/procedure/mri?showϭ5090#year. 2017 Accessed January 23, 2018 52. Auditor General of British Columbia. Health Funding Explained 2. 40. Canadian Institute for Health Information. Wait Times for Priority March 2017. http://www.bcauditor.com/sites/default/files/publications/ Procedures in Canada, 2017. https://www.cihi.ca/sites/default/files/ reports/FINAL_HFE2_2.pdf. Accessed March 31, 2018 document/wait-times-report-2017_en.pdf. Accessed May 28, 2018 53. OECD. Magnetic resonance imaging (MRI) units (indicator). https:// 41. Ministry of Health and Long-term Care. Surgery and Diagnostic Im- data.oecd.org/healtheqt/magnetic-resonance-imaging-mri-units. aging Results. Diagnostic Scans. http://www.ontariowaittimes. htm. Accessed January 30, 2018 com/SurgeryDI/EN/Data.aspx?viewϭ1&Typeϭ0&Modalityϭ3& 54. OECD. Computed tomography (CT) scanners. https://www.oecd- ModalityTypeϭ22,25&cityϭtoronto&pcϭ&distϭ0&hosptIDϭ ilibrary.org/social-issues-migration-health/computed-tomography-ct- 0&strϭ&periodϭ0&expandϭ0. Accessed April 22, 2018 scanners/indicator/english_bedece12-en. Accessed January 30, 2018 42. Legislative Assembly of Ontario. Ontario Best Practice Guidelines for 55. Diagnostic Exams. Organisation for Economic Co-operation and Devel- Managing the Flow of Patients Requiring an MRI or CT Examination. opment. Health Care Utilisation. https://stats.oecd.org/index.aspx? http://www.ontla.on.ca/library/repository/mon/23006/292809.pdf. Ac- queryidϭ24879. Accessed April 12, 2017 cessed May 6, 2018 56. Kaiser Family Foundation. Uninsured Rates for the Nonelderly by 43. British Columbia Government. Fair access to public health care for all Race/Ethnicity. https://www.kff.org/uninsured/state-indicator/rate-by- British Columbians. https://news.gov.bc.ca/releases/2018HLTH0027- raceethnicity/?dataViewϭ0¤tTimeframeϭ1&sortModelϭ 000559. Accessed May 6, 2018 %7B%22colId%22:%22Location%22,%22sort%22:%22asc%22%7D. 44. Medicare Protection Act, RSBC 1996, c. 286, ss. 18(1), 18(2). http:// Accessed December 7, 2017 www.bclaws.ca/civix/document/id/complete/statreg/E3tlc96286. Ac- 57. The World Bank. GDP (Current US$). https://data.worldbank.org/ cessed April 22, 2018 indicator/NY.GDP.MKTP.CD?endϭ2015&locationsϭCA-US& 45. Kaiser Family Foundation. Key Facts about the Uninsured Popula- startϭ2015&viewϭbar. Accessed January 26, 2018 tion. September 19, 2017. https://www.kff.org/uninsured/fact-sheet/ 58. Barnes S. Health Care Access for the Uninsured in Ontario: Symposium key-facts-about-the-uninsured-population/. Accessed October 20, Report. Wellesley Institute. http://www.wellesleyinstitute.com/wp- 2017 content/uploads/2017/01/Health-Care-Access-for-the-Uninsured- 46. Clarke TC, Norris T, Schiller JS. Early Release of Selected Estimates Symposium-Report.pdf. Accessed April 5, 2018 Based on Data from the 2016 National Health Interview Survey. Di- 59. Government of New Brunswick. New Brunswick Physicians’ Manual. vision of Health Interview Statistics, National Center for Health May 18, 2017. http://www2.gnb.ca/content/dam/gnb/Departments/ Statistics. https://www.cdc.gov/nchs/data/nhis/earlyrelease/early h-s/pdf/en/Physicians/new_brunswick_physicians_manual.pdf. Ac- release201705.pdf. Accessed July 20, 2017 cessed April 8, 2018 47. Palmer K, Tepper J, Nolan M. Indigenous Health Services Often 60. Master Agreement between the Medical Society of Prince Edward Island Hampered by Legislative Confusion. healthydebate. September 21, and the Government of Prince Edward Island and Health PEI: April 1, 2017. http://healthydebate.ca/2017/09/topic/indigenous-health. Ac- 2015–March 31, 2019. http://www.gov.pe.ca/photos/original/doh_ cessed January 24, 2018 masteragree.pdf. Accessed April 22, 2018 48. California’s Medical Malpractice Insurance History. https://www. 61. Medical Care Plan, Dept. of Health & Community Services (Newfound- gallaghermalpractice.com/state-resources/california-medical- land & Labrador). Medical Payment Schedule. June 1, 2013. http:// malpractice-insurance/. Accessed January 24, 2018 www.health.gov.nl.ca/health/mcp/providers/Full_MCP_Payment_ 49. Doctors of BC. 2016 CMPA Rebates. https://www.doctorsofbc.ca/ Schedule-03_13_14.pdf. Accessed April 8, 2018 sites/default/files/2016cmpatable.pdf. Accessed April 3, 2018 62. OECD.Stat. Health Status. http://stats.oecd.org/index.aspx?Data 50. Jena AB, Seabury S, Lakdawalla D, et al. Malpractice risk accord- SetCodeϭHEALTH_STAT. Accessed December 6, 2017 ing to physician specialty. N Engl J Med 2011;365:629–36 63. IMV Medical Information Division. IMV 2016 CT Benchmark Report. CrossRef Medline Des Plaines; IMV; Accessed March 14, 2018 51. Perspective. Canadian Medical Protective Association. March 2017. 64. IMV Medical Information Division. IMV 2016 MR Benchmark Re- https://www.cmpa-acpm.ca/static-assets/pdf/advice-and-publications/ port. Des Plaines; IMV; Accessed March 4, 2018

1790 Valand Oct 2018 www.ajnr.org ORIGINAL RESEARCH ADULT BRAIN

Clinical Value of Hybrid TOF-PET/MR Imaging–Based Multiparametric Imaging in Localizing Seizure Focus in Patients with MRI-Negative Temporal Lobe Epilepsy

X K. Shang, X J. Wang, X X. Fan, X B. Cui, X J. Ma, X H. Yang, X Y. Zhou, X G. Zhao, and X J. Lu

ABSTRACT BACKGROUND AND PURPOSE: Temporal lobe epilepsy is the most common type of epilepsy. Early surgical treatment is superior to prolonged medical therapy in refractory temporal lobe epilepsy. Successful surgical operations depend on the correct localization of the epileptogenic zone. This study aimed to evaluate the clinical value of hybrid TOF-PET/MR imaging–based multiparametric imaging in localizing the epileptogenic zone in patients with MR imaging-negative for temporal lobe epilepsy.

MATERIALS AND METHODS: Twenty patients with MR imaging-negative temporal lobe epilepsy who underwent preoperative evaluation and 10 healthy controls were scanned using PET/MR imaging with simultaneous acquisition of PET and arterial spin-labeling. On the basis of the standardized uptake value and cerebral blood flow, receiver operating characteristic analysis and a logistic regression model were used to evaluate the predictive value for the localization. Statistical analyses were performed using statistical parametric mapping. The values of the standardized uptake value and cerebral blood flow, as well as the asymmetries of metabolism and perfusion, were compared between the 2 groups. Histopathologic findings were used as the criterion standard.

RESULTS: Complete concordance was noted in lateralization and localization among the PET, arterial spin-labeling, and histopatho- logic findings in 12/20 patients based on visual assessment. Concordance with histopathologic findings was also obtained for the remaining 8 patients based on the complementary PET and arterial spin-labeling information. Receiver operating characteristic analysis showed that the sensitivity and specificity of PET, arterial spin-labeling, and combined PET and arterial spin-labeling were 100% and 81.8%, 83.3% and 54.5%, and 100% and 90.9%, respectively. When we compared the metabolic abnormalities in patients with those in healthy controls, hypometabolism was detected in the middle temporal gyrus (P Ͻ .001). Metabolism and perfusion asymmetries were also located in the temporal lobe (P Ͻ .001).

CONCLUSIONS: PET/MR imaging–based multiparametric imaging involving arterial spin-labeling may increase the clinical value of local- izing the epileptogenic zone by providing concordant and complementary information in patients with MR imaging-negative temporal lobe epilepsy.

ABBREVIATIONS: AI ϭ asymmetry index; ASL ϭ arterial spin-labeling; EZ ϭ epileptogenic zone; FCD ϭ focal cortical dysplasia; HS ϭ hippocampal sclerosis; SPM ϭ statistical parametric mapping; SUV ϭ standardized uptake value; SUVr ϭ standardized uptake value ratio; TLE ϭ temporal lobe epilepsy

pilepsy is a common chronic neurologic disorder character- to prolonged medical therapy in refractory TLE.2 Successful Eized by recurrent spontaneous seizures. It has an incidence operations depend on the correct localization of the epilepto- of 50 per 100,000 persons per year.1 Temporal lobe epilepsy genic zone (EZ). MR imaging is a powerful tool in identifying (TLE) is the most common type of epilepsy. A published ran- the lesions causing epilepsy, such as hippocampal sclerosis domized trial reported that early surgical treatment is superior (HS). However, approximately 16% of patients with TLE have a normal MR imaging appearance.3 Histopathologic studies have shown that many focal cortical dysplasias (FCDs) are Received April 4, 2018; accepted after revision July 18. small or subtle and are difficult to identify visually using MR From the Departments of Nuclear Medicine (K.S., J.W., B.C., J.M., H.Y., J.L.), Neuro- 4 surgery (X.F., G.Z.), and Radiology (J.L.), Xuanwu Hospital, Capital Medical Univer- imaging. FCDs are identified as the most common histo- sity, Beijing, China; and Department of Radiology (Y.Z.), Johns Hopkins University, Baltimore, Maryland. Please address correspondence Jie Lu, MD, Department of Nuclear Medicine, Xu- This work was supported by the National Natural Science Foundation of China anwu Hospital, Capital Medical University, 45 Changchunjie, Xicheng District, Bei- (Grant No. 81522021) and the National Key Research and Development Program of jing 100053, China; e-mail: [email protected] China (Grant No. 2016YFC0103000). Indicates open access to non-subscribers at www.ajnr.org Paper previously presented, in part, at: Annual Meeting of the Society of Nuclear Medicine and Molecular Imaging, June 10–14, 2017; Denver, Colorado. http://dx.doi.org/10.3174/ajnr.A5814

AJNR Am J Neuroradiol 39:1791–98 Oct 2018 www.ajnr.org 1791 Table 1: Study population demographics Age Age at Epilepsy Side of Patient No. Sex (yr) Onset (yr) Duration (yr) Seizure Frequency Operation Histopathology 1 F 26 14 12 4–5 Times/mo Left Left temporal region (FCD type I) 2 M 28 7 21 1–2 Times/day Left Left temporal region (FCD type I) 3 M 31 26 5 2–4 Times/mo Left Left temporal region (FCD type I) 4 M 23 14 9 2–5 Times/mo Left Left temporal region (FCD type Ib) 5 M 29 15 14 6–8 Times/mo Left Left temporal region (FCD type IIIa-HS) 6 F 24 12 12 2–3 Times/day Left Left temporal region (FCD type IIIa-HS) 7 F 38 17 21 3–5 Times/mo Left Left temporal region (FCD type IIIa-HS) 8 F 27 2 25 2–3 Times/mo Right Right temporal region (FCD type IIIa-HS) 9 M 23 16 7 3–4 Times/mo Right Right temporal region (FCD type I) 10 F 14 8 6 2–3 Times/day Left Left temporal region (FCD type I) 11 M 16 10 6 6–9 Times/mo Left Left temporal region (FCD type I) 12 M 26 14 12 3–4 Times/mo Right Right temporal region (FCD type I) 13 M 29 5 24 1 Time/day Right Right temporal region (FCD type I) 14 F 16 13 3 1 Time/mo Right Right temporal region (FCD type Ib) 15 M 24 11 13 2–4 Times/mo Right Right temporal region (FCD type IIIa-HS) 16 M 35 12 23 3–5 Times/mo Right Right temporal region (FCD type Ic) 17 M 19 14 5 3 Times/mo Right Right temporal region (FCD type Ib) 18 M 22 15 7 7–8 Times/mo Right Right temporal region (FCD type Ic) 19 M 17 1 16 1–2 Times/mo Left Left temporal region (FCD type Ic) 20 F 52 12 40 2–3 Times/mo Left Left temporal region (FCD type Ic) pathologic abnormality in MR imaging-negative TLE.5 There- the diagnostic value. However, as far as we know, no study has fore, additional functional imaging is needed for localization. revealed the relationship between metabolism and perfusion The increase in neuronal activity triggered by an epileptic sei- with the EZ confirmed by histopathology in hybrid PET/MR zure is associated with an increase in neuronal metabolism and imaging. This pilot study aimed to investigate the clinical value regional blood flow.6 Therefore, analyses of cerebral perfusion of localizing the EZ using hybrid TOF-PET/MR imaging– and the metabolic status are widely used in presurgical evalu- based multiparametric imaging in patients with MR imaging- ation to identify the EZ.7,8 negative TLE. [18F] FDG-PET reveals areas of interictal cerebral hypome- tabolism associated with epileptic activity and epileptogenic lesions and is being used for presurgical evaluation of the EZ.9 MATERIALS AND METHODS PET has been shown beneficial in patients with MR imaging Subjects negative epilepsy or nonconcordant electroencephalography Twenty patients (13 males and 7 females; 26.5 Ϯ 9.48 years of and neuroimaging findings.10,11 Moreover, FDG-PET-posi- age) who were diagnosed with medically refractory TLE and tive, MR imaging-negative TLE patients had excellent surgical who underwent standard anterior temporal lobectomy with en outcomes.12,13 Arterial spin-labeling (ASL) MR imaging is a bloc resection between September 2015 and April 2017 at Xu- noncontrast perfusion technique increasingly used to evaluate anwu Hospital were retrospectively evaluated. The inclusion the brain cerebral blood flow (CBF). It uses the magnetically criteria were as follows: 1) TLE confirmed by histopathology, labeled arterial blood as an endogenous contrast agent. Previ- 2) no structural abnormalities revealed on MR images, and 3) ous studies have shown increased perfusion during the peri- no seizure for at least 72 hours before PET/MR imaging. The ictal period, decreased perfusion during the postictal period, exclusion criteria were as follows: 1) abnormalities observed in and hemispheric hemodynamic asymmetry in patients with MR images, and 2) serious medical or nonepilepsy neurologic epilepsy.14,15 Thus, ASL perfusion MR imaging might help to disorders. All patients received the same standardized preop- confirm the location and extent of the EZ in the presurgical erative assessment protocol, including a detailed clinical and work-up of epilepsy.16,17 medical history, neuropsychologic examination, long-term ic- Previous studies were obtained with separate PET, MR imag- tal and interictal scalp electroencephalography surveillance, ing, and coregistered PET/MR imaging. Hybrid PET/MR imaging and anatomic and functional neuroimaging studies, which in- provides simultaneous acquisition in the same physiologic or cluded a brain MR imaging, PET/CT, and hybrid PET/MR im- pathophysiologic states and may be advantageous in various neu- aging. Patient demographics, clinical characteristics, surgical rologic disorders.18,19 A few studies with PET/MR imaging that approach, and histopathologic findings are summarized in evaluated refractory focal epilepsy have been reported.20-22 Table 1. Ding et al20 showed specific patterns of metabolic abnormality For comparison, 10 healthy controls were recruited (6 men and asymmetry in patients with epilepsy, which may help to and 4 women, 44.1 Ϯ 8.0 years of age). All healthy controls were understand the etiopathogenesis. Shin et al21 evaluated the im- right-handed with normal brain MR imaging findings. No famil- proved accuracy of hybrid PET/MR imaging compared with ial or personal history of neurologic or psychiatric diseases was separate MR imaging and PET/CT in localizing the EZ. Boscolo noted. The clinical study was approved by the ethics committee of Galazzo et al22 reported a high level of correlation between PET Xuanwu Hospital, and written informed consent was obtained and ASL and found that the statistical approach could improve from all participants before the study. 1792 Shang Oct 2018 www.ajnr.org Surgical Approach and Outcome with abnormalities of metabolism and perfusion. Any disagree- All patients underwent standard anterior temporal lobectomy, ment between the 2 observers was resolved by consulting a third including 11 who had an operation on the left side and 9 who had reader to reach a final consensus. an operation on the right side. Nine patients were evaluated after a postoperative follow-up period of at least 1 year. The Engel Statistical Parametric Mapping Analysis of PET and ASL classification was used in the evaluation of surgical outcomes Data (class I: free of disabling seizures; class II: rare disabling seizures All standardized uptake value (SUV) and CBF images were pro- [almost seizure-free]; class III: worthwhile improvement; and cessed and analyzed using statistical parametric mapping (SPM8; class IV: no worthwhile improvement).23 There were 7 patients http://www.fil.ion.ucl.ac.uk/spm/software/spm12) and Matlab with Engel class I and 2 with Engel class II results. 7.14 (MathWorks, Natick, Massachusetts) with a dedicated in- house code developed for this study. PET images were spatially Histopathologic Examination and Findings normalized to the Montreal Neurological Institute template The laboratory protocols used for specimen preparation were space. The images were then smoothed using an isotropic similar in all samples as described previously.24 Tissue sections Gaussian kernel with a full width at half maximum of 8 mm for were fixed overnight in 10% buffered formalin and then orien- all directions. PET data in the Montreal Neurological Institute tated and cut perpendicularly to the cortical surface. Following space were transformed into maps representing the standard- routine paraffin embedding, 4- or 8-␮m-thick sections were ized uptake value ratio (SUVr), which were normalized by scal- stained with hematoxylin-eosin and Luxol fast blue. The selected ing to a common value (50) for all scans. Similarly, each CBF sections were also tested for immunohistochemical reactions. map was registered to a specific template in the Montreal Neu- Histopathologic diagnoses were made by an experienced neuro- rological Institute space with a resolution of 2 ϫ 2 ϫ 2mm3 pathologist. There were 15 cases of FCD type I and 5 of FCD type using a nonlinear registration and then smoothed using an III–HS. 8-mm full width at half maximum Gaussian kernel. According to the histopathologic findings, FDG-PET SUV images and PET/MR Imaging Acquisition ASL MR imaging perfusion CBF maps of the right hemisphere Each subject was scanned with an imaging protocol consisting of needed to be left-right flipped for further data analysis to en- injecting FDG (mean, 264.0 Ϯ 46.8 MBq), with the scan being sure a homogeneous group with all patients having the EZ on initiated 40 minutes after the injection. All studies were per- the same side. formed on an integrated simultaneous Signa PET/MR imaging system (GE Healthcare, Milwaukee, Wisconsin). The PET bed Analysis of the Asymmetry Index of FDG-PET and ASL MR position included a simultaneous 18-second 2-point Dixon scan Imaging for MR imaging–based attenuation correction as well as addi- An asymmetry index (AI), which was used for detecting left-right tional diagnostic MR images. The attenuation map for the head asymmetries, was calculated for the FDG-PET SUV and ASL MR frame was created with atlas-based methods, in which a coregis- imaging CBF based on the following equation: AI ϭ 2 (Left Ϫ tered MR imaging–CT atlas dataset was used to derive a pseu- Right) / (Left ϩ Right).25 The normalization of the SUV and CBF do-CT image from the patient’s MR image. images allowed identification of left-right asymmetries in the ce- The ordered subsets expectation maximization algorithm rebral hemispheres; voxelwise AIs for both the SUV and CBF was used for PET image reconstruction, and the detailed pa- maps of each patient were then calculated. After we calculated the ␮ rameters were the following: 8 iterations, 32 subsets, and full mean ( AI) and SD (SDAI) of the overall AI map, a voxelwise AI z ϭ Ϫ ␮ width at half maximum of a Gaussian filter of 3.0 mm. The PET score map (ZAI) was derived as follows: ZAI (ValueAI AI)/ ϫ 22 images were reconstructed to a matrix of 192 192, and the SDAI. The metabolic and perfusion asymmetries were evaluated slice thickness was 2.44 mm. Imaging parameters of the ASL by referring to the normal distribution values of the healthy perfusion MR images were as follows: TR ϭ 4852 ms, TE ϭ controls. 10.7 ms, TI ϭ 2025 ms, FOV ϭ 24 ϫ 24 cm2, gap ϭ 0 mm, matrix size ϭ 512 ϫ 8, postlabel delay ϭ 2 seconds. CBF maps Statistical Analysis were generated on-line from the console using an AW4.6 A ␬ test was used to determine the degree of concordance be- Workstation (GE Healthcare). We also obtained routine ana- tween the 2 readers. Correlation analysis between regional tomic acquisitions: axial T2-weighted fast spin-echo (TR ϭ SUVr and CBF was used, and the Spearman rank correlation 9600 ms, TE ϭ 149 ms, matrix size ϭ 256 ϫ 256, slice thick- coefficient was obtained. On the basis of the values of the SUVr ness ϭ 3.0 mm, gap ϭ 1.0 mm); axial T1-weighted fast spin- and CBF, receiver operating characteristic analysis was used to echo (TR ϭ 3286 ms, TE ϭ 24 ms, matrix size ϭ 288 ϫ 256, assess the predictive value of each parameter. The highest area slice thickness ϭ 3.0 mm, gap ϭ 1.0 mm). under the curve and sensitivity and specificity were obtained. A logistic regression model with stepwise regression was used to Visual Inspection select the optimal model for the location of the EZ. Statistical All PET and CBF maps were evaluated in a blinded manner by at analyses were performed using SPSS 21.0 (IBM, Armonk, New least 2 experienced neuroradiologists independently, without any York). A P value Ͻ .05 was considered statistically significant. knowledge of the patients, operations, histopathology, or fol- For statistical parametric mapping (SPM) analysis, the low-up results. They were requested to identify the brain lobes 2-sample t test was used for comparing the difference in SUV, AJNR Am J Neuroradiol 39:1791–98 Oct 2018 www.ajnr.org 1793 Table 2: Summary of findings of PET and ASL in hybrid PET/MR imaging and histopathology Patient No. PET Findings in PET/MR Imaging ASL Findings in PET/MR Imaging Histopathology 1 Normal Left temporal Left temporal region (FCD type I) 2 Left parietal and temporal Left frontal, parietal, and temporal Left temporal region (FCD type I) 3 Left temporal Negative Left temporal region (FCD type I) 4 Left temporal Left temporal Left temporal region (FCD type Ib) 5 Left temporal Negative Left temporal region (FCD type IIIa-HS) 6 Left temporal Left temporal Left temporal region (FCD type IIIa-HS) 7 Left frontal, parietal, and temporal Left frontal, parietal, and temporal Left temporal region (FCD type IIIa-HS) 8 Right temporal Negative Right temporal region (FCD type IIIa-HS) 9 Right temporal Right temporal Right temporal region (FCD type I) 10 Left temporal Left temporal Left temporal region (FCD type I) 11 Left frontal, parietal, and temporal Left frontal, parietal, and temporal Left temporal region (FCD type I) 12 Right temporal Right temporal Right temporal region (FCD type I) 13 Left frontal and right temporal Right temporal Right temporal region (FCD type I) 14 Right temporal Negative Right temporal region (FCD type Ib) 15 Right temporal Right temporal Right temporal region (FCD type IIIa-HS) 16 Left temporal Left temporal Right temporal region (FCD type Ic) 17 Right temporal Right temporal Right temporal region (FCD type Ib) 18 Right temporal Right temporal Right temporal region (FCD type Ic) 19 Right temporal Negative Left temporal region (FCD type Ic) 20 Left temporal Negative Left temporal region (FCD type Ic)

CBF, and AI in the whole brain between patients and healthy controls. Hypometabolism, hypoperfusion, and AI were re- garded as statistically significant using an uncorrected P value of Ͻ .001 with a minimum cluster size of 50 voxels.

RESULTS The FDG-PET, ASL perfusion MR imaging, and histopathologic findings are summarized in Table 2. The concordance between the 2 neuroradiologists in the interpretation of the PET and ASL CBF maps was good (␬ ϭ 0.86, 0.74).

Comparison of Histopathologic Findings with PET and ASL Hypometabolism was observed by visual assessment of PET im- ages in 19/20 patients. Concordant lateralization with the EZ was confirmed by histopathology in 18/20 patients, and 15/18 patients showed focal hypometabolism concordant with the localization of the EZ. The remaining 3/18 patients had at least 1 hypometa- FIG 1. Receiver operating characteristic curves for SUVr, CBF, and bolic area, which showed partial concordance with histopatho- combined SUVr and CBF to predict EZ. The combined PET and ASL logic findings. obtain the highest area under the curve (0.970) with high sensitivity (100%) and specificity (90.9%). PET has more diagnostic information In ASL perfusion MR imaging, the readers agreed on the pres- with an area under the curve of 0.926, compared with ASL (area under ence of hypoperfusion in 14 of 20 patients. In the 14 patients, 11 the curve of 0.679). The combined PET and ASL show the best per- were in complete agreement with the localization of the EZ. In the formance in specificity for predicating EZ. other 3 patients, ASL and histopathologic findings had partial concordance with 1 common lobe identified. Six of 8 patients (patients 3, 5, 8, 14, 19, and 20) showed The best performance was FDG-PET SUVr with an area under normal perfusion in ASL images; however, concordant hypo- the curve of 0.932, a sensitivity of 100%, and a specificity of 81.8%, metabolic areas in PET images were observed. A representative followed by ASL perfusion MR imaging CBF with an area under patient with normal finding in ASL image and hypometabo- the curve of 0.636, a sensitivity of 83.3%, and a specificity of 54.5% lism in PET image is shown in Fig 3. In addition, patient 1 (Fig 1). showed a PET image with findings that seemed normal on visual inspection; however, focal hypoperfusion was identified Evaluation of Combined FDG-PET and ASL MR Imaging in as the EZ by histopathology. Bilateral brain regions (right tem- the Location of EZ poral and left frontal lobe) with hypometabolism were seen in A complete concordance in lateralization and localization patient 13. However, only the right temporal lobe with hypo- among the PET, ASL, and histopathologic results was found in perfusion was identified as the EZ (Fig 4). 12/20 patients. A representative image set is shown in Fig 2. Scatterplots of the mean SUVr and CBF values from FDG-PET Concordance of histopathologic results with either FDG-PET and ASL MR imaging in a series of ROIs with hypometabolism or ASL MR imaging was also obtained in the remaining 8 and hypoperfusion are presented in Fig 5. The correlation be- patients. tween PET and ASL across 20 patients was good (r ϭ 0.587, P Ͻ 1794 Shang Oct 2018 www.ajnr.org FIG 2. FCD type Ib in a 23-year-old patient with a history of seizures, onset at 14 years of age. A, T1-weighted axial image has normal findings. B and C, PET and PET-T1WI fused images (arrow) indicate a well-defined area of focal hypometabolism in the left temporal lobe region. D and E, ASL and ASL-T1WI fused images (arrow) show hypoperfusion in the same brain region. After a left anterior temporal lobectomy, histopathologic findings showed FCD type Ib. After a postoperative follow-up of at least 1 year, the patient was classified as having an Engel class I outcome.

FIG 3. FCD type IIIa–HS in a 29-year-old patient with a history of seizures, onset at 15 years of age. A, T1-weighted axial image has normal findings. B and C, PET and PET-T1WI fused images (arrows) indicate a well-defined area of focal hypometabolism in the left temporal lobe region. However, ASL and ASL-T1WI fused images (D and E, arrows) have normal findings in the same brain region. After a left anterior temporal lobectomy, histopathologic findings of the surgical specimen were consistent with FCD type IIIa–HS. This patient with TLE had an Engel class I outcome after Ͼ1 year of follow-up.

FIG 4. FCD type I in a 26-year-old patient with a history of seizure onset at 14 years of age. The T1-weighted axial image (A), PET image (B), and fused image (C) have normal findings. D and E, ASL and ASL-T1WI fused images (circles) show a well-defined area of focal hypoperfusion in the left temporal lobe region. After a left temporal lobe resection, pathologic findings were consistent with FCD type I. The patient had a follow-up time of Ͼ1 year, showing Engel class I outcome.

.001). Significantly improved logistic regression models were focal area of functional alteration using SPM analysis was de- identified for combining FDG-PET SUVr and ASL perfusion MR tected. Similarly, statistical analysis automatically identified the imaging CBF to predicate localization with a sensitivity of 100% areas of decreased perfusion in all patients compared with the and a specificity of 90.9% (Fig 1). healthy controls. However, no areas with a statistically significant hypoperfusion were detected in the patient group. SPM Analyses of the Patient Group The Z map of the assessment of metabolic and perfusion Figure 6 illustrates the results from the statistical SPM analysis AI between patients and healthy controls in FDG-PET SUV images. abnormalities is shown in Fig 7. The metabolism asymmetry The interictal metabolic activities in patients were compared with mainly located in the middle temporal gyrus was found to be those in healthy controls using voxelwise normalization methods: statistically significant compared with that in the control Hypometabolism of the middle temporal gyrus was mainly de- group, showing agreement with the histopathologic findings tected, which reached perfect agreement with the histopathologic (P Ͻ .001). When we compared the patient group with the findings (P Ͻ .001). As discussed, metabolic abnormalities were control group, a statistically significant asymmetry of perfu- mainly identified in the temporal lobe region by visual inspection, sion was located in the superior temporal gyrus and insula, in which was concordant with the results of SPM analysis. A more line with the histopathologic results (P Ͻ .001). AJNR Am J Neuroradiol 39:1791–98 Oct 2018 www.ajnr.org 1795 firmed by histopathology. Further- more, PET and ASL could provide concordant and complementary infor- mation in localizing the EZ. The value of localization using FDG- PET based on visual inspection varied from 36% to 73% in extratemporal epi- lepsy and reached up to 90% exclusively in temporal lobe cases.26,27 The maxi- mum value for clinical management of FDG-PET can be achieved in patients with MR imaging-negative TLE because it can correctly localize lesions in 80% of the patients.28 For FCD, FDG-PET is more sensitive than MR imaging.29 Our study found that the sensitivity and specificity of FDG-PET were 100% and 81.8%, respectively. There were 3 patients who showed a widespread area of hypometabolism extending be- yond the EZ and 2 patients who showed incorrect lateralization and lo- FIG 5. Regional comparison across the 20 patients for ROIs with hypometabolism in PET and calization, in agreement with the find- hypoperfusion in ASL (r ϭ 0.587, P Ͻ .001). ings of previous studies.10,30 Many previous ASL studies have re- ported interictal hypoperfusion at the site of seizure in patients with TLE and found that interictal ASL might help in localizing the EZ in FCD.31,32 The recommended postlabel delay was used in this study to ensure that the brain parenchyma was sufficiently perfused, which could help to avoid false reduction of relative CBF as estimated with ASL.33 The results of the ASL perfusion MR imaging study showed that the sensitivity and specificity were 83.3% and 54.5%, respectively. The hypoperfusion regions ex- tending beyond the EZ in 3 patients were likely due to confound- ing factors, such as age, leukoaraiosis, and so forth.34 Previous studies reported that areas of hypoperfusion on the ASL MR imaging were well-associated with the hypometa- bolic areas on the PET.35,36 Our study with simultaneous ac- quisition of FDG-PET and ASL MR imaging also showed good overall correlation between them. Combined FDG-PET and ASL MR imaging was studied to find the optimal model for evaluating the value of EZ localization. Receiver operating characteristic analysis and logistic regression models are widely used to evaluate and optimize the performance of clin- ical diagnostic tests, especially for the predictive value of neu- rologic diseases.37,38 A combined PET and ASL model was identified with a higher area under the curve compared with FIG 6. SPM analysis of [18F] FDG-PET images in localizing the EZ be- tween patients and healthy controls. The hypometabolic region is PET or ASL only. Moreover, the specificity of combined FDG- mainly identified in the middle temporal gyrus. The threshold P value PET and ASL MR imaging was improved, which suggested that is set at .001. multiparametric imaging is potentially valuable for EZ local- ization. In terms of individual patients, there were 6 patients DISCUSSION with hypometabolism in PET but normal findings in ASL and 2 This study demonstrated the value of hybrid PET/MR imaging cases showing that ASL can also provide complementary val- in localizing the EZ in patients with MR imaging-negative TLE ues. These findings suggest that the combined use of PET and using the simultaneous acquisition of PET and ASL. The met- ASL can yield excellent performance in lateralization and abolic abnormalities seen in the PET images and abnormal localization. perfusion in ASL images corresponded well with the EZ con- Multiparametric imaging is now widely practiced in presurgi- 1796 Shang Oct 2018 www.ajnr.org aging abnormalities. Eight percent of all specimens were microscopically re- ported nonlesional or negative for ep- ilepsy.43 Because the pathologic find- ings in MR imaging-negative TLE were all FCDs in our study, multicenter studies involving other pathologic types will be needed to make the re- sults more comprehensive and of sta- tistical significance in the future.

CONCLUSIONS Hybrid PET/MR imaging plays an in- creasingly important role in the study of refractory focal epilepsy, especially with MR imaging-negative TLE. Comple- FIG 7. Brain regions of metabolism and perfusion asymmetry of patients compared with controls mentary information with simultaneous by SPM analysis in PET (A) and ASL images (B), respectively. The regions of metabolism and perfusion asymmetry are mainly identified in the middle temporal gyrus and superior temporal acquisition of metabolism and perfu- gyrus, respectively. The threshold P value is set at .001. sion was shown to be useful in localizing the EZ for MR imaging-negative TLE. cal evaluation to identify the EZ, especially for patients with MR 39 imaging-negative epilepsy. However, the nonsynchronized and ACKNOWLEDGMENTS nonsimultaneous acquisition of different functional parameters We thank Judy Buchanan, Division of Nuclear Medicine, Johns might lead to potential biases in the EZ localization. The hybrid Hopkins University, and Abby Barre, Laboratory for the Study of PET/MR imaging can provide simultaneous acquisition of PET the Brain Basis of Individual Differences, Massachusetts General and ASL in the same physiologic or pathophysiologic states. Hospital, for their helpful editing and suggestions. Although visual analysis is widely used in clinical practice, it is subjective and relies on the knowledge and experience of physi- cians. SPM has proved to be an objective and useful method for REFERENCES quantitative analysis in epilepsy with FDG-PET and ASL MR im- 28,40 1. Duncan JS, Winston GP, Koepp MJ, et al. Brain imaging in the as- aging. Metabolic abnormalities were mainly found in the sessment for epilepsy surgery. Lancet Neurol 2016;15:420–33 middle temporal gyrus when comparing patients with healthy CrossRef Medline controls. However, no statistically significant hypoperfusion was 2. Engel J Jr, McDermott MP, Wiebe S, et al; Early Randomized Surgical observed in the qualitative assessment of CBF maps, potentially Epilepsy Trial (ERSET) Study Group. Early surgical therapy for due to the limiting size of the patient cohort. In addition, the AI is drug-resistant temporal lobe epilepsy: a randomized trial. JAMA 2012;307:922–30 CrossRef Medline also suggested for analysis. Previous studies used voxelwise or 3. Capraz IY, Kurt G, Akdemir O¨ ,etal.Surgical outcome in patients region-wise asymmetries to evaluate metabolism and perfusion with MRI-negative, PET-positive temporal lobe epilepsy. Seizure abnormalities between the 2 hemispheres.41,42 Our results show 2015;29:63–68 CrossRef Medline that the asymmetries of metabolism and perfusion were mainly 4. Tan YL, Kim H, Lee S, et al. Quantitative surface analysis of com- located in the temporal lobe. These findings suggest that PET and bined MRI and PET enhances detection of focal cortical dysplasias. Neuroimage 2018;166:10–18 CrossRef Medline ASL had excellent performance. 5. Blu¨mcke I, Thom M, Aronica E, et al. The clinicopathologic spec- There were several limitations to this study. First, the num- trum of focal cortical dysplasias: a consensus classification pro- ber of the patients and healthy controls was relatively small, posed by an ad hoc Task Force of the ILAE Diagnostic Methods and the participants were not age-matched because the partic- Commission. Epilepsia 2011;52:158–74 CrossRef Medline ipants had to be exposed to radioactive material. Future studies 6. Setoain X, Carren˜o M, Pavı´a J, et al. PET and SPECT in epilepsy [in Spanish]. Rev Esp Med Nucl Imagen Mol 2014;33:165–74 CrossRef with larger cohorts are needed for further confirmation of the Medline findings. Second, visual assessment combined with SPM anal- 7. Schwartz TH. Neurovascular coupling and epilepsy: hemodynamic ysis should be used with each patient in localizing the EZ. markers for localizing and predicting seizure onset. Epilepsy Curr Third, the spectrum of histopathologic changes observed in 2007;7:91–94 CrossRef Medline tissue specimens from the patients with MR imaging negative 8. Won HJ, Chang KH, Cheon JE, et al. Comparison of MR imaging with PET and ictal SPECT in 118 patients with intractable epilepsy. for epilepsy could be divided into 4 categories as follows: FCD, AJNR Am J Neuroradiol 1999;20:593–99 Medline HS, abnormalities in white matter, and microscopically non- 9. Chassoux F, Rodrigo S, Semah F, et al. FDG-PET improves surgical lesional (eg, astrogliosis). The most common histopathologic outcome in negative MRI Taylor-type focal cortical dysplasias. abnormality identified is FCD. Increased numbers of hetero- Neurology 2010;75:2168–75 CrossRef Medline topic white matter neurons can be encountered in epileptic 10. Kumar A, Chugani HT. The role of radionuclide imaging in epi- lepsy, Part 1: sporadic temporal and extratemporal lobe epilepsy. brain tissue, which are usually not visible at the MR imaging J Nucl Med Technol 2017;45:14–21 CrossRef Medline level. Five percent to ten percent of patients with histopatho- 11. Uijl SG, Leijten FS, Arends JB, et al. The added value of [18F]-fluoro- logically confirmed TLE with HS do not present with MR im- D-deoxyglucose positron emission tomography in screening for AJNR Am J Neuroradiol 39:1791–98 Oct 2018 www.ajnr.org 1797 temporal lobe epilepsy surgery. Epilepsia 2007;48:2121–29 CrossRef 28. Zhu Y, Feng J, Wu S, et al. Glucose metabolic profile by visual as- Medline sessment combined with statistical parametric mapping analysis in 12. YangPF, Pei JS, Zhang HJ, et al. Long-term epilepsy surgery out- pediatric patients with epilepsy. J Nucl Med 2017;58:1293–99 comes in patients with PET-positive, MRI-negative temporal lobe CrossRef Medline epilepsy. Epilepsy Behav 2014;41:91–97 CrossRef Medline 29. Lee KK, Salamon N. [18F] fluorodeoxyglucose-positron-emission 13. LoPinto-Khoury C, Sperling MR, Skidmore C, et al. Surgical out- tomography and MR imaging coregistration for presurgical evalu- come in PET-positive, MRI-negative patients with temporal lobe ation of medically refractory epilepsy. AJNR Am J Neuroradiol 2009; epilepsy. Epilepsia 2012;53:342–48 CrossRef Medline 30:1811–16 CrossRef Medline 14. Oishi M, Ishida G, Morii K, et al. Ictal focal hyperperfusion demon- 30. Rheims S, Jung J, Ryvlin P. Combination of PET and magnetoen- strated by arterial spin-labeling perfusion MRI in partial epilepsy cephalography in the presurgical assessment of MRI-negative epi- status. Neuroradiology 2012;54:653–56 CrossRef Medline lepsy. Front Neurol 2013;4:188 CrossRef Medline 15. Boscolo Galazzo I, Storti SF, Del Felice A, et al. Patient-specific de- 31. Lim YM, Cho YW, Shamim S, et al. Usefulness of pulsed arterial spin tection of cerebral blood flow alterations as assessed by arterial spin labeling MR imaging in mesial temporal lobe epilepsy. Epilepsy Res labeling in drug-resistant epileptic patients. PLoS One 2015;10: 2008;82:183–89 CrossRef Medline e0123975 CrossRef Medline 32. Blauwblomme T, Boddaert N, Che´maly N, et al. Arterial spin label- 16. Kim BS, Lee ST, Yun TJ, et al. Capability of arterial spin labeling MR ing MRI: a step forward in non-invasive delineation of focal cortical imaging in localizing seizure focus in clinical seizure activity. Eur J Radiol 2016;85:1295–303 CrossRef Medline dysplasia in children. Epilepsy Res 2014;108:1932–39 CrossRef 17. Sierra-Marcos A, Carren˜o M, Setoain X, et al. Accuracy of arterial Medline spin labeling magnetic resonance imaging (MRI) perfusion in de- 33. Haller S, Zaharchuk G, Thomas DL, et al. Arterial spin labeling per- tecting the epileptogenic zone in patients with drug-resistant neo- fusion of the brain: emerging clinical applications. Radiology 2016; cortical epilepsy: comparison with electrophysiological data, struc- 281:337–56 CrossRef Medline tural MRI, SISCOM and FDG-PET. Eur J Neurol 2016;23:160–67 34. Deibler AR, Pollock JM, Kraft RA, et al. Arterial spin-labeling in CrossRef Medline routine clinical practice, Part 2: hypoperfusion patterns. AJNR Am J 18. Garibotto V, Heinzer S, Vulliemoz S, et al. Clinical applications of Neuroradiol 2008;29:1235–41 CrossRef Medline hybrid PET/MRI in neuroimaging. Clin Nucl Med 2013;38:e13–18 35. Pendse N, Wissmeyer M, Altrichter S, et al. Interictal arterial spin- CrossRef Medline labeling MRI perfusion in intractable epilepsy. J Neuroradiol 2010; 19. Schwenzer NF, Stegger L, Bisdas S, et al. Simultaneous PET/MR im- 37:60–63 CrossRef Medline aging in a human brain PET/MR system in 50 patients: current state 36. Storti SF, Boscolo Galazzo I, Del Felice A, et al. Combining ESI, ASL of image quality. Eur J Radiol 2012;81:3472–78 CrossRef Medline and PET for quantitative assessment of drug-resistant focal epi- 20. Ding YS, Chen BB, Glielmi C, et al. A pilot study in epilepsy patients lepsy. Neuroimage 2014;102(Pt 1):49–59 CrossRef Medline using simultaneous PET/MR. Am J Nucl Med Mol Imaging 2014;4: 37. Lascano AM, Perneger T, Vulliemoz S, et al. Yield of MRI, high- 459–70 Medline density electric source imaging (HD-ESI), SPECT and PET in epi- 21. Shin HW, Jewells V, Sheikh A, et al. Initial experience in hybrid lepsy surgery candidates. Clin Neurophysiol 2016;127:150–55 PET-MRI for evaluation of refractory focal onset epilepsy. Seizure CrossRef Medline 2015;31:1–4 CrossRef Medline 38. Shaffer JL, Petrella JR, Sheldon FC, et al; Alzheimer’s Disease Neuro- 22. Boscolo Galazzo I, Mattoli MV, Pizzini FB, et al. Cerebral metabo- imaging Initiative. Predicting cognitive decline in subjects at risk for lism and perfusion in MR-negative individuals with refractory fo- Alzheimer disease by using combined cerebrospinal fluid, MR im- cal epilepsy assessed by simultaneous acquisition of (18)F-FDG aging, and PET biomarkers. Radiology 2013;266:583–91 CrossRef PET and arterial spin labeling. Neuroimage Clin 2016;11:648–57 Medline CrossRef Medline 39. Ferna´ndez S, Donaire A, Sere`s E, et al. PET/MRI and PET/MRI/ 23. Noe K, Sulc V, Wong-Kisiel L, et al. Long-term outcomes after non- SISCOM coregistration in the presurgical evaluation of refractory lesional extratemporal lobe epilepsy surgery. JAMA Neurol 2013;70: focal epilepsy. Epilepsy Res 2015;111:1–9 CrossRef Medline 1003–08 CrossRef Medline 40. Chen G, Lei D, Ren J, et al. Patterns of postictal cerebral perfusion in 24. Wang DD, Piao YS, Blumcke I, et al. A distinct clinicopathological variant of focal cortical dysplasia IIId characterized by loss of layer idiopathic generalized epilepsy: a multi-delay multi-parametric ar- 4 in the occipital lobe in 12 children with remote hypoxic-ischemic terial spin labelling perfusion MRI study. Sci Rep 2016;6:28867 injury. Epilepsia 2017;58:1697–1705 CrossRef Medline CrossRef Medline 25. Gaillard WD, Berl MM, Duke ES, et al. fMRI language dominance and 41. Van Bogaert P, Massager N, Tugendhaft P, et al. Statistical paramet- FDG-PET hypometabolism. Neurology 2011;76:1322–29 CrossRef ric mapping of regional glucose metabolism in mesial temporal Medline lobe epilepsy. Neuroimage 2000;12:129–38 CrossRef Medline 26. Lee JJ, Lee SK, Lee SY, et al. Frontal lobe epilepsy: clinical character- 42. Pizzini FB, Farace P, Manganotti P, et al. Cerebral perfusion altera- istics, surgical outcomes and diagnostic modalities. Seizure 2008;17: tions in epileptic patients during peri-ictal and post-ictalphase: 514–23 CrossRef Medline PASL vs DSC-MRI. Magn Reson Imaging 2013;31:1001–05 CrossRef 27. Archambaud F, Bouilleret V, Hertz-Pannier L, et al. Optimizing sta- Medline tistical parametric mapping analysis of 18F-FDG PET in children. 43. So EL, Ryvlin P. MRI-Negative Epilepsy: Evaluation and Surgical Man- EJNMMI Res 2013;3:2 CrossRef Medline agement. Cambridge: Cambridge University Press, 2015

1798 Shang Oct 2018 www.ajnr.org ORIGINAL RESEARCH ADULT BRAIN

Longitudinal Persistence of Meningeal Enhancement on Postcontrast 7T 3D-FLAIR MRI in Multiple Sclerosis

X S.N. Jonas, X I. Izbudak, X A.A. Frazier, and X D.M. Harrison

ABSTRACT BACKGROUND AND PURPOSE: Preliminary research has demonstrated that postgadolinium 3D-FLAIR MR imaging at 7T may be a valuable tool for detecting abnormal meningeal enhancement and inflammation in MS; however, researchers have not systematically investigated its longitudinal persistence. We hypothesized that persistence of meningeal enhancement in MS varies on the basis of pattern of enhancement as well as demographic and clinical factors such as treatment status, disease phenotype, and disability score.

MATERIALS AND METHODS: Thirty-one subjects with MS were prospectively scanned before and after intravenous contrast adminis- tration at 2 time points, approximately 1 year apart. Fifteen subjects in the cohort were scanned at another time approximately 1 year later. Foci of enhancement were categorized into 4 subtypes: subarachnoid spread/fill, subarachnoid nodular, vessel wall, and dural foci. We reviewed follow-up scans to determine whether foci changed between time points and then compared persistence with demographic and clinical variables.

RESULTS: Persistence ranged from 71% to 100% at 1 year and 73% to 100% at 2 years, depending on the enhancement pattern. Subarachnoid spread/fill and subarachnoid nodular subtypes persisted less often than vessel wall and dural foci. Persistence was not significantly different between those on/off treatment and those with progressive/nonprogressive disease phenotypes. The number of persisting foci was significantly different in subjects with/without increasing Expanded Disability Status Scale scores (median, 12 versus 7.5, P ϭ .04).

CONCLUSIONS: Longitudinal persistence of meningeal enhancement on 3D-FLAIR at 7T in MS varies by pattern of enhancement and correlates with worsening disability; however, it is not significantly different in those on/off treatment or in those with progressive/ nonprogressive disease phenotypes.

ABBREVIATIONS: EDSS ϭ Expanded Disability Status Scale; MP2RAGE ϭ magnetization-prepared rapid acquistion of 2 gradient echoes

S is a chronic inflammatory demyelinating disorder classi- nation, cortical neuroaxonal loss, microglial activation, and oligo- Mcally affecting white matter within the brain and spinal dendrocyte dysfunction.1-6 Visualization of meningeal inflamma- cord. In the past few decades, an additional pathophysiologic tion on MR imaging has become an active, and somewhat mechanism—meningeal inflammation—has been elucidated in controversial, area of recent investigation.7 Landmark studies have MS, which is now believed to directly contribute to cortical demyeli- demonstrated that gadolinium-enhanced 3D-FLAIR sequences, which have long been useful for identifying meningeal infection and carcinomatosis, can also be used to image meningeal disease in the Received March 20, 2018; accepted after revision July 20. MS population. At a magnetic field strength of 3T, Absinta et al7 From the Department of Radiology (S.N.J., A.A.F.), University of Maryland Medical found that approximately 25% of patients with MS demonstrated Center, Baltimore, Maryland; Department of Radiology/Neuroradiology (I.I), Johns Hopkins University, Baltimore, Maryland; and Department of Neurology, University leptomeningeal enhancement on gadolinium-enhanced FLAIR. of Maryland School of Medicine (D.M.H.), Baltimore, Maryland. Protocols for imaging meningeal enhancement were im- This study was funded, in part, by grants from EMD Serono and the National Insti- proved by Zivadinov et al,8 who showed the benefit of both pre- tutes of Health (National Institute of Neurological Disorders and Stroke 1K23NS072366-01A1). and postcontrast acquisitions and generating subtraction images Please address correspondence to Samuel N. Jonas, MD, Department of Radiology, when assessing meningeal enhancement, because these tech- University of Maryland Medical Center, 22 South Greene St, Baltimore, MD 21201; e-mail: [email protected] niques decrease false-positives and reduce interpretation time. Indicates open access to non-subscribers at www.ajnr.org Recent preliminary research has also suggested that 7T MR imag- Indicates article with supplemental on-line tables. ing may be more sensitive than 3T for detecting meningeal en- http://dx.doi.org/10.3174/ajnr.A5796 hancement. Although no direct 3T-versus-7T comparisons have

AJNR Am J Neuroradiol 39:1799–1805 Oct 2018 www.ajnr.org 1799 Table 1: MRI sequence parameters Sequence Resolution (mm) TR TI TE Parallel Imaging Flip Angle Time (min:sec) ϫ ϫ ϭ ϭ ϭ ϫ ϭ MP2RAGE 0.7 0.7 0.7 TRvolume 8.25 s TI1 1s 1.97 ms SENSE 2 2FA1 7° 9:46 ϭ ϭ ϭ TRTFE 6.9 ms TI2 3.3 s FA2 5° MPFLAIR 0.7 ϫ 0.7 ϫ 0.7 8000 ms 2077 ms 400 ms SENSE ϭ 2 ϫ 3 90° 10:48 Note:—SENSE indicates sensitivity encoding; MPFLAIR, magnetization-prepared FLAIR; FA, flip angle; TFE, turbo field echo. been made in the same study population, up to 90% of patients were acquired before the administration of contrast and again after with MS undergoing contrast-enhanced brain MR imaging at 7T the intravenous administration of 0.1 mmol/kg of gadoteridol (Pro- demonstrated at least 1 enhancing focus.9 This result closely ap- Hance; Bracco Diagnostics, Princeton, New Jersey). Magnetization- proximates the 89% of patients with MS reported to show some prepared rapid acquistion of 2 gradient echoes (MP2RAGE) images element of leptomeningeal inflammation at postmortem examina- were initiated approximately 3 minutes after contrast administra- tion.10,11 Given this radiologic-pathologic concordance, it is conceiv- tion, and magnetization-prepared FLAIR images were initiated ap- able that 7T 3D-FLAIR may soon provide a noninvasive in vivo proximately 20 minutes after contrast administration. method of detecting and accurately quantifying the extent of menin- geal inflammation in patients with MS. Image Processing and Analysis Meningeal enhancement was noted to be a persistent phe- MP2RAGE images were processed to create a T1-weighted image nomenon in prior 3T studies7; however, at 7T, where sensitivity and a T1 map.14 Images were then manipulated using Medical for meningeal enhancement in MS appears to be significantly Image Processing, Analysis, and Visualization (Version 7.2; http:// higher, it remains unknown whether smaller, more subtle foci of mipav.cit.nih.gov). Magnetization-prepared FLAIR images under- enhancement wax and wane in a predictable pattern across time went N4 inhomogeneity correction before analysis.15 Pre- and or whether they remain longitudinally stable. Also unknown is the postcontrast magnetization-prepared FLAIR images were regis- degree to which enhancement persistence with time is associated tered to the precontrast T1 map. A magnetization-prepared with previously described enhancement shape and morphology, FLAIR subtraction image was created by direct subtraction of the including subarachnoid spread/fill and subarachnoid nodular registered pre- and postcontrast images. 7,9 patterns. Because prior studies have shown that the prevalence The magnetization-prepared FLAIR subtraction image, of meningeal enhancement varies with enhancement morphol- alongside the pre- and postcontrast magnetization-prepared 7,9 ogy, in this study, we hypothesized that persistence of menin- FLAIR images, was reviewed by 2 independent raters (a postgrad- geal enhancement in MS would vary on the basis of the morphol- uate year 4 radiology resident and an academic MS neurologist) ogy of enhancement as well as demographic and clinical factors who were blinded to subject identity, disease state, and treatment such as treatment status, disease phenotype, and disability scores. regimen. Hyperintensities noted on the subtraction image were Greater understanding of the imaging and clinical characteristics located on anatomic images and demarcated, if present, in the of meningeal enhancement is necessary if these features are to aid meningeal space on postcontrast images only. All foci were local- in the diagnosis of and prognosis for patients with MS. ized in 3 orthogonal planes before notation. When needed, coreg- istered MP2RAGE T1-weighted images were used for confirma- MATERIALS AND METHODS tion of anatomic locations. The pattern of enhancement was Standard Protocol Approval and Informed Consent categorized on the basis of location and morphology and stratified The institutional review boards at the authors’ institutions ap- into 1 of 4 subtypes. Subarachnoid spread/fill foci were character- proved this Health Insurance Portability and Accountability Act– ized by the presence of contrast in the subarachnoid space distrib- compliant, prospective study. Written, informed consent was ob- uted in an amorphous manner (Fig 1A). Subarachnoid nodular tained from all participants. foci were characterized by small, round areas of contrast, usually 1–2 voxels (0.7–1.4 mm) and were adherent to the pial surface Participants Thirty-one volunteers, 26–61 years of age, with diagnoses of re- (Fig 1B). Vessel wall enhancement was characterized by contrast lapsing-remitting MS, secondary-progressive MS, or primary- outlining the outer margin of a large meningeal vessel with a sig- progressive MS according to the 2010 revised McDonald Criteria nal void in the lumen of the vessel, often resulting in a character- were recruited.12 Exclusion criteria included contraindications to istic tram-track appearance (Fig 1C). Dural foci were character- contrast-enhanced MR imaging. ized by discrete regions of enhancement clearly situated along the dural surface without extension into the subarachnoid space (Fig MR Imaging Protocol 1D). Following both independent reviews, a consensus review was Study participants were prospectively scanned at 2 time points ap- performed under the supervision of an expert third rater (an ac- proximately 1 year apart on a 7T Achieva scanner (Philips Health- ademic neuroradiologist with 12 years of experience). After con- care, Best, the Netherlands) with a volume-transmit/32-channel sensus review, follow-up images underwent linear registration head coil (Nova Medical, Wilmington, Massachusetts). Fifteen pa- (with 9 df) to baseline images. Consensus regions of contrast en- tients in the cohort were scanned at an additional third visit approx- hancement on baseline images were reviewed for their presence or imately 1 year later. Scans were obtained between September 9, 2014, absence on follow-up scans. The total number of foci per sub- and August 21, 2017. Dielectric padding was used for improved im- ject that persisted between scans was compared among differ- age homogeneity.13 Scanning parameters are listed in Table 1. Images ent morphologies of meningeal enhancement and correlated 1800 Jonas Oct 2018 www.ajnr.org FIG 1. Original illustration depicting the 4 morphologies of meningeal enhancement seen in this analysis. Subarachnoid spread/fill pattern (represented by green in A) is an amorphous and ill-defined collection of contrast pooling within the cerebral sulci. The subarachnoid nodular pattern (B) is defined as a punctate, discrete site of meningeal enhancement located within the cerebral sulci abutting the pial surface. The vessel wall pattern (C) is characterized by extension of contrast along the outer margin of large meningeal vessels with a preserved internal flow void creating a characteristic tram-track appearance. The dural pattern (D) is a circumscribed, rounded focus of contrast situated along the dural margin without extension into the subarachnoid space. The perivascular, tubular white structures (seen in schematics A, B, and D) represent the recently discovered meningeal lymphatic system. Reaccumulation of leaked contrast from the CSF into these meningeal lymph channels is a potential mechanism for the venous rim pattern (C). with demographic and clinical data. Additionally, the propor- demographic and clinical variables using the Wilcoxon rank sum tion of baseline foci per subject that persisted to follow-up statistic. We computed the Spearman rank correlation for correla- scans was also compared with morphologic, demographic, and tion testing. All statistical tests were performed with a significance clinical factors. threshold of P Ͻ .05. Due to the small sample size and exploratory nature of this study, adjustment for multiple comparisons was not Disability Measures performed. The Kurtzke Expanded Disability Status Scale (EDSS) was used to characterize disability.16 EDSS progression was defined as an in- crease of the EDSS score at follow-up of Ն1.0 if the baseline EDSS RESULTS score was Ͻ5.0 or an increase of Ն0.5 if the baseline EDSS score We recruited 31 patients with MS; most had the relapsing-remit- was Ն5.0. The Modified Fatigue Impact Scale was used to assess ting MS phenotype (n ϭ 21, 68%), though 7 subjects had second- MS-related fatigue.17,18 The Symbol Digit Modalities Test was ary-progressive MS (23%) and 3 subjects had primary-progres- used to assess cognitive functioning.19 These tests were adminis- sive MS (10%) (Table 2). No subject had a comorbid tered at each study visit. neuroinflammatory disorder. Most subjects were on disease- modifying therapy (n ϭ 25, 81%). This was a relatively stable and Statistical Analysis moderately disabled patient population with a median of 0 Statistical analysis was performed in Stata 10.1 IC (StataCorp, Col- (range, 0–3) relapses in the year before enrollment and a median lege Station, Texas). Nonparametric testing was used due to the non- Expanded Disability Status Scale score of 3 (range, 1–6.5). En- normal distribution of data. We performed group comparisons for hancing white matter lesions were seen in 3 subjects on review of AJNR Am J Neuroradiol 39:1799–1805 Oct 2018 www.ajnr.org 1801 Table 2: Cohort baseline characteristicsa association (␳ ϭϪ0.31, P ϭ .09) be- Characteristics tween the proportion of persisting foci Age at enrollment (yr) 49 (26–61) overall and the interval change in Sym- Sex 11/31 Men (35%), 20/31 women (65%) bol Digit Modalities Test scores at 1 year Disease subtype at enrollment 21/31 (68%) RR, 7/31 (23%) SP, (On-line Table 3). Surprisingly and 3/31 (10%) PP Disease duration at enrollment (mo) 109 (8–461) counterintuitively, we observed a posi- Patients with new relapses in past 30 days 1/31 (3%) tive correlation (␳ ϭ 0.45, P ϭ .01) be- No. of relapses in past year per subject 0 (0–3) tween the proportion of enhancing Modified Fatigue Impact Scale score at enrollment 43 (0–78) meningeal foci that persisted at 1 year Symbol Digit Modality Test at enrollment 50 (35–81) and baseline Symbol Digit Modalities Expanded Disability Status Scale score at enrollment 3 (1–6.5) Immunomodulatory treatment status at baseline Test scores (On-line Table 3). This asso- On treatment 25/31 (81%) ciation was driven by the correlation Not on treatment 6/31 (19%) (␳ ϭ 0.48, P ϭ .01) between the propor- Treatment type at baseline tion of subarachnoid spread/fill subtype Interferon 3/25 (12%) that persisted at 1 year and baseline Glatiramer 6/25 (24%) Natalizumab 2/25 (8%) Symbol Digit Modalities Test scores. We Teriflunomide 1/25 (4%) also observed 15 foci of meningeal en- Fingolimod 4/25 (16%) hancement that developed in the inter- Dimethyl fumarate 9/25 (36%) val between baseline and follow-up No. of subjects who switched between 10/31 scans. The morphologies of these 15 foci disease-modifying therapies from baseline to follow-up scans were as follows: 6 subarachnoid spread/ fill, 4 subarachnoid nodular, 2 vessel Note:—RR indicates relapsing-remitting MS; SP, secondary-progressive MS; PP, primary-progressive MS. a Median values are shown with the range of observed values in parentheses. wall, and 3 dural foci.

Table 3: Anatomic distribution within the brain of enhancing DISCUSSION meningeal foci at baseline In this study, we catalogued 2 enhancement patterns described in No. of Foci Percentage of Brain Region at Baseline Foci at Baseline prior analyses (subarachnoid spread/fill and subarachnoid nodu- 7,9 Right frontal 60 21.1 lar) in addition to describing 2 new patterns of meningeal en- Left frontal 64 22.5 hancement for the first time: vessel wall enhancement and dural Right parietal 44 15.5 foci. Previous studies without precontrast comparison sequences Left parietal 44 15.5 excluded from consideration regions of postcontrast hyperinten- Right occipital 20 7.0 sity in/near the dural sinuses, large subarachnoid veins, and the Left occipital 24 8.4 Right temporal 16 5.6 basal meninges to reduce false-positives because these structures Left temporal 8 2.8 often manifest precontrast T1 or FLAIR hyperintensity.7,9 Using Right cerebellum 2 0.7 techniques similar to those in the recent investigation by Zivadi- Left cerebellum 2 0.7 nov et al,8 we coregistered and subtracted pre- and postcontrast magnetization-prepared FLAIR sequences in all cases. Given this T1-weighted images, with 2 subjects having 1 enhancing lesion protocol, we did not have to exclude any structures a priori, and and 1 subject having 2 enhancing lesions. we were confident in our ability to differentiate true vessel wall At baseline, a total of 284 enhancing foci were identified across and dural foci enhancement from intrinsically increased signal. all 31 subjects. Table 3 lists the anatomic distribution of these foci Which anatomic/pathologic substrates are represented by vessel within the brain. Most (Ͼ98%) of the observed foci were located wall and dural foci is unknown, but most interesting, both closely supratentorially. Figure 2 shows the percentage of enhancing match what was recently described for visualization of meningeal 20-22 meningeal foci identified at baseline that persisted at later time lymphatics by FLAIR MR imaging. Thus, these findings may points. Figures 3 and 4 provide examples of persisting and resolv- represent gadolinium absorption by lymphatic structures after ing enhancing meningeal foci from each group. Table 4 and On- leakage into the CSF. The accumulation of gadolinium signal line Tables 1–3 compare the persistence of meningeal enhance- alongside the outer wall of vessels in the vessel wall pattern is also ment with demographic and clinical variables. We found no very reminiscent of the expected location and direction of drain- significant difference in the total number or proportion of longi- age of solutes from brain parenchyma along the recently de- tudinally persistent enhancing meningeal foci between those on scribed glymphatic system.23 Alternatively, it is also possible that or off treatment or between those with progressive phenotypes vessel wall and dural foci could represent the reaccumulation, (primary-progressive MS and secondary-progressive MS) versus under hydrostatic pressure, of CSF-leaked gadolinium back into a relapsing phenotype (relapsing-remitting MS). However, we did the venous system. They could also feasibly represent the actual find significantly more (P ϭ .04) persistent foci in Expanded Dis- sites of blood-brain and blood-CSF barrier disturbance secondary ability Status Scale progressors (median, 12; range, 1–15) com- to ongoing inflammation.24 Because age- and sex-matched pared with those who were not progressors (median, 7.5; range, healthy controls were not used in this study, the specificity of 1–24). We also observed a nonsignificant trend toward a negative dural and vessel wall enhancements to MS is unknown. Future 1802 Jonas Oct 2018 www.ajnr.org work is needed to determine whether such findings are specific to However, the lack of difference is not surprising because prior MS, neuroinflammatory disease in general, or are seen in all studies have also failed to show differences in meningeal enhance- patients. ment between those who are and are not taking disease-modifying We found no significant difference in the total number or medications.7 Our data reinforce the notion that current immu- proportion of persisting meningeal enhancement per subject be- nomodulatory medications may not adequately control menin- tween those on treatment and those off treatment. Lack of a sig- geal inflammation. Of note, none of our subjects received a course nificant difference between subjects on/off treatment may, in of corticosteroids during the study. However, 10 of 31 subjects in part, be explained by the low statistical power of our study because this cohort switched between disease-modifying therapies from a relatively small number of untreated subjects were included. baseline to follow-up scans, including 2 subjects who changed to rituximab and 1 subject who switched to alemtuzumab—both monoclonal anti- bodies that impact B-cell function. De- spite such changes, most foci remained stable. Given the sample size of this report, we would not want to comment on the persistence (or lack thereof) of foci with individual therapy changes because con- clusions from 1 or 2 examples would not be generalizable. Future comparative studies are needed to determine whether changes in any specific disease-modifying therapies or monoclonal regimens alter the longitudinal persistence of meningeal enhancement. Surprisingly, we did not detect a sig- nificant difference in the persistence of FIG 2. Graph displaying the percentages of baseline enhancing meningeal foci that persist 1 year later (gray bars) and 2 years later (black bars). All 31 participants were scanned at baseline and at meningeal enhancement between MS 1 year, but 2-year data are limited to 15 participants. At 1 year, persistence was noted in 253/284 subjects with progressive phenotypes (89%) overall foci, 91/114 (80%) subarachnoid spread/fill foci, 10/14 (71%) subarachnoid nodular (primary-progressive MS and second- foci, 104/107 (97%) vessel wall foci, and 46/46 (100%) dural nodular foci. At 2 years, persistence was noted in 132/161 (82%) overall foci, 45/62 (73%) subarachnoid spread/fill foci, 6/7 (86%) subarach- ary-progressive MS) and those with re- noid nodular foci, 55/55 (100%) vessel wall foci, and 34/34 (100%) dural foci.

FIG 3. Examples of persisting foci of meningeal enhancement on delayed postcontrast FLAIR at 7T. Sagittal reformatted images show subarach- noid spread/fill enhancement that persists from December 15, 2015 (A), to March 3, 2017 (B), in a 58-year-old woman with relapsing-remitting MS. Axial images show subarachnoid nodular enhancement that persists from October 8, 2014 (C), to March 3, 2017 (D), in a 49-year-old man with relapsing-remitting MS. Axial images show vessel wall enhancement that persists from March 14, 2016 (E), to April 4, 2017 (F), in a 57-year-old man with primary-progressive MS. Axial images show dural enhancement that persists from May 9, 2016 (G), to May 31, 2017 (H), in a 44-year-old woman with secondary-progressive MS. Note that no intrinsic signal was observed in these locations on precontrast acquisitions (not shown). AJNR Am J Neuroradiol 39:1799–1805 Oct 2018 www.ajnr.org 1803 lapsing-remitting MS. This finding runs counter to the previously foci) at 1 year with disability progression (by the Expanded Dis- proposed theory that meningeal enhancement may be a substrate ability Status Scale) during the same time period. If postcontrast specific to progressive MS, with the associated cortical demy- meningeal enhancement on magnetization-prepared FLAIR is in- elination and volume loss representing a distinctly late marker deed representative of meningeal inflammation, this finding may of disease.1,7 Indeed, previous studies have shown that the indicate that persistent rather than transient meningeal inflam- presence of leptomeningeal enhancement at 3T was 1.7-fold mation is required to affect prognosis in patients with MS. This higher in progressive MS compared with relapsing-remitting notion is supported by postmortem data showing that the devel- MS, and postmortem findings of meningeal inflammation opment of structures that support ongoing inflammation, such as were more profound in those with secondary-progressive ectopic lymphoid follicular tissue, in the meninges of patients MS.1,7 However, our 7T data showed no difference in the fre- with MS is associated with earlier onset of disease, shorter diag- quency of longitudinal persistence of meningeal enhancement nosis-death interval, and more severe cortical pathology.1,25 If between patients with MS with progressive and relapsing this relationship can be confirmed in larger studies, perhaps the phenotypes. elimination of persistently enhancing meningeal foci can become Although the persistence of enhancing foci was not related to a target outcome for patients with MS. clinical phenotype, we did detect a significant relationship be- While we found a significant relationship between changes in tween the persistence of foci (especially subarachnoid spread/fill Expanded Disability Status Scale scores and the persistence of enhancement, we were unable to detect any similar association between imaging findings and MS-related fatigue or cog- nitive deficits (On-line Tables). This dif- ference may be due to the small sample size of our study or the short follow-up duration. Still to be investigated is the rate at which foci of meningeal enhance- ment develop with time, whether the rate of development is associated with enhancement morphology, and whether the rate of development is associated with demographic and clinical parame- ters. This study focused on subjects with MS only; future work is needed to eluci- date the rate of longitudinal persistence of meningeal enhancement in other neuroinflammatory diseases.26 Finally, our study is limited by possible false dis- FIG 4. Examples of resolving foci of meningeal enhancement on delayed postcontrast FLAIR at covery because we did not perform mul- 7T. Coronal reformatted images show subarachnoid spread/fill enhancement that resolves be- tiple-comparison correction, given the tween October 23, 2014 (A), and February 26, 2016 (B), in a 49-year-old woman with relapsing- exploratory nature of this investigation. remitting MS. Coronal reformatted images show subarachnoid nodular enhancement within the cerebellar folia that resolves between October 8, 2014 (C), and February 19, 2016 (D), in a 49-year- Therefore, our results will require repli- old man with relapsing-remitting MS. Sagittal formatted images show vessel wall enhancement cation before widespread acceptance of that resolves from May 9, 2016 (E), to May 31, 2017 (F), in a 44-year-old woman with secondary- these conclusions. Despite these limita- progressive MS. Note that no foci of meningeal enhancement classified as a dural subtype resolved in this study. tions, these preliminary results provide

Table 4: Wilcoxon rank sum test for longitudinal persistence of meningeal enhancement versus demographic and clinical factorsa On Not on EDSS EDSS Treatment Treatment Progressive Nonprogressive Progressor at Nonprogressor (n = 25) (n =6) MS (n = 10) MS (n = 21) 1yrb (n =7) at1yrb (n = 24) Total No. of overall foci persisting at 1 yr 9 (1–24) 9 (1–15) 8 (1–24) 9 (1–15) 12 (1–15) 7.5 (1–24)c per subject Total No. of subarachnoid spread/fill foci 2 (0–9) 2 (0–6) 2 (0–9) 2 (0–9) 5 (1–9) 2 (0–9) persisting at 1 yr per subject Total No. of subarachnoid nodular foci 0 (0–2) 0 (0–1) 0 (0–0) 0 (0–2)c 0 (0–0) 0 (0–2) persisting at 1 yr per subject Total No. of vessel wall foci persisting at 3 (0–11) 3.5 (1–6) 2 (0–6) 3 (0–11) 3 (0–11) 3 (0–7) 1 yr per subject Total No. of dural foci persisting at 1 yr 1 (0–9) 1 (0–6) 1 (0–9) 1 (0–6) 1 (0–6) 1 (0–9) per subject a Median values are listed with the range of observed values in parentheses. b Criteria for EDSS progressor status is listed in the “Materials and Methods” section. c P Ͻ .05.

1804 Jonas Oct 2018 www.ajnr.org important new insight into the longitudinal activity of meningeal nant for B cell recruitment to the CSF during neuroinflammation. enhancement in MS. J Neuroinflammation 2012;9:93 CrossRef Medline 6. Nielsen AS, Kinkel RP, Tinelli E, et al. Focal cortical lesion detection in multiple sclerosis: 3 Tesla DIR versus 7 Tesla FLASH-T2. J Magn CONCLUSIONS Reson Imaging 2012;35:537–42 CrossRef Medline Here we describe the results of a prospective, systematic investi- 7. Absinta M, Vuolo L, Rao A, et al. Gadolinium-based MRI character- gation into the longitudinal persistence of meningeal enhance- ization of leptomeningeal inflammation in multiple sclerosis. Neu- ment in MS using 7T 3D-FLAIR. Given our pre- and postcontrast rology 2015;85:18–28 CrossRef Medline 8. Zivadinov R, Ramasamy DP, Hagemeier J, et al. Evaluation of lepto- techniques, we are able to include, for the first time, vessel wall meningeal contrast enhancement using pre-and postcontrast sub- and dural foci subtypes, which persist most frequently; their ap- traction 3D-FLAIR imaging in multiple sclerosis. AJNR Am J Neu- pearance very closely matches recent descriptions of meningeal roradiol 2018;39:642–47 CrossRef Medline lymphatics or the glial lymphatics system.19-22 Longitudinal per- 9. Harrison DM, Wang KY, Fiol J, et al. Leptomeningeal enhancement at sistence of meningeal enhancement is not significantly different 7T in multiple sclerosis: frequency, morphology, and relationship to cortical volume. J Neuroimaging 2017;27:461–68 CrossRef Medline between those on or off immunomodulatory treatment, and there 10. Howell OW, Reeves CA, Nicholas R, et al. Meningeal inflammation is not a significant difference in the rates of longitudinal persis- is widespread and linked to cortical pathology in multiple sclerosis. tence between those with progressive clinical phenotypes (prima- Brain J Neurol 2011;134(Pt 9):2755–71 CrossRef Medline ry-progressive MS and secondary-progressive MS) and those 11. Howell OW, Schulz-Trieglaff EK, Carassiti D, et al. Extensive grey without a progressive clinical phenotype (relapsing-remitting matter pathology in the cerebellum in multiple sclerosis is linked to inflammation in the subarachnoid space. Neuropathol Appl Neuro- MS). However, there is a significantly increased number of per- biol 2015;41:798–813 CrossRef Medline sistent foci in subjects who have worsening Expanded Disability 12. Polman CH, Reingold SC, Banwell B, et al. Diagnostic criteria for Status Scale scores at 1 year compared with those who do not, multiple sclerosis: 2010 revisions to the McDonald criteria. Ann suggesting that persistently enhancing meningeal foci may be an Neurol 2011;69:292–302 CrossRef Medline in vivo imaging marker for ongoing meningeal inflammation 13. Haines K, Smith NB, Webb AG. New high dielectric constant mate- rials for tailoring the B1؉ distribution at high magnetic fields. J causing clinical progression. Magn Reson 2010;203:323–27 CrossRef Medline 14. Marques JP, Kober T, Krueger G, et al. MP2RAGE, a self bias-field Disclosures: Samuel N. Jonas—UNRELATED: Support for Travel to Meetings for the corrected sequence for improved segmentation and T1-mapping at Study or Other Purposes: 2018 Radiological Society of North America Resident high field. Neuroimage 2010;49:1271–81 CrossRef Medline Travel Grant, Comments: $500; Provision of Writing Assistance, Medicines, Equip- 15. Tustison NJ, Avants BB, Cook PA, et al. N4ITK: improved N3 bias cor- ment, or Administrative Support: Department of Radiology Pagination Editor. Izlem rection. IEEE Trans Med Imaging 2010;29:1310–20 CrossRef Medline Izbudak—UNRELATED: Consultancy: Alexion, Comments: Neuromyelitis Optica Re- lapse Adjudication Committee MRI reads; Grants/Grants Pending: Siemens, Biogen, 16. Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an Comments: Siemens DTI of the spinal cord prospective research study, Biogen MS expanded disability status scale (EDSS). Neurology 1983;33:1444–52 PATHS MRI scan-rescan study.* Aletta A. Frazier—UNRELATED: Consultancy: Amer- CrossRef Medline ican Institute for Radiologic Pathology, Comments: Section Chief, Cardiovascular 17. Multiple Sclerosis Council for Clinical Practice Guidelines. Fa- Radiology; Associate Chief, Chest Radiology, American Institute for Radiologic Pa- tigue and multiple sclerosis: evidence-based management strat- thology, a program of the American College of Radiology, Silver Spring, Maryland; egies for fatigue in multiple sclerosis. Paralyzed Veterans of Amer- Employment: Clinical Professor of Radiology, University of Maryland School of Med- ica 1998:1–33 icine, Comments: Clinical Professor of Diagnostic Radiology and Nuclear Medicine, 18. Flachenecker P, Kumpfel T, Kallmann B, et al. Fatigue in multiple University of Maryland School of Medicine, Baltimore, Maryland. Daniel M. sclerosis: a comparison of different rating scales and correlation to Harrison—RELATED: Grant: EMD-Serono, Comments: research grant*; UNRELATED: Consultancy: EMD-Serono, Genentech, Genzyme, Biogen, Comments: honoraria for clinical parameters. Mult Scler 2002;8:523–26 CrossRef Medline 1-time advisory board meetings; Payment for Lectures Including Service on Speakers 19. Smith A. Symbol Digit Modalities Test: Manual. Los Angeles: Western Bureaus: Houston Methodist Hospital; Other: American College of Physicians, Com- Psychological Services; 1973 ments: textbook chapter writing and multiple-choice question development for 20. Aspelund A, Antila S, Proulx ST, et al. A dural lymphatic vascular Medical Knowledge Self Assessment Program (MKSAP). *Money paid to the system that drains brain interstitial fluid and macromolecules. J institution. Exp Med 2015;212:991–99 CrossRef Medline 21. Raper D, Louveau A, Kipnis J. How do meningeal lymphatic vessels drain the CNS? Trends Neurosci 2016;39:581–86 CrossRef Medline REFERENCES 22. Absinta M, Ha SK, Nair G, et al. Human and nonhuman primate 1. Magliozzi R, Howell O, Vora A, et al. Meningeal B-cell follicles in meninges harbor lymphatic vessels that can be visualized noninva- secondary progressive multiple sclerosis associate with early onset sively by MRI. Elif 2017;6 CrossRef Medline of disease and severe cortical pathology. Brain J Neurol 2007;130 23. Plog BA, Nedergaard M. The glymphatic system in central nervous (Pt 4):1089–104 Medline system health and disease: past, present, and future. Annu Rev 2. Lucchinetti CF, Popescu BF, Bunyan RF, et al. Inflammatory cortical Pathol 2018;13:379–94 CrossRef Medline demyelination in early multiple sclerosis. N Engl J Med 2011;365: 24. Ortiz GG, Pacheco-Moise´s FP, Macı´as-Islas MA, et al. Role of the 2188–97 CrossRef Medline blood-brain barrier in multiple sclerosis. Arch Med Res 2014;45: 3. Choi SR, Howell OW, Carassiti D, et al. Meningeal inflammation 687–97 CrossRef Medline plays a role in the pathology of primary progressive multiple scle- 25. Zivadinov R, Ramasamy DP, Vaneckova M, et al. Leptomeningeal rosis. Brain J Neurol 2012;135(Pt 10):2925–37 CrossRef Medline contrast enhancement is associated with progression of cortical at- 4. Serafini B, Rosicarelli B, Magliozzi R, et al. Detection of ectopic B-cell rophy in MS: a retrospective, pilot, observational longitudinal follicles with germinal centers in the meninges of patients with sec- study. Mult Scler 2017;23:1336–45 CrossRef Medline ondary progressive multiple sclerosis. Brain Pathol 2004;14:164–74 26. Absinta M, Cortese IC, Vuolo L, et al. Leptomeningeal gadolinium CrossRef Medline enhancement across the spectrum of chronic neuroinflammatory 5. Kowarik MC, Cepok S, Sellner J, et al. CXCL13 is the major determi- diseases. Neurology 2017;88:1439–44 CrossRef Medline

AJNR Am J Neuroradiol 39:1799–1805 Oct 2018 www.ajnr.org 1805 ORIGINAL RESEARCH ADULT BRAIN

Automated Integration of Multimodal MRI for the Probabilistic Detection of the Central Vein Sign in White Matter Lesions

X J.D. Dworkin, X P. Sati, X A. Solomon, X D.L. Pham, X R. Watts, X M.L. Martin, X D. Ontaneda, X M.K. Schindler, X D.S. Reich, and X R.T. Shinohara

ABSTRACT BACKGROUND AND PURPOSE: The central vein sign is a promising MR imaging diagnostic biomarker for multiple sclerosis. Recent studies have demonstrated that patients with MS have higher proportions of white matter lesions with the central vein sign compared with those with diseases that mimic MS on MR imaging. However, the clinical application of the central vein sign as a biomarker is limited by interrater differences in the adjudication of the central vein sign as well as the time burden required for the determination of the central vein sign for each lesion in a patient’s full MR imaging scan. In this study, we present an automated technique for the detection of the central vein sign in white matter lesions. ␲ MATERIALS AND METHODS: Using multimodal MR imaging, the proposed method derives a central vein sign probability, ij, for each ␺ lesion, as well as a patient-level central vein sign biomarker, i. The method is probabilistic in nature, allows site-specific lesion segmen- tation methods, and is potentially robust to intersite variability. The proposed algorithm was tested on imaging acquired at the University of Vermont in 16 participants who have MS and 15 participants who do not.

RESULTS: By means of the proposed automated technique, participants with MS were found to have significantly higher values of ␺ than ␺ ϭ Ϯ ␺ ϭ Ϯ Ͻ those without MS ( MS 0.55 0.18; non-MS 0.31 0.12; P .001). The algorithm was also found to show strong discriminative ability between patients with and without MS, with an area under the curve of 0.88.

CONCLUSIONS: The current study presents the first fully automated method for detecting the central vein sign in white matter lesions and demonstrates promising performance in a sample of patients with and without MS.

ABBREVIATIONS: CVS ϭ central vein sign; MIMoSA ϭ Method for InterModal Segmentation Analysis

ultiple sclerosis is an inflammatory demyelinating disease brain and spinal cord. Currently, assessment of MR imaging fac- Mof the central nervous system characterized by lesions in the tors heavily in the diagnosis of MS, with much importance placed on the distribution (dissemination in space) and time course of lesions (dissemination in time)1 in patients presenting with clin- Received April 5, 2018; accepted after revision June 25. ical symptoms typical for MS. However, current imaging-based From the Department of Biostatistics, Epidemiology, and Informatics (J.D.D., diagnostic criteria favor sensitivity over specificity, making mis- M.L.M., R.T.S.), Perelman School of Medicine, University of Pennsylvania, Philadel- diagnosis of MS relatively common.2,3 Misdiagnosis is especially phia, Pennsylvania; Translational Neuroradiology Section (P.S., M.K.S., D.S.R.), Na- tional Institute of Neurological Disorders and Stroke, National Institutes of Health, prevalent among disorders that demonstrate white matter lesions Bethesda, Maryland; Departments of Neurological Sciences (A.S.) and Radiology similar to those found in MS.4,5 (R.W.), Larner College of Medicine at the University of Vermont, Burlington, Ver- mont; Center for Neuroscience and Regenerative Medicine (D.L.P.), Henry M. Jack- As a means of distinguishing MS lesions from white matter son Foundation, Bethesda, Maryland; Mellen Center for Multiple Sclerosis Treat- abnormalities arising from other diseases, the identification of a ment and Research (D.O.), Cleveland Clinic, Cleveland, Ohio; and Department of Neurology (D.S.R.), Johns Hopkins University School of Medicine, Baltimore, vein traversing the center of a lesion has been proposed as a diag- Maryland. nostic tool because inflammatory demyelination in the MS white This work was supported, in part, by the National Institutes of Health: grants RO1 6,7 NS085211, R01 NS060910, and R21 NS093349 from the National Institute of Neuro- matter is perivenular. The potential for this marker to be used logical Disorders and Stroke, R01 MH 112847 from the National Institute of Mental Health, and R01 EB 017255 from the National Institute of Biomedical Imaging and Please address correspondence to Jordan D. Dworkin, Department of Biostatistics, Bioengineering. The research was also supported by the Intramural Research Pro- Epidemiology, and Informatics, Perelman School of Medicine, University of Penn- gram of National Institute of Neurological Disorders and Stroke (Z01 NS003119) and sylvania, 218 Blockley Hall, 423 Guardian Drive Philadelphia, PA, 19104; e-mail: grants RG-1507-05243 and RG-1707-28586 from the National Multiple Sclerosis [email protected]; @jddwor Society. Indicates open access to non-subscribers at www.ajnr.org The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies. http://dx.doi.org/10.3174/ajnr.A5765

1806 Dworkin Oct 2018 www.ajnr.org in the diagnosis of MS has been advanced by recent developments in MR imaging pulse sequences, which have enabled detailed im- aging of veins in the brain.8-10 Using these sequences, researchers have provided strong evidence that higher proportions of MS le- sions show the central vein sign (CVS) compared with lesions resulting from other disease processes commonly mistaken for MS.7,11-16 This finding has been demonstrated for neuromyelitis optica spectrum disorder, systemic autoimmune diseases, cere- bral small vessel disease, Susac syndrome, and migraine. While further replication in a prospective setting is still necessary, a high proportion of brain MR imaging lesions demonstrating the CVS appear to have potential as a biomarker with high specificity for MS. Unfortunately, important barriers limit the feasibility of the clinical application of the CVS. Two such limitations are the pres- ence of intra- and interrater variability in the subjective assess- ment of the CVS and the time required to adjudicate the CVS in every MR imaging lesion per patient. Recent studies have at- tempted to mitigate the time burden associated with CVS assess- ment by limiting the number of lesions examined.13,17 However, these techniques have the potential to increase variability and FIG 1. A, Axial slice of a lesion on T2-FLAIR volume. B, Axial slice of a have generally not been as successful as the evaluation of the pro- lesion on T2*-EPI volume. C, MIMoSA lesion-probability map. D, Dis- portion of the CVS in all MR imaging lesions per patient.7,18 Most tance-to-lesion-boundary mask with the vesselness filter overlay. The lesion-level CVS probability following permutation was 0.975. important, in studies that adjudicate all lesions per patient, opti- mal proportion cutoffs have differed across study sites and disease lesions to the average can be weighted by the noise in their T2*- comparisons.7,13,18 This variability highlights the need for thor- EPI intensities to account for scan motion. Figure 1 demonstrates ough comparison and optimization of these cutoffs across sam- the steps of the algorithm on a sample lesion. Most important, ples and diseases, yet the same issues of rater subjectivity and while figures are necessarily presented in 2D space, all methods temporal burden make this type of research difficult. Thus, the undertaken for this procedure are conducted in 3D volumetric current study introduces an algorithm for the automatic determi- space and simultaneously consider all 3 planes of the image. nation of the CVS in white matter lesions and presents a fully automated patient-level diagnostic biomarker. In this article, we Vesselness Filtering describe the CVS-detection pipeline, present statistical measures Vein maps in the brain are created to later determine the presence of judgment accuracy, and discuss the implications and next steps or absence of veins in each lesion. To do this, we applied the for this line of research. Frangi vesselness filter20 to the unregistered T2*-EPI volume (for the application to data, this study used the Convert3D Tool; MATERIALS AND METHODS https://sourceforge.net/projects/c3d/), producing a map of scores CVS Detection Algorithm of Ն0, with scores of 0 implying no vesselness qualities. The To adjudicate the CVS for each lesion in a given participant, we Frangi filter is a vessel-enhancement algorithm based on the perform several steps. We first present the overall summary and Hessian matrix at each voxel, in which the second-order structure then address each step, with associated rationale, in detail below. of the image is obtained through convolution with derivatives of To perform the algorithm, we require a T1-weighted volume, T2- Gaussian kernels. The scores are calculated using the eigenvalues weighted FLAIR volume, and T2*-weighted segmented echo-pla- of the Hessian matrix, specifically picking up on tubular struc- nar imaging volume: 1) A map of the veins present in the T2*-EPI tures that are darker (or lighter, depending on the implementa- volume is created using a process referred to as “vesselness filter- tion) than their surroundings. After being obtained in the unreg- ing,” and the vein map is rigidly registered to the T1 volume. 2) istered T2*-EPI space, these vesselness maps are then rigidly White matter lesions are segmented using the T1- and T2-FLAIR registered to the T1 space. volumes. 3) Clear lesion boundaries are then determined using a process that removes ambiguous boundary voxels. 4) Periven- Lesion Segmentation tricular lesions are removed from candidacy, per guidelines given To determine the location and shape of white matter lesions, we by the North American Imaging in Multiple Sclerosis Coopera- performed automatic lesion segmentation on coregistered T1- tive.19 5) A permutation procedure is performed to determine and T2-FLAIR volumes. For the application to data, this study whether identified veins occur in the center of a given lesion to a used the Method for InterModal Segmentation Analysis greater degree than would be expected by chance. This yields a (MIMoSA) model21 in the R statistical environment.22 The lesion probability of a CVS for each lesion j in patient i’s scan, denoted segmentation algorithm produces a map containing the probabil- ␲ ij. Lesion-level CVS probabilities are then averaged to obtain a ity that each voxel is part of a lesion. For the results presented in ␺ patient-level CVS biomarker, denoted i. 6) Contributions of the this article, a threshold of 0.30 was applied to this probability map AJNR Am J Neuroradiol 39:1806–13 Oct 2018 www.ajnr.org 1807 to create a binary lesion mask. The threshold of 0.30 was chosen by future advances in methods for segmenting and counting dis- because previous work has found it to be a conservative cutoff that tinct veins. can limit the amount of false-positive lesion tissue.23,24 Following the definition of a lesion positive for CVS (CVSϩ) given by the CVS Permutation Procedure North American Imaging in Multiple Sclerosis Cooperative,19 we In lesions that contain central veins, one would expect above- removed from candidacy lesions detected by the MIMoSA model average coherence between the centrality of voxels within the le- of Ͻ3 mm in any plane. sion and their vesselness score. The proposed permutation proce- dure takes advantage of that expectation to examine the degree to Lesion-Boundary Determination which the most veinlike voxels of a lesion are more concentrated Thresholding of the lesion-probability map often results in patho- in the center of the lesion than one might expect to observe by logically distinct lesions being connected by ambiguous boundary chance. First, a vein-center coherence score for lesion j in patient voxels. For these lesions to be properly assessed for the CVS, the i’s scan, Cij, is calculated by summing the products of the distance- proposed algorithm addresses this pseudoconfluence through a to-nearest-lesion-boundary of each voxel (ie, centrality) score, d , and its Frangi vesselness score, f . The coherence formula is recently described technique that removes voxels that are con- ijv ijv given by necting pathologically distinct lesions.24 The technique works by finding regions in which the texture of the lesion-probability map ϭ ͸ ϫ Cij v⑀V dijv fijv, resembles the center of a lesion. Therefore, the centers that it produces are maintained and used for investigating the CVS for where V is the set of all voxels in lesion j. Thus, higher values of the remainder of this algorithm. Further detail on the implemen- this score indicate that the highest vesselness values within the lesion tend to occur in the same voxels as the highest centrality tation of this method can be found in the original publication.24 values. Because the North American Imaging in Multiple Sclerosis A lesion-specific null distribution of coherence scores is cre- guidelines call for the exclusion of confluent lesions, the removal ated using 1000 random permutations to determine the degree to of connecting voxels may represent a deviation from these guide- which this score deviates from chance in cases in which there is no lines in cases of true confluence. However, many lesions that biologic correspondence between vesselness and location within would be judged discrete by expert raters are often merged by lesions. For each permutation, p, the vesselness scores of the vox- automated segmentation methods.25 This merging can result els in lesion j are randomly resampled without replacement, yield- in drastic and unrealistic degrees of pseudoconfluence in au- ing a randomly ordered set of values, V*. A null coherence score is Ն p tomated lesion masks, sometimes resulting in 50 distinct le- then calculated using the formula, sions being merged into Ͻ10 lesion components.24 Thus, rely- ϭ ͸ ϫ ing on automated determinations of confluence in automated C*ijp v⑀V,r⑀V*p dijv fijr. lesion masks would likely result in the exclusion of many or This permutation procedure is performed 1000 times, resulting in most eligible lesions. a sample of 1000 null coherence scores. The lesion-level CVS probability, ␲ , is then calculated as the proportion of chance Periventricular Lesion Exclusion ij (null-distributed) CVS scores that are smaller than the observed The density and branching nature of veins near the ventricles score, given by makes assessment for the CVS difficult in periventricular lesions, especially in cases in which Ͼ1 distinct vein traverses the lesion. 1 ␲ ϭ ͸1000 ͉ Ͻ ͉ ϭ I C* C . Thus, the North American Imaging in Multiple Sclerosis Cooper- ij 1000 p 1 ijp ij ative recommends excluding lesions with Ͼ1 vein or with branch- To obtain a subject-level CVS biomarker, ␺ , these probabilities ing veins.19 The proposed algorithm addresses this consideration i are averaged over all lesions observed in patient i. The formula for by excluding periventricular lesions because periventricular le- ␺ , is given by sions typically contain multiple veins. This exclusion is done by i 1 performing tissue-class segmentation on the T1 volumes (for the ␺ ϭ ͸ ␲ i j⑀L ij, application to data, this study used the FMRIB Automated NL Segmentation Tool; FAST; http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/ 26 where NL is the number of candidate lesions in patient i’s scan. fast) , expanding the CSF region of the brain by 3 mm and elim- ␺ The biomarker, i, can be roughly interpreted as the proportion inating lesions from the lesion-center mask that overlap the ex- of the patient’s lesions that demonstrate the CVS. panded CSF region. The choice of a 3-mm expansion was made on the basis of visual inspection of randomly selected T2*-EPI Optional Noise Weighting volumes, for which 3 mm appeared to include most of the branch- When one takes the average of the CVS probabilities for a patient’s ing vein structure discussed in the consensus statement, without lesions, some lesions may have more reliable estimates than oth- removing too much of the deep white matter. Notably, although ers. A more stable biomarker can potentially be obtained by this technique excludes periventricular lesions, it does not exclude weighting each lesion’s contribution to the biomarker by the other lesions that may have multiple veins. This issue represents a amount of noise in the voxels of the lesion on the T2*-EPI vol- second deviation from the North American Imaging in Multiple ume. To estimate the level of noise in a lesion, we first constructed Sclerosis recommendations, which could potentially be addressed a “noiseless” T2*-EPI by performing anisotropic diffusion on the 1808 Dworkin Oct 2018 www.ajnr.org original scan.27 This procedure results in a smoothed volume that Table 1: Demographics of the study samplea maintains tissue boundaries and other image gradients. Then, for Demographics voxels in the lesion, the difference is obtained between the original MS (n ϭ 10) T2*-EPI and the smoothed T2*-EPI. A noise value is finally cal- Age (yr) 44 (16) culated by dividing the sum of the squared voxel differences by the Sex 9/10 Female Disease duration (yr) 9 (7) total number of voxels in the lesion. For lesion j, this value is Disease subtype 10/10 RRMS defined as MS with comorbidities (n ϭ 10) Age (yr) 43 (9) 1 ϭ ͸ ͑ Ϫ s͒2 Sex 9/10 Female nij v⑀V Iv Iv NV Disease duration (yr) 9 (6) s Disease subtype 10/10 RRMS where Iv and Iv are the intensity and smoothed intensity of voxel v, Migraine (n ϭ 10) respectively. The desired reliability weight is simply the inverse of Age (yr) 47 (13) this noise value, Sex 10/10 Female Misdiagnosed with MS (n ϭ 10) 1 Age (yr) 53 (7) ϭ wij . Sex 9/10 Female nij Note:—RRMS indicates relapsing-remitting MS. ␺w a A weighted subject-level biomarker, i , is then calculated by The numbers in parentheses are standard deviations. summing the products of the CVS probabilities of the lesions with their weights and dividing by the sum of the weights. The care, Best, the Netherlands) with a 32-channel dStream head coil. weighted biomarker is given by FLAIR and T1 volumes were obtained with 1-mm isotropic reso- ͸ ␲ ϫ lution, and T2*-EPI volumes were obtained with 0.55-mm isotro- ⑀ w ␺w ϭ j L ij ij pic resolution. N4 bias correction29 was performed on all images, i ͸ j⑀L wij. and the T2-FLAIR volume for each participant was interpolated to a voxel size of 1 mm3 and rigidly coregistered to the T1 volume. Implementation and Software Extracerebral voxels were removed from the T1 volume using a Accompanying this article, code for the central vein detection skull-stripping procedure,30 and the brain mask was applied to algorithm has been made freely available on-line (https://github. the T2-FLAIR volume. com/jdwor/cvs). One file, centralveins_full.R, contains code to run all preprocessing and analysis steps described in the previous Motion-Exclusion Criteria. Because head motion might occur section. This file serves to increase the understanding of all steps during the T2*-EPI scan, potentially producing uninterpretable used in this study and to provide a straightforward tool that can be images, each participant’s T2*-EPI scan was manually rated for applied to raw images. A second file, centralveins_simple.R, con- motion in the relevant white matter regions. Scans were scored tains code to be run directly on a probability map and a vein map. from 1 to 5: One indicated “perfect, no artifacts, and excellent This file serves to improve implementations across different sites signal-to-noise,” 2 indicated “only 1 minor artifact that does not and scanners, for which researchers and clinicians may have pre- obscure any vessels in supratentorial white matter,” 3 indicated ferred pipelines for preprocessing and lesion segmentation. Fol- “more than 1 artifact that does not obscure any vessels in supra- lowing preprocessing and structure segmentation, the cen- tentorial white matter,” 4 indicated “more than 1 artifact that tralveins_simple function was found to take an average of 17.7 Ϯ does obscure some vessels in supratentorial white matter,” and 5 9.1 minutes and was roughly broken down as a 10-minute base- indicated “severe artifacts or bad signal-to-noise that does ob- line with an additional 20 seconds per lesion when run without scure most vessels in supratentorial white matter.” It was decided parallelization. Finally, a third file, helperfunctions.R, provides a priori that scans that were rated 5 would be removed for the additional functions used within the previous 2 files. primary analysis because scans with that degree of motion may be unusable in clinical practice as well. Validation Performance Assessment. Because the CVS shows great promise Data. For this study, data were analyzed for 40 research partici- as a diagnostic biomarker, the performance of this algorithm in pants recruited from the University of Vermont neurology clinic distinguishing MS and non-MS is of primary interest. To deter- ␺ ␺w as part of a study aiming to improve the diagnostic specificity for mine whether the automated biomarkers, i and i , replicate the MS.17 Participants were between 20 and 67 years of age, and 37 findings from previous work that the distribution of manually were women. Ten had MS and no comorbidities known to pro- adjudicated central vein proportion differs between MS and its duce MR imaging white matter abnormalities; 10 had MS and mimics, we used t tests to compare the automated CVS values for comorbidities known to produce MR imaging white matter ab- patients with and without MS. To determine the diagnostic utility ␺ ␺w normalities; 10 had migraine with MR imaging white matter ab- of i and i , we estimated the area under the curve values of the normalities and no other white matter comorbidities; and 10 were receiver operating characteristic curves. The presence of a differ- ␺ ␺w previously incorrectly diagnosed with MS and had MR imaging ence in performance between i and i was tested with the white matter abnormalities and a variety of diagnoses (Table 1). DeLong test for comparing the areas under correlated receiver Whole-brain 3D-T2-FLAIR, T1, and T2*-EPI28 volumes were operating characteristic curves31 using the pROC package in the R acquired on a 3T dStream MR imaging scanner (Philips Health- statistical environment.22,32 Sensitivity and specificity were calcu- AJNR Am J Neuroradiol 39:1806–13 Oct 2018 www.ajnr.org 1809 FIG 2. Boxplots of the patient-level central vein sign biomarker score by diagnostic group. The score can be interpreted as the proportion of lesions that are CVSϩ according to the method described in this article. Groups shaded gray do not have an MS diagnosis, whereas groups shaded gold do. A, Boxplots for the unweighted biomarker. B, Boxplots for the noise-weighted biomarker. Points outside the boxplots represent outliers within their respective groups.

lated using the 40% cutoff,7 under which inflammatory demyeli- To determine the diagnostic utility of the automated biomark- Ն ␺ ␺w nation is diagnosed if 40% of white matter lesions exhibit the ers, i and i , we estimated receiver operating characteristic CVS, as well as the more recently proposed 50% cutoff.18 Addi- curves and calculated their areas under the curve. For the un- ␺ tionally, locally optimal cutoffs were determined, and their sensi- weighted case, i yielded an area under the curve of 0.84 (Fig 3A). ␺ tivity and specificity values were compared with those obtained On the basis of the 40% rule, applying a cutoff of 0.40 to i yielded using established cutoffs. a sensitivity of 0.94 and a specificity of 0.67. On the basis of the Finally, these cutoffs were compared with the performance of 50% rule, applying a cutoff of 0.50 to this biomarker yielded a proportion cutoffs applied to manual determinations of the CVS sensitivity of 0.56 and a specificity of 0.80. Three locally optimal 7,13,18 in previous research, as well as the performance of 3 recently cutoffs appear to occur at 0.38, at which sensitivity was 1.00 and proposed clinical decision rules that do not require the assessment specificity was 0.67; at 0.44, at which sensitivity was 0.75 and spec- of the full set of lesions in a scan. The first such rule, referred to as ificity was 0.73; and at 0.50, at which sensitivity was 0.56 and 13 the rule of 6, states that inflammatory demyelination is diag- specificity was 0.80 (Table 2). Ͼ nosed if there are 6 lesions with the CVS or if more than half of ␺w For the noise-weighted case, i yielded an area under the lesions show the CVS. The second and third, referred to as se- w curve of 0.88 (Fig 3B). Applying a cutoff of 0.40 to ␺ yielded a lect315 and select3*,17 state that inflammatory demyelination is i sensitivity of 0.75 and a specificity of 0.73. Applying a cutoff of diagnosed if the CVS is found in at least 2 of 3 lesions preselected 9 0.50 yielded a sensitivity of 0.56 and a specificity of 0.93. Two on T2-FLAIR and FLAIR* imaging, respectively. ␺w locally optimal cutoffs for i appeared to occur at 0.37, at which sensitivity was 0.94 and specificity was 0.73, and at 0.46, at which RESULTS sensitivity was 0.63 and specificity was 0.93 (Table 2). Although Following manual ratings of scan noise due to motion, 9 partici- pants were excluded and 31 remained for the primary analysis. Of the weighting appeared to produce marginally improved perfor- the remaining 31 participants, 16 had MS and 15 did not. Auto- mance, no significant difference was found using the DeLong test ϭ ϭ mated CVS detection was performed on these 31 participants us- (Z 0.77, P .22). Robustness analysis on the full sample of 40 ing the algorithms and software packages described in the previ- participants after reintroducing the motion-obscured scans ␺ ␺w ous section. Two-sample t tests were run to determine whether showed area under the curve values of 0.77 and 0.81 for i and i , the automated CVS scores differed between the 16 cases with MS respectively. ϭ Ϯ Previous studies that used CVS proportions within patients’ and 15 cases without MS. In both the unweighted (MMS 0.56 ϭ Ϯ Ͻ ϭ 0.17; Mnon-MS 0.37 0.12; P .01) and weighted (MMS full sets of lesions obtained optimal sensitivity/specificity of 1.00/ Ϯ ϭ Ϯ Ͻ 0.55 0.18; Mnon-MS 0.31 0.12; P .001) variants of the 1.00 when comparing cases of MS with undiagnosed cases with- algorithm, the within-patient average CVS probabilities were out MS,7 patients with microangiopathic lesions,13 and patients higher in patients with MS compared with patients without MS. with inflammatory vasculopathies.18 Prior research on a subset of See Fig 2 for breakdowns across all 4 groups. the current sample was unable to obtain perfect discrimination 1810 Dworkin Oct 2018 www.ajnr.org intuition. This study sought to address these issues by introducing an algorithm for automated CVS detection that could, in principle, following further valida- tion, be applied in clinical practice. In the primary analysis, the algo- rithm was tested on a cohort of 16 pa- tients with MS (8 with and 8 without other white matter comorbidities) and 15 patients without MS (8 with migraine and 7 misdiagnosed with MS). The fully automated technique replicated previ- ous work that used manual adjudica- tions7,11-16 by demonstrating that pro- portions of lesions with the CVS differ significantly between MS and its mim- ics. Additionally, the automated bio- ␺ ␺w markers, i and i , were found to have strong diagnostic ability, with areas un- der the curve of 0.84 and 0.88 and opti- mal sensitivity/specificity of approxi- mately 0.94/0.70. There is also great promise for this algorithm to perform consistently across study sites and FIG 3. Receiver operating characteristic curves of MS diagnosis based on patient-level auto- MR imaging scanners because in-house mated CVS biomarker scores. The receiver operating characteristic curve for the unweighted biomarker is shaded blue, and the receiver operating characteristic curve for the weighted bio- preprocessing and lesion-segmentation marker is shaded green. The area under the curve (AUC) values for both curves are displayed in methods can be easily substituted and their respective colors. the remaining steps (obtaining vessel- Table 2: Diagnostic performance of weighted and unweighted biomarkersa ness scores, finding lesion centers, and Threshold Sensitivity Specificity calculating CVS probabilities) do not re- ␺ quire parameter tuning. Unweighted biomarker ( i) 0.38 1.00 (0.75–1.00) 0.67 (0.42–0.84) 0.40 0.94 (0.70–1.00) 0.67 (0.42–0.84) Most important, the automated bio- 0.44 0.75 (0.50–0.90) 0.73 (0.47–0.89) markers presented in this study did not 0.50 0.56 (0.25–0.65) 0.80 (0.58–0.94) ␺w perform as well as previously obtained Weighted biomarker ( i ) 0.37 0.94 (0.70–1.00) 0.73 (0.42–0.84) 0.40 0.75 (0.50–0.90) 0.73 (0.42–0.84) proportions of the CVS based on man- 0.46 0.63 (0.40–0.80) 0.93 (0.68–1.00) ual ratings of all lesions in patients’ 0.50 0.56 (0.35–0.75) 0.93 (0.74–1.00) scans. Specifically, the 40% and 50% a Data in column 2 represent the given cutoff values; data in parentheses are the relevant 95% confidence intervals. cutoffs used in prior manually rated studies often achieved perfect discrimi- between patients with MS and those with migraine when adjudi- nation between patients with and without MS,7,18 which the au- 15 cating the CVS for all lesions. Compared with cutoffs that used tomated biomarkers were not able to replicate. However, previ- the full set of lesions, decision rules based on a subset of lesions ous work in a subset of the current sample showed that manual were generally less discriminative between participants with and ratings of all lesions did not fully distinguish patients with migraine without MS. The rule of 6 did obtain a sensitivity/specificity of from those with MS and no white-matter comorbidities.15 This find- 1.00/1.00 for distinguishing patients with MS and those with ing suggests that the patients without MS in the current sample might 13 small-vessel ischemia, yet in the current sample of MS, mi- be more difficult to distinguish from those with MS using the CVS graine, and misdiagnosed patients, the select3 procedure obtained alone than the patients without MS in the studies that did obtain a sensitivity/specificity of 0.81/0.95 and the select3* procedure perfect discrimination. obtained a sensitivity/specificity of 0.81/0.83.17 Additionally, although the sensitivity and specificity obtained by these biomarkers were lower than those in the manually ob- DISCUSSION Preliminary studies have proposed and validated CVS as a prom- tained CVS proportions, the biomarkers performed comparably 17 ising biomarker for differentiation of MS from other diseases that with decision rules that use only a subset of lesions in a scan. cause MR imaging white matter abnormalities.7,15,19 Yet concerns Thus, while automated adjudication of every lesion in a scan is not remain regarding the heavy temporal burden on manual adjudi- yet as accurate as manual adjudication of every lesion in a scan, the cation of CVS as well as the subjective differences that may arise in proposed automated method shows promise as an alternative to response to variation in the adjudicators’ time constraints and other clinically feasible methods for identifying inflammatory de- AJNR Am J Neuroradiol 39:1806–13 Oct 2018 www.ajnr.org 1811 myelination. Further study and refinement of this technique have REFERENCES the potential to yield biomarkers that are both feasible for use in 1. McDonald WI, Compston A, Edan G, et al. Recommended diagnos- the clinic and comparable in accuracy and reliability with CVS tic criteria for multiple sclerosis: guidelines from the International proportions obtained by manual adjudication. Panel on the Diagnosis of Multiple Sclerosis. Ann Neurol 2001;50: 121–127 Medline There are several important limitations to the proposed algo- 2. Solomon AJ, Klein EP, Bourdette D. “Undiagnosing” multiple rithm. First, biomarker values were found to be lower than previ- sclerosis: the challenge of misdiagnosis in MS. Neurology 2012;78: ously reported CVS proportions for patients with MS and higher 1986–91 CrossRef Medline than previously reported CVS proportions for those without MS. 3. Solomon AJ, Bourdette DN, Cross AH, et al. The contemporary spec- It is possible that this effect is due to errors in lesion segmentation, trum of multiple sclerosis misdiagnosis: a multicenter study. Neu- which would pull the values of participants with and without MS rology 2016;87:1393–99 CrossRef Medline 4. Aliaga ES, Barkhof F. MRI mimics of multiple sclerosis. Handb Clin toward each other due to the assessment of noninformative false- Neurol 2014;122:291–316 CrossRef Medline positive lesions. Because this method allows in-house lesion seg- 5. Miller DH, Weinshenker BG, Filippi M, et al. Differential diagnosis mentation algorithms to be applied, the impact of false-positive of suspected multiple sclerosis: a consensus approach. Mult Scler lesions could potentially be mitigated in practice. It is also possible 2008;14:1157–74 CrossRef Medline that the effect is due to false-positives or false-negatives in auto- 6. Fog T. On the vessel-plaque relationships in the brain in multiple mated CVS assessment. Future work will use manual lesion-level sclerosis. Acta Neurol Scand Suppl 1964;40:(Suppl 10):9–15 Medline 7. Tallantyre EC, Dixon JE, Donaldson I, et al. Ultra-high-field imaging assessments to tease apart these potential sources. distinguishes MS lesions from asymptomatic white matter lesions. Additionally, the exclusion of 9 of the 40 subjects due to noise Neurology 2011;76:534–39 CrossRef Medline in the T2*-EPI scan represents a potential weakness of this auto- 8. Reichenbach JR, Venkatesan R, Schillinge DJ, et al. Small vessels in mated method. However, robustness analysis found that the per- the human brain: MR venography with deoxyhemoglobin as an in- formance of the method on the full sample was not drastically trinsic contrast agent. Radiology 1997;204:272–77 CrossRef Medline 9. Sati, P, George IC, Shea C, et al. FLAIR*: a combined MR contrast reduced compared with the high-quality subset. This finding sug- technique for visualizing white matter lesions and parenchymal gests that in clinical practice, a great deal of motion would not veins. Radiology 2012;265:926–32 CrossRef Medline render a scan useless but instead may be an additional consider- 10. Campion T, Smith RJ, Altmann DR, et al. FLAIR* to visualize veins ation for clinicians when interpreting the results of the algorithm. in white matter lesions: a new tool for the diagnosis of multiple sclerosis? Eur Radiol 2017;27:4257–63 CrossRef Medline 11. Kilsdonk ID, Wattjes MP, Lopez-Soriano A, et al. Improved differ- CONCLUSIONS entiation between MS and vascular brain lesions using FLAIR* at 7 Although the potential clinical implications of an automated tool Tesla. Eur Radiol 2014;24:841–49 CrossRef Medline for CVS adjudication call for further study and refinement of such 12. Kister I, Herbert J, Zhou Y, et al. Ultrahigh-field MR (7 T) imaging of techniques, the current study demonstrates the promising perfor- brain lesions in neuromyelitis optica. Mult Scler 2013;2013:398259 mance of a fully automated method for detecting CVS in white CrossRef Medline 13. Mistry N, Abdel-Fahim R, Samaraweera A, et al. Imaging central matter lesions. To our knowledge, this is the first automated tech- veins in brain lesions with 3-T T2*-weighted magnetic resonance nique for this challenging aspect of MS diagnosis and represents imaging differentiates multiple sclerosis from microangiopathic an important step forward toward a specific MR imaging bio- brain lesions. Mult Scler 2016;22:1289–96 CrossRef Medline marker for MS lesions. 14. Sinnecker T, Do¨rr J, Pfueller CF, et al. Distinct lesion morphology at 7-T MRI differentiates neuromyelitis optica from multiple sclero- Disclosures: Jordan D. Dworkin—RELATED: Grant: National Institutes of Health*. sis. Neurology 2012;79:708–14 CrossRef Medline Andrew Solomon—UNRELATED: Consultancy: Teva Pharmaceuticals, Biogen, EMD 15. Solomon AJ, Schindler MK, Howard DB, et al. “Central vessel sign” Serono, Genentech, Comments: consulting and/or advisory boards; Grants/Grants on 3T FLAIR* MRI for the differentiation of multiple sclerosis from Pending: Biogen, Comments: funding for research*. Dzung L. Pham—RELATED: migraine. Ann Clin Transl Neurol 2016:3:82–87 CrossRef Medline Grant: National Multiple Sclerosis Society, Comments: support from NMSS RG- 16. Wuerfel J, Sinnecker T, Ringelstein EB, et al. Lesion morphology at 7 1507-05243*; UNRELATED: Travel/Accommodations/Meeting Expenses Unrelated Tesla MRI differentiates Susac syndrome from multiple sclerosis. to Activities Listed: Magnetic Resonance Imaging in MS, Comments: gave an invited Mult Scler 2012;18:1592–99 CrossRef Medline talk at the Magnetic Resonance Imaging in MS plenary meeting and workshop in April 2018. Although about MS, the talk was unrelated to the work in this article. Daniel 17. Solomon AJ, Watts R, Ontaneda D, et al. Diagnostic performance of Ontaneda—UNRELATED: Consultancy: Biogen Idec, Genzyme, Merck; Expert Testi- central vein sign for multiple sclerosis with a simplified three-le- mony: Novartis, Genzyme, Genentech*. Daniel S. Reich—RELATED: Grant: National sion algorithm. Mult Scler 2018;24:750–57 CrossRef Medline Multiple Sclerosis Society, Conrad N. Hilton Foundation; UNRELATED: Expert Testi- 18. Maggi P, Absinta M, Grammatico M, et al. Central vein sign differ- mony: Leventhal & Puga; Grants/Grants Pending: National Multiple Sclerosis Soci- entiates multiple sclerosis from central nervous system inflamma- ety, Conrad N. Hilton Foundation, Adelson Medical Research Foundation*; Payment tory vasculopathies. Ann Neurol 2018;83:283–94 CrossRef Medline for Development of Educational Presentations: At the Limits; Travel/Accommoda- 19. Sati P, Oh J, Constable RT, et al; NAIMS Cooperative. The central tions/Meeting Expenses Unrelated to Activities Listed: National MS Society, Race to Erase MS Foundation, Adelson Medical Research Foundation, ECTRIMS, LACTRIMS, vein sign and its clinical evaluation for the diagnosis of multiple ACTRIMS, Consortium of MS Centers, MS Society of Canada, Neurological Research sclerosis: a consensus statement from the North American Imaging Support Society (Istanbul), Alberta MS Network, University of British Columbia, in Multiple Sclerosis Cooperative. Nat Rev Neurol 2016;12:714–22 Mayo Clinic, Gordon Research Conferences, Comments: invitations from nonprofit CrossRef Medline foundations for travel to present scientific lectures or act as federal liaison to an 20. Frangi AF, Niessen WJ, Vincken KL, et al. Multiscale vessel enhance- advisory board.* Russell T. Shinohara—RELATED: Grant: National Institutes of ment filtering. In: Wells WM, Colchester A, Delp S, eds. Proceedings Health, National Multiple Sclerosis Society, Race to Erase MS*; UNRELATED: Board of Medical Image Computing and Computer-Assisted Intervention– Membership: Genentech/Roche*; Consultancy: Genentech/Roche*; Grants/ Grants Pending: National Institutes of Health*; Travel/Accommodations/Meeting MICCAI’98, Cambridge, Massachusetts. October 11–18, 1998 Expenses Unrelated to Activities Listed: European Committee for Treatment and 21. Valcarcel AM, Linn KA, Vandekar SN, et al. MIMoSA: an automated Research in Multiple Sclerosis, Comments: travel grant for conference attendance. method for intermodal segmentation analysis of multiple sclerosis *Money paid to the institution. brain lesions. J Neuroimaging 2018;28:389–98 CrossRef Medline 1812 Dworkin Oct 2018 www.ajnr.org 22. R: A Language and Environment for Statistical Computing. http:// 27. Perona P, Malik J. Scale-space and edge detection using anisotropic www.r-project.org. Accessed June 19, 2018 diffusion. IEEE Trans Pattern Anal Mach Intell 1990;12:629–39 23. Sweeney EM, Shinohara RT, Shiee N, et al. OASIS is automated sta- CrossRef tistical inference for segmentation, with applications to multiple 28. Sati P1, Thomasson DM, Li N, et al. Rapid, high-resolution, whole- sclerosis lesion segmentation in MRI. Neuroimage Clin 2013;2: brain, susceptibility-based MRI of multiple sclerosis. Mult Scler 402–13 CrossRef Medline 2014;20:1464–70 CrossRef Medline 24. Dworkin JD, Linn KA, Oguz I, et al; North American Imaging in 29. Tustison NJ, Avants BB, Cook PA, et al. N4ITK: improved N3 bias cor- Multiple Sclerosis Cooperative. An automated statistical technique rection. IEEE Trans Med Imaging 2010;29:1310–20 CrossRef Medline for counting distinct multiple sclerosis lesions. AJNR Am J Neuro- 30. Carass A, Wheeler MB, Cuzzocreo J, P, et al. A joint registration and radiol 2018;39:626–33 CrossRef Medline segmentation approach to skull stripping. http://www.iacl.ece.jhu. 25. Oguz I, Carass A, Pham DL, et al. Dice overlap measures for edu/ϳprince/pubs/2007-ISBI-Carass-SkullStripping.pdf. Accessed objects of unknown number: application to lesion segmenta- June 19, 2018 tion. In: Crimi A, Bakas S, Kuijf H, eds. Brainlesion: Glioma, Mul- 31. DeLong E, DeLong DM, Clarke-Pearson DL. Comparing the areas tiple Sclerosis, Stroke and Traumatic Brain Injuries. Berlin: Spring- under two or more correlated receiver operating characteristic er-Verlag International Publishing; 2018;10670:3–14 curves: a nonparametric approach. Biometrics 1988;44:837–45 26. Zhang Y, Brady M. Smith S. Segmentation of brain MR images CrossRef Medline through a hidden Markov random field model and the expectation- 32. Robin X, Turck N, Hainard A, et al. pROC: an open-source package -maximization algorithm. IEEE Trans Med Imaging 2001;20:45–57 for R and S؉ to analyze and compare ROC curves. BMC Bioinfor CrossRef Medline matics 2011;12:77 CrossRef Medline

AJNR Am J Neuroradiol 39:1806–13 Oct 2018 www.ajnr.org 1813 ORIGINAL RESEARCH ADULT BRAIN

Predicting Genotype and Survival in Glioma Using Standard Clinical MR Imaging Apparent Diffusion Coefficient Images: A Pilot Study from The Cancer Genome Atlas

X C.-C. Wu, X R. Jain, X A. Radmanesh, X L.M. Poisson, X W.-Y. Guo, X D. Zagzag, X M. Snuderl, X D.G. Placantonakis, X J. Golfinos, and X A.S. Chi

ABSTRACT BACKGROUND AND PURPOSE: Few studies have shown MR imaging features and ADC correlating with molecular markers and survival in patients with glioma. Our purpose was to correlate MR imaging features and ADC with molecular subtyping and survival in adult diffuse gliomas.

MATERIALS AND METHODS: Presurgical MRIs and ADC maps of 131 patients with diffuse gliomas and available molecular and survival data from The Cancer Genome Atlas were reviewed. MR imaging features, ADC (obtained by ROIs within the lowest ADC area), and mean relative ADC values were evaluated to predict isocitrate dehydrogenase (IDH) mutation, 1p/19q codeletion status, MGMT promoter methylation, and overall survival.

RESULTS: IDH wild-type gliomas tended to exhibit enhancement, necrosis, and edema; Ͼ50% enhancing area (P Ͻ .001); absence of a cystic area (P ϭ .013); and lower mean relative ADC (median, 1.1 versus 1.6; P Ͻ .001) than IDH-mutant gliomas. By means of a cutoff value of 1.08 for mean relative ADC, IDH-mutant and IDH wild-type gliomas with lower mean relative ADC (Ͻ1.08) had poorer survival than those with higher mean relative ADC (median survival time, 24.2 months; 95% CI, 0.0–54.9 months versus 62.0 months; P ϭ .003; and median survival time, 10.4 months; 95% CI, 4.4–16.4 months versus 17.7 months; 95% CI, 11.6–23.7 months; P ϭ .041, respectively), regardless of World Health Organization grade. Median survival of those with IDH-mutant glioma with low mean relative ADC was not significantly different from that in those with IDH wild-type glioma. Other MR imaging features were not statistically significant predictors of survival.

CONCLUSIONS: IDH wild-type glioma showed lower ADC values, which also correlated with poor survival in both IDH-mutant and IDH wild-type gliomas, irrespective of histologic grade. A subgroup with IDH-mutant gliomas with lower ADC had dismal survival similar to that of those with IDH wild-type gliomas.

ϭ ϭ ϭ ϭ ϭ ϭ ABBREVIATIONS: IDH isocitrate dehydrogenase; max maximum; min minimum; rADC relative ADC; rADCmean mean relative ADC; TCGA The Cancer Genome Atlas; WHO ϭ World Health Organization

liomas are a heterogeneous group of tumors, and the clinical has high interobserver variation and correlates imperfectly with Gaggressiveness and prognoses are diverse among different clinical outcomes.1,2 Nevertheless, molecular markers, particu- histopathologic grades and molecular subtypes. Previous studies larly isocitrate dehydrogenase (IDH) mutational status, have been have shown that histopathologic classification of diffuse gliomas demonstrated to be significant and more robust prognostic mark- ers3 and have been incorporated into the classification of diffuse gliomas in the latest update of the World Health Organization Received April 25, 2018; accepted after revision July 2. (WHO) classification in 2016.4 IDH mutation, a powerful prog- From the Department of Radiology (C.-C.W., W.-Y.G.), Taipei Veterans General Hospital, Taipei, Taiwan, Republic of China; School of Medicine (C.-C.W., W.-Y.G.), nostic marker of improved survival in diffuse glioma, is found National Yang-Ming University, Taipei, Taiwan, Republic of China; Departments of mainly in lower grade gliomas (WHO grades II and III), but Radiology (C.-C.W., R.J., A.R.), Neurosurgery (R.J., D.P., J.G.), and Pathology (D.Z., M.S.), NYU School of Medicine, New York, New York; Department of Public also in glioblastoma (WHO grade IV), though at much lower Health Sciences and Hermelin Brain Tumor Center (L.M.P.), Henry Ford Hospital, frequency.5,6 Detroit, Michigan; and Neuro-Oncology Program (A.S.C.), Laura and Isaac Perlmut- ter Cancer Center, NYU School of Medicine and Langone Health, New York, New Preoperative and noninvasive determination of molecular York. subtyping is of great value in the clinical management of patients Chih-Chin Wu and Rajan Jain contributed equally to the article. with glioma. However, studies correlating MR imaging features Please address correspondence to Rajan Jain, MD, Department of Radiology, NYU School of Medicine, 660 First Ave, 2nd Floor, New York, NY 10016; e-mail: with IDH-mutation status and patient survival in diffuse gliomas [email protected] are scarce. Recently, we showed that the “T2-FLAIR mismatch http://dx.doi.org/10.3174/ajnr.A5794 sign,” detectable using conventional MR imaging, is a highly spe-

1814 Wu Oct 2018 www.ajnr.org cific imaging biomarker for the IDH-mutant, 1p/19q noncode- rADCmax) was calculated by dividing the tumor ADC by the ADC leted molecular subtype in lower grade gliomas.7 Wang et al8 of the contralateral normal-appearing white matter. demonstrated that the absence of contrast enhancement was as- sociated with longer progression-free and overall survival in patients Statistical Analysis with IDH1-mutated anaplastic gliomas. MR spectroscopy could de- The Kolmogorov-Smirnov test was used to determine whether tect 2-hydroxyglutarate, a metabolite that accumulates in IDH-mu- the numeric data (age and relative ADC values) for each group tant gliomas but did not discover a survival difference.9 Blood were normally distributed. Independent variables (clinical pa- 2 volume estimates obtained by MR perfusion have also provided po- rameter and MR imaging features) were compared using the ␹ tential markers for noninvasive assessment of IDH status.10 test among different molecular groups. Normally distributed ADC can be calculated from DWI, and tumors with more continuous variables (eg, age) were compared using the indepen- freely mobile water molecules and lesser cellularity have higher dent t test or ANOVA test, and non-normally distributed contin- 11 ADC values. ADC has been shown to be a valuable imaging uous variables (rADCmean, rADCmin, and rADCmax) were com- marker in the diagnosis of intracranial lesions as well as in grading pared using the Mann-Whitney U test among different molecular brain tumors.11-15 Therefore, we hypothesized that ADC values groups. The intraobserver reliability of ADC value measuring was obtained from conventional MR imaging could correlate with tested using intraclass correlation coefficients. The optimal cutoff molecular subtype and patient survival in adult diffuse gliomas. value of each rADC was obtained from receiver operating charac- teristic curve analysis when the Youden index reached a maxi- mum. Survival curves were estimated and plotted by the Kaplan- MATERIALS AND METHODS Meier method with log-rank tests to compare Kaplan-Meier This was a retrospective study using data from the publicly avail- curves among groups. Variables were first analyzed by the univar- able National Institutes of Health/National Cancer Institute– iate model. MR imaging features and clinical and molecular approved databases of The Cancer Genome Atlas (TCGA; parameters (including age, sex, WHO grade, MGMT promoter https://cancergenome.nih.gov) and The Cancer Imaging Archive methylation status, IDH mutation status, and 1p/19q codeletion (http://www.cancerimagingarchive.net/),16-18 from which all 461 status) with statistical significance in univariate analysis (P Ͻ .05) cases with imaging data were reviewed, and only cases of treatment- were entered into a Cox proportional hazards ratio model for naı¨ve diffuse gliomas (WHO grades II–IV) with available DWI and multivariate analysis. Statistical significance was defined as ADC maps were included. WHO grade, the status of 3 validated P Ͻ .05 for all tests. The statistical analyses were performed molecular prognostic markers (IDH mutation, 1p/19q codeletion, using the statistical software package SPSS 23.0 (IBM, Ar- MGMT promoter methylation), and survival data were retrieved monk, New York) and R statistical and computing software, from The Cancer Genome Atlas. MR images were reviewed, in con- Version 3.3.2 (http://www.r-project.org). sensus, by 2 board-certified neuroradiologists (with 8 and 17 years of experience) who were blinded to pathologic and molecular diagno- sis. The order of cases viewed was randomized to avoid bias. RESULTS A total of 131 (59 [45%] IDH wild-type and 72 [55%] IDH-mu- Each tumor was scored for 9 MR imaging features according to tant) gliomas were included in this study. Of the 72 IDH-mutant the following criteria modified from the Visually Accessible Rem- tumors, 26 (36%) were 1p/19q codeleted and 46 (64%) were non- brandt Images MR imaging feature set19: T2 signal intensities 1p/19q codeleted. Patients in the IDH wild-type group (mean, (higher than gray matter or mixed [the presence equal to or darker 60 Ϯ 12.2 years) were significantly older than those in the IDH- than that of gray matter part]); T2 homogeneity (homogeneous mutant group (mean, 45.1 Ϯ 13.9 years), and patients with IDH- or heterogeneous); margin (well-defined or not well-defined [ei- mutant, 1p/19q codeleted gliomas (mean, 50.7 Ϯ 13.9 years) were ther infiltrative or irregular]); edema (none to minimal or mild to older than those with IDH-mutant, non-1p/19q codeleted glio- marked); enhancing pattern (non-/minimally enhancing or en- mas (mean, 41.9 Ϯ 12.9 years) (P ϭ .01). hancing); portion of enhancing area (Ͻ50% or Ն50%); the pres- ence of cystic areas (presence or absence); and the presence of Correlation between Conventional MR Imaging Features necrotic areas (presence or absence). We investigated the rela- and Molecular Subtypes tionship among 9 different MR imaging features and 3 molecular Among the conventional MR imaging characteristics, IDH wild- markers (IDH mutation, 1p/19q codeletion, and MGMT pro- type gliomas were more likely to exhibit enhancement (P Ͻ .001), moter methylation) as well as WHO grade and overall survival. Ͼ50% enhancing area (P Ͻ .001), absence of cystic area (P ϭ Diffusion-weighted images were analyzed using OsiriX Imag- .013), the presence of necrosis (P Ͻ .001), and the presence of ing Software (http://www.osirix-viewer.com). ADC measure- edema (P Ͻ .001). Within the IDH-mutant group, there were no ments were generated by manually drawing 3 nonoverlapping MR imaging characteristics to differentiate 1p/19q codeletion sta- 2 ROIs ranging from 40 to 60 mm within the region of lowest ADC tus using the features tested (Table 1). values within the solid component of each tumor on ADC maps. The ADC value was also calculated from contralateral normal- Correlation between rADC Values and Molecular appearing white matter by drawing a single ROI with a size similar Subtypes to that of a tumoral ROI. We obtained mean, minimum (min), Median rADCmean, rADCmin, and rADCmax values of IDH wild- and maximum (max) ADCs of each tumor, respectively, by aver- type gliomas were significantly lower than those of IDH-mutant Ͻ aging the 3 ROIs; and relative ADC (rADCmean, rADCmin, and gliomas (P .001) (Fig 1 and Table 1). Within the IDH-mutant AJNR Am J Neuroradiol 39:1814–20 Oct 2018 www.ajnr.org 1815 Table 1: MR imaging features and IDH-mutation and 1p/19q codeletion status IDH Wild- IDH-Mutant, IDH-Mutant, Type IDH-Mutant P Non-1p/19q Codeleted 1p/19q Codeleted P MR imaging features (No.) (%) T2 signal intensitiesa ϾGray matter 13 (22.4%) 30 (43.5%) .012c 21 (48.8%) 9 (34.6%) .248 ϾMixed 45 (77.6%) 39 (56.5%) 22 (51.2%) 17 (65.4%) T2 homogeneitya Homogeneous 6 (10.3%) 14 (20.3%) .125 10 (23.3%) 4 (15.4%) .544 Heterogeneous 52 (89.7%) 55 (79.7%) 33 (76.7%) 22 (84.6%) Margin Well-defined 32 (54.2%) 27 (38.0%) .065 20 (44.4%) 7 (26.9%) .143 Mixed 27 (45.8%) 44 (62.0%) 25 (55.6%) 19 (73.1%) Edema No-to-minimal 20 (33.9%) 57 (79.2%) Ͻ.001c 38 (82.6%) 19 (73.1%) .339 Mild-to-marked 39 (66.1%) 15 (20.8%) 8 (17.4%) 7 (26.9%) Enhancing patternb None/minimally enhancing 4 (7.0%) 30 (42.3%) Ͻ.001c 19 (41.3%) 11 (44.0%) .826 Enhancing 55 (93.2%) 41 (56.9%) 27 (58.7%) 14 (56.0%) Proportion of enhancing areab Ͻ50% 14 (23.7%) 63 (88.7%) Ͻ.001c 40 (87.0%) 23 (92.0%) .704 Ն50% 45 (76.3%) 8 (11.3%) 6 (13.0%) 2 (8.0%) Cystic area Presence 5 (8.5%) 28 (39.4%) .013c 26 (56.5%) 17 (65.4%) .461 Absence 54 (91.5%) 43 (59.7%) 20 (43.5%) 9 (34.6%) Necrotic areab Presence 49 (83.1%) 16 (22.5%) Ͻ.001c 38 (82.6%) 17 (68.0%) .159 Absence 10 (16.9%) 55 (77.5%) 8 (17.4%) 8 (32.0%) Ͻ c rADCmean (median) 1.1 1.6 .001 1.7 1.5 .071 Ͻ c rADCmin (median) 0.9 1.4 .001 1.4 1.3 .178 Ͻ c rADCmax (median) 1.3 1.8 .001 1.9 1.7 .106 a There were 4 cases lacking T2 MR images. b One case in the IDH-mutant group lacked postcontrast studies. c Statistically significant (P Ͻ .05).

FIG 1. Boxplot representation of rADCmean values by glioma IDH genotype. 1816 Wu Oct 2018 www.ajnr.org Table 2: Comparison of overall survival by relative ADC and glioma IDH-mutation status the optimal cutoff value of 1.08 for a b ␤ Variable Case No. Survival Time Coefficient ( ) P Value rADCmean could differentiate survival

IDH mutant, high rADCmean 63 62.0 differences within both IDH-mutant and vs IDH mutant, low rADCmean .141 .003 IDH wild-type gliomas. Patients having vs IDH wild-type, high rADC .06 Ͻ.001 mean IDH-mutant gliomas with an rADC vs IDH wild-type, low rADC .24 Ͻ.001 mean mean below the cutoff value of 1.08 had poorer IDH mutant, low rADCmean 9 24.2 (0.0–55.0) survival than those with an rADCmean vs IDH wild-type, high rADCmean .45 .614 above 1.08 (P Ͻ .001). In addition, me- IDH wild-type, high rADCmean 30 17.7 (11.6–23.7) vs IDH wild-type, low rADCmean .25 .041 dian survival associated with IDH-mu-

IDH wild-type, low rADCmean 29 10.4 (4.4–16.4) tant gliomas with a low rADCmean was vs IDH mutant, low rADCmean .51 .179 very poor. Survival time for this group a Ն Ͻ High rADCmean: rADCmean values 1.08; low rADCmean: rADCmean values 1.08. was similar to that of those with IDH b Survival time is expressed as median (95% CI) (months). wild-type gliomas with either high or

low rADCmean (Table 2 and Fig 2). Fi- glioma cohort, rADC values trended lower in 1p/19q codeleted nally, the rADCmean cutoff value of 1.08 could also distinguish a gliomas than in noncodeleted gliomas; however, this trend did survival difference within IDH wild-type gliomas (P Ͻ .041). not reach statistical significance. Receiver operating characteristic analysis identified an rADC of 1.2 as the optimal cutoff value mean DISCUSSION to differentiate IDH wild-type and IDH-mutant gliomas irrespec- Mutations in the IDH genes are among the most important diag- tive of WHO grade, with the best combination of sensitivity nostic and prognostic markers of diffuse gliomas.20 Patients with (81.9%) and specificity (74.6%) and area under the curve (0.790; IDH-mutant gliomas have significantly longer survival compared 95% CI, 0.707–0.869; P Ͻ .001). In the analysis of intraobserver with those with IDH wild-type gliomas, and management of these reliability, the intraclass correlation coefficients indicated a 2 molecular subgroups differs significantly. Previous studies have good correlation of the first evaluator (intraclass correlation investigated the potential of various conventional and advanced coefficient, 0.951; 95% CI, 0.829–0.987; P Ͻ .001) and the MR imaging characteristics, including perfusion, diffusion ten- second evaluator (intraclass correlation coefficient, 0.926; 95% sor, and MR spectroscopy, in identifying genetic subtypes of dif- CI, 0.785–0.975; P Ͻ .001). fuse gliomas.10,21,22 Here, our results indicate that rADC values correlate with IDH mutation status as well as survival in both Correlation between rADC and Overall Survival IDH-mutant and IDH wild-type diffuse gliomas, independent of Median overall survival was 25.4 months (95% CI, 19.0–31.7 their WHO grade. Additionally, using rADC, we could identify a months), and overall cumulative survival rates were 79% at 1 year, particularly poor prognosis subset of IDH-mutant gliomas, with 59% at 2 years, 42% at 3 years, and 36% at 5 years in all 131 cases. Univariate survival analysis found survival to be significantly re- outcomes similar to those in patients IDH wild-type disease. Most lated to 6 MR imaging features, including T2 homogeneity, important, determining the rADC value is a simple approach that enhancing pattern, enhancing areas, presence of necrosis, requires no specialized software; hence, our findings potentially have immediate clinical impact. rADCmean, and rADCmin values, in addition to age, WHO grade, MGMT promoter methylation status, IDH-mutation status, and The mechanism by which IDH-mutant and IDH wild-type 1p/19q codeletion status. gliomas differ in terms of the rADC is not clear; however, it may be related to tumor cellularity. In many previous studies, DWI has Multivariate analysis identified IDH status and rADCmean as the only prognostic factors that independently impacted overall shown utility for preoperative grading and outcome of gliomas survival after considering the WHO grade, MGMT promoter and for evaluating the response to therapy in patients with 13,23-27 methylation, and 1p/19q codeletion status and adjusting for pa- glioblastoma. ADC values provide quantitative informa- tient age. Specifically, IDH-mutant gliomas had significantly lon- tion that reflects Brownian motion of water molecules within a ger overall survival (median, 62.0 months) than IDH wild-type scanned area and are determined by many factors. Mainly, differ- gliomas (median, 14.7 months) (P Ͻ .001 by log-rank test; age- ences in ADC have been attributed to tumor cellularity but also to adjusted hazard ratio, 13.5; 95% CI, 4.8–38.4; P Ͻ .001 by multi- the presence of necrosis or cysts and water content in interstitial 11,13,28 variate Cox analysis). Overall cumulative survival rates were 97% space. ADC has been shown to correlate inversely with tu- at 1 year, 71% at 3 years, and 61% at 5 years in IDH-mutant mor cellularity on histologic examination, one of the main fea- gliomas, and 76% at 1 year and 13% at 3 years in IDH wild-type tures of the WHO classification of brain tumors. Our results dem- gliomas. Gliomas with higher rADCmean had longer overall sur- onstrate that most IDH-mutant gliomas exhibit higher rADCmean ϭ vival compared with those with lower rADCmean (P .001 by values and MR imaging features accordant with their low-grade univariate regression; age-adjusted hazard ratio, 0.17; 95% CI, features, while most IDH wild-type gliomas show necrosis and a

0.1–0.6; P ϭ .004 by multivariate Cox analysis). No statistically lower rADCmean in solid portions, likely representing higher cel- significant differences were noted for the remaining MR imaging lularity, which is associated with higher grade features. features, WHO grade, MGMT promoter methylation status, and Previous studies have observed an association between ADC 1p/19q codeletion status in multivariate analysis. and IDH status in gliomas. One study of 37 anaplastic astrocyto- By means of the area under a time-dependent receiver operat- mas showed that the minimum ADC (cutoff point, 0.95 ϫ 10Ϫ3 ing characteristic curve for prediction of survival at 12 months, mm2/s) had acceptable discrimination (area under the curve, AJNR Am J Neuroradiol 39:1814–20 Oct 2018 www.ajnr.org 1817 FIG 2. Kaplan-Meier survival curve subclassification for the present study by relative ADC and IDH genotype. Censored patients are annotated by a asterisks or plus signs. Results of the analysis are provided in Table 2 . IDHmut indicates IDH-mutant; IDHwt, IDH wild-type.

0.711; 95% CI, 0.534–0.887) to predict the IDH status.29 Another ated with relatively decreased global DNA methylation.32 To- retrospective study of 112 cases by Tan et al30 demonstrated that gether, these data clearly indicate that a subgroup of IDH-mutant fractional anisotropy and ADC from diffusion tensor imaging can gliomas behaves as aggressively as their IDH wild-type counter- detect IDH1 mutation in astrocytomas, with the ratio of ADCmin parts. Although whether the malignant subgroups across these being the best metric for detecting IDH mutation, regardless of datasets represent the same biology is unknown, our results sug- the WHO grade. We found that rADCmean can differentiate IDH gest that rADC values can potentially identify this aggressively wild-type from IDH-mutant gliomas with excellent discrimina- behaving IDH-mutant subgroup. tion, regardless of WHO grade. Our study also emphasizes MR Furthermore, our study highlights how detection of robust imaging and ADC values correlating well with molecular subtype. imaging-phenotype correlations can be significantly improved by The main novel finding of our study is that preoperative analyzing glioma datasets by molecular subtype rather than by rADC values can distinguish favorable and unfavorable prognosis histopathologic classification. We evaluated MR imaging features within both IDH-mutant and IDH wild-type glioma subgroups. and prognosis in diffuse gliomas across lower and higher grades in

While IDH-mutant gliomas generally behave less aggressively and the current study and demonstrated the power of rADCmean to have a better prognosis compared with their IDH wild-type coun- differentiate IDH-mutation status and discrete survival sub- terparts, we identified a small subset (12.5%) of IDH-mutant glio- groups beyond WHO grade. While many previous studies have mas with low rADCmean values and poor overall survival, which demonstrated an inverse relationship between ADC and astrocy- was only slightly better (24 months) than that of IDH wild-type toma grade,12,13,28 other studies have shown substantial overlap gliomas but was not statistically significant. Concordantly, a study of ADC values between high-versus-low-grade gliomas3,24,33,34 by Jiao et al31 revealed that a small subgroup (11.7%) of patients and no significant differences between grade II versus III35 or with IDH-mutant gliomas across all grades had a dismal progno- grade III versus IV.36,37 These observed variations of ADC in pre- sis (median survival of 22 months), more similar to IDH wild-type dicting tumor grade are likely due to limitations that make the gliomas and glioblastomas in their cohort. These tumors had dis- exact histopathologic classification challenging, with high inter- tinct genetic characteristics, lacking the typical concurrent genetic observer variability2 and molecular constituent and clinical be- alterations observed in IDH-mutant gliomas. In addition, a recent havior being likely different in tumors with the same histopatho- study of The Cancer Genome Atlas identified a small subset logic grade.3,4,38,39 Similar to a recent meta-analysis by Zulfiqar et (5.5%) of IDH-mutant gliomas with markedly worse survival al,27 which showed that low ADC values correlate independently than other IDH-mutant gliomas, and these tumors were associ- with poor survival in malignant astrocytomas (grades III and IV), 1818 Wu Oct 2018 www.ajnr.org 2016-00221//C32595GG*; Patents (Planned, Pending or Issued): method for treating we found an inverse relationship between rADCmean values and prognosis for both IDH-mutant and IDH wild-type tumors inde- high-grade gliomas*. *Money paid to the Institution. pendent of WHO grade. One limitation of our study is its retrospective design, which REFERENCES was necessary to include a relatively large number of patients and 1. Coons SW, Johnson PC, Scheithauer BW, et al. Improving diagnostic to correlate with survival, which is relatively long in patients with accuracyandinterobserverconcordanceintheclassificationandgrad- ing of primary gliomas. Cancer 1997;79:1381–93 CrossRef Medline IDH mutation. A second limitation is that the studied patients 2. van den Bent MJ. Interobserver variation of the histopathological had been scanned on different MR imaging magnet types, and diagnosis in clinical trials on glioma: a clinician’s perspective. Acta ADC maps were generated by diffusion-weighted imaging or dif- Neuropathol 2010;120:297–304 CrossRef Medline fusion tensor imaging of all collected data. However, a previous 3. Cancer Genome Atlas Research Network; Brat DJ, Verhaak RG, Al- study has verified that ADC datasets from 3-directional DWI and dape KD, et al. Comprehensive, integrative genomic analysis of dif- fuse lower-grade gliomas. N Engl J Med 2015;372:2481–98 CrossRef 6-directional diffusion tensor imaging could be analyzed to- Medline 40 gether. We calculated the rADC to minimize the differences 4. Louis DN, Perry A, Reifenberger G, et al. The 2016 World Health among absolute ADC values across platforms. Third, the treat- Organization Classification of Tumors of the Central Nervous ment regimen applied to each patient was not available to us in System: a summary. Acta Neuropathol 2016;131:803–20 CrossRef many cases. This issue might have potentially impacted the out- Medline 5. Vigneswaran K, Neill S, Hadjipanayis CG. Beyond the World Health come and survival in each case. However, IDH status has been Organization grading of infiltrating gliomas: advances in the mo- repeatedly shown to be an independent marker of prognosis in lecular genetics of glioma classification. Ann Transl Med 2015;3:95 independent datasets.4 Finally, the ADC value has previously been CrossRef Medline reported to predict 1p/19q codeletion status, a marker of oligo- 6. Turkalp Z, Karamchandani J, Das S. IDH mutation in glioma: new insights and promises for the future. JAMA Neurol 2014;71:1319–25 dendroglioma, in lower grade gliomas.35,41 In the study by John- 41 CrossRef Medline son et al, the ADC values were calculated from sampling both 7. Patel SH, Poisson LM, Brat DJ, et al. T2-FLAIR mismatch, an imag- the highest and lowest ADC areas. However, we did not detect a ing biomarker for IDH and 1p/19q status in lower grade gliomas: a significant correlation between rADC values and 1p/19q codele- TCGA/TCIA project. Clin Cancer Res 2017;23:6078–85 CrossRef tion status. Further investigation of the optimal methods of mea- Medline 8. Wang YY, Wang K, Li SW, et al. Patterns of tumor contrast enhance- suring ADC and physiology correlates of ADC values in geneti- ment predict the prognosis of anaplastic gliomas with IDH1 muta- cally defined oligodendroglioma is needed. tion. AJNR Am J Neuroradiol 2015;36:2023–29 CrossRef Medline Our results require independent confirmation, incorporating 9. Natsumeda M, Igarashi H, Nomura T, et al. Accumulation of 2-hy- emerging molecular markers and accounting for different treat- droxyglutarate in gliomas correlates with survival: a study by 3.0- ment strategies. However, our results expand on and refine the Tesla magnetic resonance spectroscopy. Acta Neuropathol Commun 2014;2:158 CrossRef Medline existing correlation between DWI and tumor genetic markers and 10. Kickingereder P, Sahm F, Radbruch A, et al. IDH mutation status is highlight its potential role as an independent imaging biomarker associated with a distinct hypoxia/angiogenesis transcriptome sig- that can aid in substratification of patients with gliomas, both nature which is non-invasively predictable with rCBV imaging in IDH-mutant and IDH wild-type. Here, we were able to identify a human glioma. Sci Rep 2015;5:16238 CrossRef Medline subset of aggressive IDH-mutant gliomas using ADC values easily 11. Gupta RK, Cloughesy TF, Sinha U, et al. Relationships between cho- line magnetic resonance spectroscopy, apparent diffusion coeffi- obtained from common clinical MR images. Ongoing accumula- cient and quantitative histopathology in human glioma. J Neuroon- tion of tumorigenesis knowledge, together with imaging studies col 2000;50:215–26 CrossRef Medline stratified by molecular subgroup rather than histopathologic fea- 12. Yamasaki F, Kurisu K, Satoh K, et al. Apparent diffusion coefficient tures, will likely identify additional robust genetic-imaging-clini- of human brain tumors at MR imaging. Radiology 2005;235:985–91 CrossRef Medline cal phenotype correlations that will improve early detection of 13. Lee EJ, Lee SK, Agid R, et al. Preoperative grading of presumptive clinically meaningful glioma molecular subtypes. low-grade astrocytomas on MR imaging: diagnostic value of mini- mum apparent diffusion coefficient. AJNR Am J Neuroradiol 2008; CONCLUSIONS 29:1872–77 CrossRef Medline 14. Hilario A, Ramos A, Perez-Nun˜ezA,etal.The added value of appar- We demonstrate that ADC values obtained from DWI correlate ent diffusion coefficient to cerebral blood volume in the preopera- with IDH-mutation status and overall survival in adult diffuse tive grading of diffuse gliomas. AJNR Am J Neuroradiol 2012;33: gliomas. IDH wild-type gliomas showed low ADC values and 701–07 CrossRef Medline poor survival compared with IDH-mutant gliomas. Within IDH- 15. Zhang L, Min Z, Tang M, et al. The utility of diffusion MRI with quantitative ADC measurements for differentiating high-grade mutant gliomas, a small subgroup with lower ADC values had from low-grade cerebral gliomas: evidence from a meta-analysis. dismal survival, similar to that in IDH wild-type gliomas. ADC J Neurol Sci 2017;373:9–15 CrossRef Medline values correlated with survival in patients with IDH-mutant and 16. TCGA-GBM collection from The Cancer Imaging Archive (TCIA). IDH wild-type gliomas regardless of WHO grade. Preoperative https://wiki.cancerimagingarchive.net/display/Public/TCGA-GBM. ADC estimates may corroborate with molecular subtypes as a Accessed April 01, 2016 17. TCGA-LGG collection from The Cancer Imaging Archive (TCIA). prognostic marker and potentially enhance risk stratification, es- https://wiki.cancerimagingarchive.net/display/Public/TCGA-LGG. pecially within IDH-mutant gliomas. Accessed April 01, 2016 18. Clark K, Vendt B, Smith K, et al. The Cancer Imaging Archive Disclosures: Dimitris Placantonakis—UNRELATED: Grants/Grants Pending: National (TCIA): maintaining and operating a public information reposi- Institutes of Health, R01 NS102665, NY State Stem Cell Program - DOH01-STEM5- tory. J Digit Imaging 2013;26:1045–57 CrossRef Medline AJNR Am J Neuroradiol 39:1814–20 Oct 2018 www.ajnr.org 1819 19. Wiki for the VASARI feature set The National Cancer Institute Web rospective study of 112 cases. AJNR Am J Neuroradiol 2014;35: site. https://wikicancerimagingarchivenet/display/Public/VASARIϩ 920–27 CrossRef Medline ResearchϩProject. Updated July 25, 2016 31. Jiao Y, Killela PJ, Reitman ZJ, et al. Frequent ATRX, CIC, FUBP1 and 20. Eckel-Passow JE, Lachance DH, Molinaro AM, et al. Glioma groups IDH1 mutations refine the classification of malignant gliomas. On- based on 1p/19q, IDH, and TERT promoter mutations in tumors. cotarget 2012;3:709–22 CrossRef Medline N Engl J Med 2015;372:2499–508 CrossRef Medline 32. Ceccarelli M, Barthel FP, Malta TM, et al. Molecular profiling reveals 21. Xiong J, Tan W, Wen J, et al. Combination of diffusion tensor im- biologically discrete subsets and pathways of progression in diffuse aging and conventional MRI correlates with isocitrate dehydroge- glioma. Cell 2016;164:550–63 CrossRef Medline nase 1/2 mutations but not 1p/19q genotyping in oligodendroglial 33. Lam WW, Poon WS, Metreweli C. Diffusion MR imaging in glioma: tumours. Eur Radiol 2016;26:1705–15 CrossRef Medline does it have any role in the pre-operation determination of grading 22. Leu K, Ott GA, Lai A, et al. Perfusion and diffusion MRI signatures of glioma? Clin Radiol 2002;57:219–25 CrossRef Medline in histologic and genetic subtypes of WHO grade II-III diffuse glio- 34. Murakami R, Hirai T, Sugahara T, et al. Grading astrocytic tumors mas. J Neurooncol 2017;134:177–88 CrossRef Medline by using apparent diffusion coefficient parameters: superiority 23. Murakami R, Hirai T, Kitajima M, et al. Magnetic resonance imaging of a one- versus two-parameter pilot method. Radiology 2009;251: of pilocytic astrocytomas: usefulness of the minimum apparent dif- 838–45CrossRef Medline fusion coefficient (ADC) value for differentiation from high-grade 35. Khayal IS, Vandenberg SR, Smith KJ, et al. MRI apparent diffusion gliomas. Acta Radiol 2008;49:462–67 CrossRef Medline coefficient reflects histopathologic subtype, axonal disruption, and 24. Higano S, Yun X, Kumabe T, et al. Malignant astrocytic tumors: clinical tumor fraction in diffuse-type grade II gliomas. Neuro Oncol 2011; importance of apparent diffusion coefficient in prediction of grade 13:1192–201 CrossRef Medline and prognosis. Radiology 2006;241:839–46 CrossRef Medline 36. Kang Y, Choi SH, Kim YJ, et al. Gliomas: histogram analysis of ap- 25. Murakami R, Sugahara T, Nakamura H, et al. Malignant supraten- parent diffusion coefficient maps with standard- or high-b-value torial astrocytoma treated with postoperative radiation therapy: diffusion-weighted MR imaging–correlation with tumor grade. prognostic value of pretreatment quantitative diffusion-weighted Radiology 2011;261:882–90 CrossRef Medline MR imaging. Radiology 2007;243:493–99 CrossRef Medline 26. Yamasaki F, Sugiyama K, Ohtaki M, et al. Glioblastoma treated with 37. Kitis O, Altay H, Calli C, et al. Minimum apparent diffusion coeffi- postoperative radio-chemotherapy: prognostic value of apparent cients in the evaluation of brain tumors. Eur J Radiol 2005;55:393– diffusion coefficient at MR imaging. Eur J Radiol 2010;73:532–37 400 Medline CrossRef Medline 38. Olar A, Wani KM, Alfaro-Munoz KD, et al. IDH mutation status and 27. Zulfiqar M, Yousem DM, Lai H. ADC values and prognosis of ma- role of WHO grade and mitotic index in overall survival in grade lignant astrocytomas: does lower ADC predict a worse prognosis II-III diffuse gliomas. Acta Neuropathol 2015;129:585–96 CrossRef independent of grade of tumor? A meta-analysis. AJR Am J Roent- Medline genology 2013;200:624–29 CrossRef Medline 39. Reuss DE, Mamatjan Y, Schrimpf D, et al. IDH mutant diffuse and 28. Sugahara T, Korogi Y, Kochi M, et al. Usefulness of diffusion-weighted anaplastic astrocytomas have similar age at presentation and little MRIwithecho-planartechniqueintheevaluationofcellularityinglio- difference in survival: a grading problem for WHO. Acta Neuro- mas. J Magn Reson Imaging 1999;9:53–60 CrossRef Medline pathol 2015;129:867–73 CrossRef Medline 29. Wasserman JK, Nicholas G, Yaworski R, et al. Radiological and path- 40. Khayal IS, McKnight TR, McGue C, et al. Apparent diffusion coeffi- ological features associated with IDH1–R132H mutation status and cient and fractional anisotropy of newly diagnosed grade II glio- early mortality in newly diagnosed anaplastic astrocytic tumours. mas. NMR Biomed 2009;22:449–55 CrossRef Medline PLoS One 2015;10:e0123890 CrossRef Medline 41. Johnson DR, Diehn FE, Giannini C, et al. Genetically defined oligo- 30. Tan WL, Huang WY, Yin B, et al. Can diffusion tensor imaging dendroglioma is characterized by indistinct tumor borders at MRI. noninvasively detect IDH1 gene mutations in astrogliomas? A ret- AJNR Am J Neuroradiol 2017;38:678–84 CrossRef Medline

1820 Wu Oct 2018 www.ajnr.org ORIGINAL RESEARCH ADULT BRAIN

Near-Term Decrease in Brain Volume following Mild Traumatic Injury Is Detectible in the Context of Preinjury Volumetric Stability: Neurobiologic Insights from Analysis of Historical Imaging Examinations

X A.E. Goldman-Yassen, X K.X. Chen, X D. Edasery, X K. Hsu, X K. Ye, and X M.L. Lipton

ABSTRACT BACKGROUND AND PURPOSE: Neurodegeneration after mild traumatic brain injury may manifest as decreasing regional brain volume that evolves from months to years following mild traumatic brain injury and is associated with worse clinical outcomes. We hypothesized that quantitative brain volume derived from CT of the head, performed for clinical indications during routine care, would change with time and provide insights into the putative neuroinflammatory response to mild traumatic brain injury.

MATERIALS AND METHODS: We searched the electronic medical record of our institution for NCCTs of the head performed in patients with mild traumatic brain injury and included those who also underwent NCCTs of the head 1 month to 1 year before and after mild traumatic brain injury for an indication unrelated to trauma. Controls underwent 3 sequential NCCTs of the head with indications unrelated to trauma. The whole-brain and intracranial volume groups were computed using ITK-SNAP. Brain volumes normalized to intracranial volumes were compared across time points using the Wilcoxon signed-rank test.

RESULTS: We identified 48 patients from 2005 to 2015 who underwent NCCTs of the head in the emergency department for mild traumatic brain injury and had NCCTs of the head performed both before and after mild traumatic brain injury. Median normalized brain volumes significantly decreased on the follow-up study post-mild traumatic brain injury (0.86 versus 0.84, P Ͻ .001) and were similar compared with pre-mild traumatic brain injury studies (0.87 versus 0.86, P ϭ .927). There was no significant difference between normalized brain volumes in the 48 controls.

CONCLUSIONS: A decrease in brain volume following mild traumatic brain injury is detectable on CT and is not seen in similar patients with non-mild traumatic brain injury during a similar timeframe. Given the stability of brain volume before mild traumatic brain injury, CT volume loss may represent the subtle effects of neurodegeneration.

ABBREVIATIONS: IQR ϭ interquartile range; mTBI ϭ mild traumatic brain injury; NCCTH ϭ NCCT of the head; TBI ϭ traumatic brain injury

raumatic brain injury (TBI) affects an estimated 1.7 million patients, a minority of patients experience persistent symp- Tpatients annually in the United States and is a major cause toms, which may be disabling.4 of morbidity and mortality.1,2 Most, 70%–90% of cases of TBI, Conventional CT and MR imaging examinations performed are classified as mild TBI (mTBI), which is also known as con- as part of routine clinical care typically reveal no visible indica- 3 cussion. Although mTBI-related symptoms resolve in most tions of trauma in patients with mTBI. Neuroinflammation, pro- posed as a mechanism underlying persistent dysfunction and brain structural changes following mTBI, however, might mani- Received March 27, 2018; accepted after revision June 29. fest as subtle brain swelling on acute timeframe imaging.5 Al- From the Department of Radiology (A.E.G.-Y., K.X.C., D.E., K.H.), Montefiore Medi- cal Center, Bronx, New York; Department of Epidemiology and Population Health though quantitative studies of brain volume have demonstrated (K.Y.), Albert Einstein College of Medicine, Bronx, New York; and Gruss Magnetic both gray and white matter volume loss after mTBI using MR Resonance Research Center Departments of Radiology, Psychiatry and Behavioral Sciences and Dominick P. Purpura Department of Neuroscience (M.L.L.), Albert imaging, published studies of brain volume in TBI have an im- Einstein College of Medicine, Bronx, New York. portant limitation: They did not have access to pre-mTBI brain Preliminary results were presented at: Annual Meeting of the American Society of Neuroradiology and the Foundation of the ASNR Symposium, April 22–27, 2017; imaging and cannot therefore assess change of brain volume from Long Beach, California. a preinjury baseline.6-16 Moreover, prior studies generally re- Please address correspondence to Michael L. Lipton, MD, PhD, Albert Einstein Col- ported change during the long-term after mTBI. Although prior lege of Medicine, 1300 Morris Park Ave, MRRC, Room 219C, Bronx, NY 10461; e-mail: [email protected]; @AdGoldmanYassen studies have shown a reduction in brain volume after mTBI, in the http://dx.doi.org/10.3174/ajnr.A5769 absence of a baseline, it is unclear whether these changes are the

AJNR Am J Neuroradiol 39:1821–26 Oct 2018 www.ajnr.org 1821 the time of mTBI (cases) or the second CT (controls). Clinical notes associated with NCCTHs performed before mTBI, at the time of mTBI, and after mTBI were re- viewed to assess the clinical state of the pa- tient at the time of each scan, including the mechanism of injury, clinical indications, and signs and symptoms.

CT Image Acquisition and Data Storage During routine clinical care, axial NCCTH images were acquired on Light- Speed VCT and LightSpeed RT16 (GE Healthcare, Milwaukee, Wisconsin) at a kilovolt (peak) ranging from 120 to 140 and milliampere ranging from 280 to 300. For each subject, the hardware and technique remained constant across all time points. We retrieved DICOM files FIG 1. Flowsheet of included and excluded cases. containing 0.625-mm-thick axial slices of the head from the PACS and stored result of resolving subtle trauma-related edema due to neuroin- them on a password-protected local hard drive. Subjects were flammation or atrophy from neurodegeneration. excluded if 0.625-mm-thick images were not available. CT is widely performed as the initial imaging test of choice for patients who present with acute mTBI, but it has not been re- Image Processing ported as a method to quantify brain volume changes.17-19 In this Processing and analysis were performed by 3 postgraduate year study, we aimed to mine existing clinical CT examinations, per- 3 or postgraduate year 4 radiology residents, supervised by an formed as a part of routine clinical care, to quantify baseline brain American Board of Radiology Certificate of Added Qualifica- volume before mTBI and assess subsequent brain volume change. tion–certified neuroradiologist. Before the processing and Because NCCT of the head (NCCTH) is ordered in patients for a analysis of all cases, a random subset of 18 cases was indepen- variety of reasons unrelated to trauma, we can compare mTBI dently segmented by 2 residents and interrater reliability was examinations with CT performed both before and following assessed. Once the protocol was considered reliable, the resi- mTBI in the same patients, to quantify brain volume changes dents then processed and analyzed the remaining cases inde- following mTBI and determine whether the change suggests res- pendently to conserve time and resources. Incomplete datasets olution of acute swelling, consistent with resolving neuroinflam- were excluded. Next, all images were first visually inspected. mation, or loss of volume, consistent with neurodegeneration. Subjects were excluded if there was evidence of acute or active disease, including hemorrhage, mass, acute or chronic infarct, MATERIALS AND METHODS and hydrocephalus. Images degraded by motion or beam- Study Population and Study Design hardening artifacts were also excluded. Patients with age-re- After obtaining approval from our institutional review board, in- lated abnormalities, such as white matter hypodensities, were cluding a waiver of informed consent for our Health Insurance not excluded. Portability and Accountability Act–compliant study, we used Clini- We assessed total brain and intracranial volumes using semi- cal Looking Glass (Streamline Health, Atlanta, Georgia) to search our automated threshold segmentation in ITK-SNAP, Version 3.6.0 electronic medical records for patients who underwent NCCTH for (www.itksnap.org), adhering to the following standardized pro- the evaluation of mTBI. Specifically, we searched for radiology re- tocol for consistency20: An initial lower threshold of 20 HU and an ports from NCCTHs ordered in the emergency department at Mon- upper threshold of 75 HU were applied. We chose these thresh- tefiore Medical Center from 2005 to 2015 with clinical indications of olds on the basis of maximal differentiation between brain paren- “trauma,” “fall,” “concussion,” or “assault.” Subjects were included if chyma and CSF on visual inspection. We then placed 10-mm they also underwent NCCTH 1–12 months before and after the seeds in the centrum semiovale, corona radiata, basal ganglia, mTBI scan for a nontraumatic indication (Fig 1). If the patient un- adjacent to the occipital horn of each lateral ventricle, and pons. derwent multiple scans in these intervals, the scan closest to the time Seeds were grown to fill the desired volume, followed by inspec- of mTBI was used. Controls were identified by searching Clinical tion of the segmentations and growing of additional seeds to fill Looking Glass for patients who underwent 3 sequential NCCTHs in parenchyma not included by active contour evolution. Structures the emergency department for nontraumatic reasons at time inter- spuriously included in the initial segmentation, such as paraspinal vals similar to those in the patients with mTBI. soft tissues or the optic nerve, were then manually edited. The Demographic information and medical comorbidities were inferior extent of the cerebellar tonsils was defined as the caudal obtained for each individual from the electronic medical records at limit of the segmentation volume. A representative final segmen- 1822 Goldman-Yassen Oct 2018 www.ajnr.org FIG 2. ITK-SNAP screenshot showing total brain volume segmentation. tation is demonstrated in Fig 2. To calculate total intracranial were calculated between the initial scan and the index scan and the volume, we started with the previously computed brain segmen- index scan and follow-up for both the patients and controls. Dif- tation. We then used a lower threshold of Ϫ20 HU and an upper ferences in absolute volume changes between patients and con- threshold of 100 HU to maximize differentiation between the trols were calculated using the Mann-Whitney U test. A 2-tailed P skull and intracranial contents. The segmentation was then grown value Ͻ .05 was considered statistically significant. to fill the intracranial volume, and structures spuriously included, such as paraspinal soft tissues or the optic nerve, were then man- RESULTS ually edited. The inferior extent of the cerebellar tonsils was also We identified 28,777 patients who underwent NCCTH (“index defined as the caudal limit of the intracranial segmentation CT”) in 1 emergency department for evaluation of traumatic head volume. injury. Of these, 751 also underwent NCCTH within 1–12 months both before (“baseline CT”) and after (“follow-up CT”) the index Statistical Analysis Statistical analysis was performed using STATA, Version 12.1 CT (Fig 1). Of the 751 patients, 48 patients had no visible CT (StataCorp, College Station, Texas). Each individual’s brain vol- abnormalities on any CT or history of head trauma at the time of ume was initially normalized to his or her total intracranial vol- either the baseline CT or follow-up CT. In addition, 48 matched ume. Continuous demographic variables were compared using control subjects were identified who underwent 3 CT examina- the Mann-Whitney U test, and categoric variables, with the ␹2 or tions, which were similarly named (baseline, index, and follow-up Fisher’s exact test when appropriate. Interrater reliability was as- CT). sessed using the intraclass correlation coefficient. Brain volumes Patient demographics did not differ between patients with were compared within individuals across timepoints using the mTBI and controls (Table 1). Similarly, there was no significant Wilcoxon signed rank test. Absolute changes in brain volumes difference in the median time between baseline, index, and fol- AJNR Am J Neuroradiol 39:1821–26 Oct 2018 www.ajnr.org 1823 Table 1: Subject demographics is the first report of quantitative analysis mTBI (n = 48) Controls (n = 48) P Value of brain volume changes derived from Median age (IQR) (yr) 63 (52–79) 71 (50–82) .202a NCCTH in patients with mTBI.21 While b Male sex percent (No.) 42% (20) 31% (15) .289 we do not propose the use of CT for rou- a Median pre- to index interval (IQR) (day) 118 (67–239) 171 (87–274) .123 tine clinical follow-up, the ubiquity of Median index to post interval (IQR) (day) 149 (79–204) 124 (61–219) .901a NCCTH performed during routine clin- a Mann-Whitney U test. b ␹2 test. ical practice in the emergency depart- ment makes these data ripe for analysis Table 2: No significant differences in comorbidities that may to characterize and gain insight into the underlying disease pro- affect brain volume between cases and controls cess. We found a significant decrease in brain volumes from the mTBI Controls P (n = 48) (n = 48) Valuea time of mTBI to follow-up 1–12 months later, with evidence of Dementia 23% (11) 25% (12) Ͼ.999 stable brain volume during 1–12 months before mTBI and across End-stage renal disease 8% (4) 8% (4) Ͼ.999 similar timeframes in controls evaluated in the emergency depart- Diabetes mellitus 48% (23) 38% (18) .409 ment for reasons other than TBI. Ͼ Hypertension 75% (36) 77% (37) .999 Changes in brain volume following mTBI have been described Cardiac arrhythmia/valvular 19% (9) 19% (9) Ͼ.999 in many previous studies to correlate with clinical sequelae of disease Seizure disorder 10% (5) 10% (5) Ͼ.999 long-term brain injury. In 2 studies that reported whole-brain Cirrhosis 8% (4) 6% (3) Ͼ.999 volume changes in patients with TBI with a range of severity, Psychiatric disorders 30% (14) 19% (9) .339 average total brain volume decreases of 7.6 mL were found at 1 a Fisher exact test. year in patients with mTBI and 157.3 mL in patients with mild and moderate TBI.6,16 For reference, normal human total brain vol- low-up scans. Additionally, patients with mTBI and controls had ume at 32 years of age averages 1273.6 mL in men and 1131.1 mL similar rates of underlying medical comorbidities (Table 2). in women and declines by about 0.2%–0.5% per year in healthy Among the 48 patients with mTBI, 36 fell (75%), 5 were assaulted individuals.22-24 We showed an absolute decrease of approxi- (10%), 2 were involved in a motor vehicle collision (4%), and 5 mately 40 mL after mTBI at a median of 149 days’ follow-up. sustained trauma by other mechanisms (10%). Broadly, nontrau- matic clinical indications in both mTBI and control groups in- Although our study and others showed significant brain volume cluded altered mental status from a variety of reasons, headache, changes, some studies failed to show any volume differences after 25,26 seizure, dizziness, and weakness. mTBI. Most studies that reported changes in regional brain Among 18 randomly selected subjects, we found excellent volume showed changes in regions such as the frontal gyri, agreement between 2 researchers, with intraclass correlation co- precuneus, temporal gyri, caudate, cingulum, and hippocam- 6,8,15,27,28 efficients of 0.998 (95% confidence interval, 0.995–0.999) for the pus. The ITK-SNAP protocol we used is robust to index CT and 0.965 (95% CI, 0.905-.986) for the follow-up CT. discrimination of bone, fluid, and soft tissue on CT but cannot When we compared subjects across time in a pair-wise fash- discriminate soft-tissue differences (eg, gray versus white mat- ion, brain volume (as a normalized fraction of total intracranial ter) sufficiently to allow regional segmentation. Currently volume) decreased from 0.86 at the time of mTBI to 0.84 at fol- available regional segmentation algorithms such as FreeSurfer low-up (P ϭ .02, Fig 3). We found no significant change in brain (http://surfer.nmr.mgh.harvard.edu) are validated for parcel- 29 volume from the baseline NCCTH to the time of mTBI and no lation of MR imaging. CT images provide substantially lower significant change in brain volume of the controls during the soft-tissue contrast than MR imaging, and we are not aware of same time intervals. any validated methods for parcellation of brain regions based We next compared the trajectory of volume changes between on CT. patients with mTBI and controls. The magnitude of change in Although decreases in brain volume have been reported after adjusted brain volume between study cases at the time of mTBI mTBI, it is unclear if these changes reflect resolution of trauma- and follow-up NCCTH was significantly larger than that of con- related edema, which might be at a level not detectable on visual trols during a similar time interval (0.012; interquartile range inspection, or development of atrophy in the wake of injury. An- [IQR], Ϫ0.001–0.027 versus 0.002; IQR, Ϫ0.008–0.016; P ϭ imal models have shown both early cortical thickening due to .021). There was no difference in the magnitude of adjusted brain transient edema after trauma followed by reduction of cortical 30,31 volume change between the baseline NCCTH for patients with thickness due to atrophy. Our findings of stable brain volume mTBI as well as controls during the same time interval (0.000; between pretrauma and index mTBI scans with subsequent de- IQR, Ϫ0.007–0.010 versus 0.004; IQR, Ϫ0.014–0.013; P ϭ .753). creases in volume at follow-up suggest neurodegeneration follow- Brain volume declined from the baseline CT following mTBI sig- ing mTBI rather than acute edema at the time of mTBI with sub- nificantly more than the change for controls during the same time sequent resolution. Most volumetric studies of mTBI do not interval (0.007; IQR, Ϫ0.001–0.026 versus Ϫ0.002; IQR, Ϫ0.015– include comparison with preinjury imaging. However, a longitu- 0.020; P ϭ .034). dinal study of ice hockey players included MR imaging at the beginning and end of a season of play as well as additional scans DISCUSSION for players who sustained mTBI during that time. Brain vol- Although CT is the recommended initial method for evaluating umes at the end of the hockey season declined in all players, patients with mild traumatic brain injury, to our knowledge, this regardless of incident mTBI, and no increase in brain volume 1824 Goldman-Yassen Oct 2018 www.ajnr.org in volume due to aging during 12 months makes it highly implausible that we would detect a change due to aging in our follow-up interval (mTBI median, 267 days; control median, 295 days). Moreover, to detect a de- cline due to normal aging during only 1 of the 2 sequential follow-up inter- vals we examined would be very un- likely. Last, our CT protocols could not be optimized prospectively, creat- ing the potential for bias due to image quality. We were, however, able to confirm that there were no changes in imaging equipment during the time- frame of each subject’s 3 CT examina- tions, that voxel size was identical across all CT examinations, and that imaging parameters for the studies we included differed minimally, if at all. Moreover, minimal changes in imag- ing parameters limitation is ubiqui- tous in long-term studies, especially FIG 3. Differences in brain volume across time points in cases and controls. retrospective analyses.

12 was detected at the time of mTBI. The decline in brain vol- CONCLUSIONS umes regardless of concussion status in these hockey players Semiautomated volumetric analysis of the NCCTHs is reliable could represent a consequence of repeat game-related brain and detected a significant decrease in normalized brain volumes trauma in the entire cohort. following mTBI in the setting of preinjury volumetric stability, In addition to mTBI, many other clinical disorders have been suggesting neurodegeneration as the primary mechanism for vol- shown to cause longitudinal change of brain volume, including ume loss. Further study is warranted to determine how these dehydration, schizophrenia, neurodegenerative disorders, and changes correlate with clinical outcomes. normal aging.32-36 Although our patient group with mTBI did exhibit comorbid medical conditions, the control group, who REFERENCES were of similar age, exhibited a similar prevalence of comorbid 1. Faul M, Xu L, Wald MM, et al. Traumatic Brain Injury in the United disorders but did not demonstrate any significant change in brain States: Emergency Department Visits, Hospitalizations and Deaths volume during similar time periods. This finding makes comor- 2002–2006. Atlanta: Centers for Disease Control and Prevention; bid disease an unlikely explanation for the effects we found. We 2010. https://www.cdc.gov/traumaticbraininjury/pdf/blue_book.pdf. considered that administration of IV fluid in the emergency de- Accessed July 19, 2017 2. Faul M, Coronado V. Epidemiology of traumatic brain injury. partment might result in an increase in brain volume, with a Handb Clin Neurol 2015;127:3–13 CrossRef Medline subsequent decrease detectable on follow-up. However, be- 3. Cassidy JD, Carroll LJ, Peloso PM, et al. Incidence, risk factors and cause all 3 CT examinations were performed in the emergency prevention of mild traumatic brain injury: results of the WHO Col- department but the only significant change occurred between laborating Centre Task Force on Mild Traumatic Brain Injury. the index and follow-up CT, and only in patients with mTBI, J Rehabil Med 2004;(43 Suppl):28–60 Medline 4. Centers for Disease Control and Prevention and National Center for we consider hydration status an unlikely explanation of the Injury Prevention and Control. Report to Congress on Mild Trau- findings. matic Brain Injury in the United States: Steps to Prevent a Serious Our study has several limitations. First, the retrospective Public Health Problem. Atlanta: Centers for Disease Control and design cannot confirm a causal relationship between incident Prevention; 2003. https://www.cdc.gov/traumaticbraininjury/pdf/ mTBI and brain volume change. Second, our reliance on clin- mtbireport-a.pdf. Accessed July 19, 2017 5. Schmidt OI, Heyde CE, Ertel W, et al. Closed head injury: an inflam- ical information collected as a part of routine clinical care may matory disease? Brain Res Brain Res Rev 2005;48:388–99 CrossRef be inadequate to determine the true nature of a patient’s clin- Medline ical condition. Third, because of the inconsistency of clinical 6. Zhou Y, Kierans A, Kenul D, et al. Mild traumatic brain injury: record entries created by different clinicians across time, we longitudinal regional brain volume changes. Radiology 2013;267: could not correlate the volume changes we identified with con- 880–90 CrossRef Medline 7. Singh P, Killgore WD. Time dependent differences in gray matter sistently measured clinical outcomes. Fourth, decline in brain volume post mild traumatic brain injury. Neural Regen Res 2016;11: volume with time could confound our results, particularly in 920–21 CrossRef Medline older individuals. However, reference to the published changes 8. Wang X, Xie H, Cotton AS, et al. Early cortical thickness change after AJNR Am J Neuroradiol 39:1821–26 Oct 2018 www.ajnr.org 1825 mild traumatic brain injury following motor vehicle collision. 23. Fotenos AF, Mintun MA, Snyder AZ, et al. Brain volume decline in J Neurotrauma 2015;32:455–63 CrossRef Medline aging: evidence for a relation between socioeconomic status, pre- 9. Lui YW, Xue Y, Kenul D, et al. Classification algorithms using mul- clinical Alzheimer disease, and reserve. Arch Neurol 2008;65:113–20 tiple MRI features in mild traumatic brain injury. Neurology 2014; CrossRef Medline 83:1235–40 CrossRef Medline 24. Fjell AM, Walhovd KB, Fennema-Notestine C, et al. One-year brain 10. Zagorchev L, Meyer C, Stehle T, et al. Differences in regional brain atrophy evident in healthy aging. J Neurosci 2009;29:15223–31 volumes two months and one year after mild traumatic brain in- CrossRef Medline jury. J Neurotrauma 2016;33:29–34 CrossRef Medline 25. Mac Donald CL, Barber J, Andre J, et al. 5-Year imaging sequelae of 11. Meier TB, Bellgowan PS, Bergamino M, et al. Thinner cortex in colle- concussive blast injury and relation to early clinical outcome. Neu- giate football players with, but not without, a self-reported history of roimage Clin 2017;14:371–78 CrossRef Medline concussion. J Neurotrauma 2016;33:330–38 CrossRef Medline 26. Farbota KD, Sodhi A, Bendlin BB, et al. Longitudinal volumetric 12. Jarrett M, Tam R, Herna´ndez-Torres E, et al. A prospective pilot changes following traumatic brain injury: a tensor-based mor- investigation of brain volume, white matter hyperintensities, and phometry study. J Int Neuropsychol Soc 2012;18:1006–18 CrossRef hemorrhagic lesions after mild traumatic brain injury. Front Neurol Medline 2016;7:11 CrossRef Medline 13. Tate DF, York GE, Reid MW, et al. Preliminary findings of cortical 27. Epstein DJ, Legarreta M, Bueler E, et al. Orbitofrontal cortical thin- thickness abnormalities in blast injured service members and their ning and aggression in mild traumatic brain injury patients. Brain relationship to clinical findings. Brain Imaging Behav 2014;8:102–09 Behav 2016;6:e00581 CrossRef Medline CrossRef Medline 28. Bernick C, Banks S. What boxing tells us about repetitive head 14. Mayer AR, Hanlon FM, Ling JM. Gray matter abnormalities in pe- trauma and the brain. Alzheimers Res Ther 2013;5:23 CrossRef diatric mild traumatic brain injury. J Neurotrauma 2015;32:723–30 Medline CrossRef Medline 29. Reuter M, Schmansky NJ, Rosas HD, et al. Within-subject template 15. Monti JM, Voss MW, Pence A, et al. History of mild traumatic brain estimation for unbiased longitudinal image analysis. Neuroimage injury is associated with deficits in relational memory, reduced hip- 2012;61:1402–18 CrossRef Medline pocampal volume, and less neural activity later in life. Front Aging 30. Lewe´n A, Fredriksson A, Li GL, et al. Behavioural and morphologi- Neurosci 2013;5:41 CrossRef Medline cal outcome of mild cortical contusion trauma of the rat brain: in- 16. MacKenzie JD, Siddiqi F, Babb JS, et al. Brain atrophy in mild or fluence of NMDA-receptor blockade. Acta Neurochir (Wien) 1999; moderate traumatic brain injury: a longitudinal quantitative anal- 141:193–202 CrossRef Medline ysis. AJNR Am J Neuroradiol 2002;23:1509–15 Medline 31. Farkas O, Povlishock JT. Cellular and subcellular change evoked by 17. Barbosa RR, Jawa R, Watters JM, et al; Eastern Association for the diffuse traumatic brain injury: a complex web of change extending Surgery of Trauma. Evaluation and management of mild traumatic far beyond focal damage. Prog Brain Res 2007;161:43–59 CrossRef brain injury: an Eastern Association for the Surgery of Trauma Medline practice management guideline. J Trauma Acute Care Surg 2012;73: 32. Duning T, Kloska S, Steinstra¨ter O, et al. Dehydration confounds the S307–14 CrossRef Medline assessment of brain atrophy. Neurology 2005;64:548–50 CrossRef 18. Tavender EJ, Bosch M, Green S, et al. Quality and consistency of guide- Medline lines for the management of mild traumatic brain injury in the emer- 33. Scahill RI, Frost C, Jenkins R, et al. A longitudinal study of brain gency department. Acad Emerg Med 2011;18:880–89 CrossRef Medline volume changes in normal aging using serial registered magnetic 19. McCrory P, Meeuwisse W, Dvor˘a´k J, et al. Consensus statement on resonance imaging. Arch Neurol 2003;60:989–94 CrossRef Medline concussion in sport-the 5th international conference on concus- sion in sport held in Berlin, October 2016. Br J Sports Med 2017;51: 34. Bernard C, Helmer C, Dilharreguy B, et al. Time course of brain 838–47CrossRef Medline volume changes in the preclinical phase of Alzheimer’s disease. Alz- 20. Yushkevich PA, Piven J, Hazlett HC, et al. User-guided 3D active con- heimers Dement 2014;10:143–51.e1 CrossRef Medline toursegmentationofanatomicalstructures:significantlyimprovedef- 35. Mathalon DH, Sullivan EV, Lim KO, et al. Progressive brain volume ficiency and reliability. Neuroimage 2006;31:1116–28 CrossRef Medline changes and the clinical course of schizophrenia in men: a longitu- 21. Shetty VS, Reis MN, Aulino JM, et al. ACR Appropriateness Criteria dinal magnetic resonance imaging study. Arch Gen Psychiatry 2001; Head Trauma. J Am Coll Radiol 2016;13:668–79 CrossRef Medline 58:148–57 CrossRef Medline 22. Allen JS, Damasio H, Grabowski TJ. Normal neuroanatomical vari- 36. Guevara C, Bulatova K, Soruco W, et al. Retrospective diagnosis of ation in the human brain: an MRI-volumetric study. Am J Phys An- parkinsonian syndromes using whole-brain atrophy rates. Front thropol 2002;118:341–58 CrossRef Medline Aging Neurosci 2017;9:99 CrossRef Medline

1826 Goldman-Yassen Oct 2018 www.ajnr.org ORIGINAL RESEARCH ADULT BRAIN

3D Black-Blood Luminal Angiography Derived from High-Resolution MR Vessel Wall Imaging in Detecting MCA Stenosis: A Preliminary Study

X X. Bai, X P. Lv, X K. Liu, X Q. Li, X J. Ding, X J. Qu, and X J. Lin

ABSTRACT BACKGROUND AND PURPOSE: 3D high-resolution vessel wall imaging is increasingly used for intracranial arterial diseases. This study compared the diagnostic performance of black-blood luminal angiography derived from 3D vessel wall imaging with source images of vessel wall imaging and TOF-MRA in detecting middle cerebral artery stenosis.

MATERIALS AND METHODS: Sixty-two patients with suspected MCA atherosclerosis underwent TOF-MRA, vessel wall imaging, and CTA. Intracranial black-blood luminal angiography was created from source images of vessel wall imaging using minimum intensity projection. The degree and length of MCA stenosis were measured on source images of vessel wall imaging, TOF-MRA, and black-blood luminal angiography and compared using CTA as a reference standard.

RESULTS: The image quality of black-blood luminal angiography was diagnostic in most patients. The intra- and interobserver agreement for both stenosis degree and length measurements was excellent for black-blood luminal angiography. It was comparable with that of source images of vessel wall imaging in grading stenosis. Compared with TOF-MRA, black-blood luminal angiography showed significantly higher sensitivity for the detection of severe stenosis (89.3% versus 64.3%, P ϭ .039) and higher specificity for the detection of occlusion (95.4% versus 84.6%, P ϭ .039). Lesion length estimated on source images of vessel wall imaging was significantly greater than that measured by CTA and black-blood luminal angiography (P Ͻ .001 and P ϭ .010).

CONCLUSIONS: Black-blood luminal angiography is better than TOF-MRA in detecting severe stenosis and occlusion of the MCA. Compared with source images of vessel wall imaging, it is more accurate in evaluating stenosis length. Black-blood luminal angiography can be produced as a derivative from vessel wall imaging and implemented as an adjunct to vessel wall imaging and TOF-MRA without extra acquisition time.

ABBREVIATIONS: BBLA ϭ black-blood luminal angiography; VWI ϭ vessel wall imaging

ntracranial atherosclerotic stenosis is an important cause of nosis with high accuracy,2 but it also exposes patients to ionizing Iischemic stroke and most often involves the proximal segment radiation. Both DSA and CTA use iodinated contrast material and 1 of the middle cerebral artery (MCA). Imaging evaluation of the are restricted in patients who have previous adverse reactions to stenosis is critical to treatment planning. Traditionally, the degree iodine and in patients with impaired renal function. Time-of- of stenosis is evaluated by luminal angiography techniques, in- flight imaging is the most commonly used unenhanced MRA cluding DSA, CTA, and MRA. DSA, although long considered the technique for intracranial arterial imaging; however, its major criterion standard in assessing intracranial atherosclerotic steno- weakness is local signal loss resulting from slow and turbulent sis, is limited by invasiveness, ionizing radiation, and high cost. flow.3 CTA has been widely used for evaluating intracranial artery ste- Recently, intracranial high-resolution vessel wall imaging (VWI) has been implemented for direct depiction of the intracra- Received April 1, 2018; accepted after revision July 8. nial arterial wall. Beyond the features of the arterial wall, dimen- From the Department of Radiology (X.B., P.L., K.L., J.L.), Zhongshan Hospital, Fudan sions of the vessel lumen could be reliably measured on the cross- University and Shanghai Institute of Medical Imaging, Shanghai, China; Depart- 4-6 ments of Neurosurgery (Q.L.) and Neurology (J.D.), Zhongshan Hospital, Fudan Uni- sectional images of VWI. With large spatial coverage, high versity, Shanghai, China; and GE Healthcare (J.Q.),Shanghai, China. signal-to-noise ratio, and flexible orientation of the reformatted Please address correspondence to Jiang Lin, MD, Department of Radiology, Zhong- images, 3D VWI is often chosen to demonstrate tortuous cerebral shan Hospital, Fudan University and Shanghai Institute of Medical Imaging, Fenglin Rd 180, Shanghai 200032, China; e-mail: [email protected] arteries. Clear demonstration of the vessel wall requires suppres- http://dx.doi.org/10.3174/ajnr.A5770 sion of the MR signal of the luminal spins to yield minimum or no

AJNR Am J Neuroradiol 39:1827–32 Oct 2018 www.ajnr.org 1827 signal from the flowing blood. Based on these characteristics of density-weighted imaging were transferred to an ADW4.6 Work- 3D VWI, we attempted to convert the already acquired vessel wall station (GE Healthcare). MIP and MPR images were created from images, referred to as source images of VWI, into 3D black-blood TOF images. Minimum intensity projection was used to generate luminal angiography (BBLA) using minimum intensity projec- BBLA from 3D-Cube proton-density-weighted imaging. tion. In this study, we used CTA as the reference standard and prospectively compared BBLA with source images of VWI and CTA with TOF-MRA for measuring both the stenosis degree and ste- All craniocervical CTAs, from the aortic arch to the distal intra- nosis length of the MCA. Our hypothesis was that this product of cranial arteries, were performed on a 320–detector row CT sys- 3D VWI would be superior to source images of VWI and TOF- tem (Aquilion ONE CT scanner; Toshiba Medical Systems, To- MRA in the evaluation of MCA stenosis. kyo, Japan) with the following scanning parameters: tube voltage, 120 kV(peak); tube current, 146–210 mA; slice thickness, 0.5 mm; MATERIALS AND METHODS FOV, 22 ϫ 22 cm; and matrix size, 512 ϫ 512. With a power Patients injector, 50–70 mL of nonionic iodinated contrast media (Ultra- From February 2017 to January 2018, sixty-two consecutive pa- vist 370, iopromide; Bayer HealthCare, Berlin, Germany) was in- tients suspected of having intracranial atherosclerotic stenosis in- jected intravenously at a rate of 5 mL/s followed by a 30-mL saline volving the MCA were prospectively included in the study. Pa- bolus. CT scanning was initiated using a bolus-tracking technique tients were excluded if they had contraindications to MR imaging at the level of the aortic arch with a trigger threshold of 150 HU. or nonatherosclerotic intracranial diseases, such as vasculitis, CTA raw images were reformatted on a Vitrea fX workstation Moyamoya disease, and arterial dissection, which were clinically (Version 3.10; Toshiba Medical Systems) using MIP and MPR 7-9 diagnosed according to their diagnostic criteria. Patients with reconstructions for stenosis evaluation. extracranial carotid artery stenosis of Ͼ50% were also excluded. All patients underwent intracranial TOF-MRA and 3D VWI, fol- Image Analysis lowed by CTA within 1 week. Institutional review board approval First, an experienced neuroradiologist assessed the image quality for the study design was obtained, and informed consent was of BBLA in demonstrating the MCA. It was graded on a 3-point obtained from all patients. scale as follows10: 1, poor visualization or nondiagnostic; 2, mod- erate visualization, adequate for diagnostic purposes; and 3, good, MR Imaging high quality for diagnostic purposes. Patients in whom image All examinations were performed on a 3T MR imaging system quality was grade 1 were excluded from the comparison. Second, (Discovery MR750; GE Healthcare, Milwaukee, Wisconsin) with 2 other experienced neuroradiologists independently quantified an 8-channel head coil. The MR imaging protocol included TOF- both the degree and length of the stenosis of the MCA from 3D- MRA and high-resolution VWI. TOF-MRA was acquired first Cube proton-density-weighted images, BBLA, TOF-MRA, and through the circle of Willis with the following parameters: TR, 23 CTA. All cases were reviewed in random order. The measure- ms; TE, 3.4 ms; flip angle, 20°; FOV, 22 ϫ 20.7 cm; matrix size, ments were performed using electronic calipers on the worksta- 320 ϫ 224; slice thickness,1.4 mm; mean acquisition time, 5 min- tion. For all imaging modalities, the Warfarin Aspirin Symptom- utes 8 seconds. Using TOF-MRA as a localizer, we performed atic Intracranial Disease criteria were used to measure the MCA pulse sequences with different image contrast weightings, includ- stenosis11,12: luminal stenosis ϭ [1- (Dstenosis / Dnormal)] ϫ ing 2D FSE T2WI, 3D-Cube T1WI (GE Healthcare), and 3D- 100%, where Dstenosis indicates the diameter of the residual lu- Cube proton-density-weighted imaging, for assessing the intra- men at the site of the most severe degree of stenosis, and Dnormal, cranial vessel walls. The parameters for targeted 2D FSE T2WI the diameter of the proximal normal artery. If the proximal artery were the following: TR, 2150 ms; TE, 50 ms; flip angle, 111°; FOV, 10 ϫ 10 cm; matrix, 384 ϫ 384; slice thickness, 2 mm; slice gap, was diseased, the diameter of the distal portion of the artery at its 0.5 mm; NEX, 12; mean acquisition time, 4 minutes 23 seconds. widest point could be used instead. In case of multiple stenosis in The parameters for 3D-Cube T1WI were as follows: TR, 600 ms; the MCA, the most severe one was chosen for evaluation. The Ͻ TE, 13 ms; FOV, 20 ϫ 20 cm; matrix size, 288 ϫ 288; slice thick- degree of stenosis was graded as normal ( 30%), mild (30%– ness, 1 mm; NEX, 0.5; mean acquisition time, 4 minutes 16 sec- 49%), moderate (50%–69%), severe stenosis (70%–99%), and 4,13 Ͻ onds. 3D-Cube proton-density-weighted imaging using variable occlusion (100%). Arteries with stenosis of 30% on all these 14 flip angle refocusing pulses was performed with the following pa- modalities were excluded from analysis. Stenosis length was de- rameters: TR, 2500 ms; TE, 25 ms; FOV, 20 ϫ 20 cm; matrix size, fined as the distance from the proximal-to-distal end of the ste- 13 288 ϫ 288, slice thickness, 1 mm; NEX, 0.5; and mean acquisition notic artery. On source images of VWI, however, the lesion time, 5 minutes 13 seconds. 3D-Cube proton-density-weighted length could not be measured directly but was estimated by imaging was performed in the sagittal plane orthogonal to the multiplying the slice thickness and the number of the slices proximal MCA with intravoxel dephasing for suppression of showing the luminal stenosis in the MCA. The lesion length the signal from the blood. The imaging volume was centered at was not measured in cases with total occlusion. One of the 2 the circle of Willis with voxel dimensions of 0.69 ϫ 0.69 ϫ 1mm neuroradiologists re-evaluated the BBLA 4 weeks later to assess and displayed with a reconstructed resolution of 0.39 ϫ 0.39 ϫ intraobserver agreement. For the degree and length of stenosis, 0.5 mm. the average of the 2 radiologists’ measurements was used for Source images from both TOF-MRA and 3D-Cube proton- the final analysis. 1828 Bai Oct 2018 www.ajnr.org Comparison of the degree of stenosis with source images of VWI, BBLA, and TOF-MRA with RESULTS CTA The image quality of BBLA in demon- Stenosis Degree on Stenosis Degree on CTA strating the MCA was graded as good in VWI/BBLA/TOF-MRA 30%–49% 50%–69% 70%–99% 100% 33 patients and moderate in 27 patients. 30%–49% 15/15/14 0/0/1 0/0/0 0/0/0 Two of 62 (3.2%) patients were ex- 50%–69% 1/1/2 20/19/14 0/0/0 0/0/0 cluded because of poor image quality 70%–99% 0/0/0 1/2/6 26/25/18 0/0/0 100% 0/0/0 0/0/0 2/3/10 15/15/15 from motion artifacts. A total of 80 dis- eased middle cerebral arteries from 60 patients (20 patients had bilateral MCA lesions) were further examined in this study. Among these patients, there were 33 men (55%) and 27 women (45%) with a mean age of 61.3 years (range, 45–80 years). Thirty-nine patients (65%) had ischemic stroke or transient isch- emic attack with hemiparesis, language disorder, or perceptual deficits. The re- maining 21 patients (35%) presented with dizziness or headache. The intra- and interobserver agree- ment for both stenosis degree (intraclass correlation coefficient ϭ 0.982–0.929, respectively) and length measurements (intraclass correlation coefficient ϭ 0.953–0.939, respectively) were excel- lent for BBLA. The Table summarizes the results of MCA stenosis degree measured with source images of VWI, BBLA, and TOF- MRA in comparison with CTA. For ste- nosis degree, there was excellent agree- ment between source images of VWI FIG 1. 3D proton-density-weighted vessel wall source image (A), black-blood luminal angiography ␬ ϭ (B), TOF-MRA (C), and CTA (D) all depict moderate stenosis of the left M1 MCA (arrow). The 3D and CTA ( 0.956; 95% CI, 0.913– vessel wall image (A) shows eccentric plaque (arrow) on the MCA wall. The inset is a magnified 0.998), excellent agreement between vessel wall image. These images are used as source images to generate luminal angiography with BBLA and CTA (␬ ϭ 0.934; 95% CI, a minimum intensity projection. 0.882–0.986), and good agreement be- tween TOF-MRA and CTA (␬ϭ 0.800; Statistical Analysis 95%CI, 0.717–0.883) (Fig 1). Sensitivity Commercially available software (MedCalc for Windows, Ver- and specificity in detecting severe stenosis were 92.9% and 98.1% sion 18; MedCalc Software, Mariakerke, Belgium) was used for with source images of VWI, 89.3% and 96.2% with BBLA, and the statistical analysis. Intra- and interobserver agreement in the 64.3% and 88.5% with TOF-MRA. Sensitivity and specificity in measurement of stenosis degree and length by BBLA was assessed detecting occlusion were 100% and 96.9% with source images of VWI, 100% and 95.4% with BBLA, and 100% and 84.6% with using the intraclass correlation coefficient. Agreement between TOF-MRA. The sensitivity and specificity of source images of the source images of VWI and CTA, between BBLA and CTA, as VWI and BBLA for the detection of severe stenosis or occlusion well as between TOF-MRA and CTA for the evaluation of MCA are comparable (sensitivity, P ϭ .500 and P ϭ 1.000; specificity, stenosis degree was calculated with the Cohen ␬ test. Sensitivity P ϭ 1.000 and P ϭ .500). Compared with TOF-MRA, BBLA and specificity in the detection of severe stenosis and occlusion of showed significantly higher sensitivity for the detection of severe the MCA were calculated for source images of VWI, BBLA, and stenosis (89.3% versus 64.3%, P ϭ .039) and higher specificity for TOF-MRA using CTA as the reference standard. A Wilcoxon the detection of occlusion (95.4% versus 84.6%, P ϭ .039). BBLA matched pairs test was performed to test whether stenosis lengths demonstrated arteries distal to severe stenosis and occlusion in 11 obtained from different imaging modalities were statistically dif- cases in which TOF-MRA did not (Fig 2). The average length of ferent. The strength of agreement of the ␬ and intraclass correla- the stenosis was 4.25 Ϯ 1.58 mm estimated with source images of tion coefficient was categorized as follows: poor, Ͻ0.20; fair, VWI, 4.10 Ϯ 1.48 mm measured on BBLA, 4.11 Ϯ 1.65 mm on 0.21–0.40; moderate, 0.41–0.60; good, 0.61–0.80; and excellent, TOF-MRA, and 3.97 Ϯ 1.63 mm on CTA. The measurement of 0.81–1.00.5 A 2-tailed P value Ͻ .05 was considered indicative of a stenosis length between BBLA and CTA (P ϭ .060) and between significant difference. TOF-MRA and CTA (P ϭ .054) was not significantly different. AJNR Am J Neuroradiol 39:1827–32 Oct 2018 www.ajnr.org 1829 Twelve of 60 patients underwent DSA before possible interventional an- gioplasty. Fourteen diseased MCAs were identified on DSA. They were 9 occlu- sions, 3 severe stenoses, 1 moderate ste- nosis, and 1 mild stenosis. CTA, BBLA, and source images of VWI correctly di- agnosed all of the 14 diseased MCAs (Fig 3) compared with DSA.

DISCUSSION The current common practice for delin- eating the intracranial vessel wall is VWI; the volumetric proton intensity– weighted images obtained from VWI were used as source images in this study. With the application of minimum in- tensity projection to these source im- ages, we further obtained BBLA, which provided additional advantages com- pared with VWI alone. Our study dem- onstrates that the presentation of steno- sis degree and length of the MCA on FIG 2. CTA (A) and black-blood luminal angiography (B) depict severe stenosis of the left M1 MCA BBLA was in excellent agreement with (arrow) and arteries distal to the stenosis. C, TOF-MRA overestimates this severe stenosis as an the observations on CTA, which sup- occlusion (arrow) and fails to show the distal arteries. D, The 3D vessel wall image shows severe stenosis of the MCA and eccentric plaque (arrow) on the vessel wall. ported the high diagnostic quality of BBLA. Compared with source images of VWI, with comparable performance in the grading of stenosis, BBLA was no- ticeably more accurate in the measure- ment of stenosis length. We also found that BBLA was superior to TOF-MRA in terms of its higher sensitivity for the de- tection of severe stenosis and higher specificity for occlusion of the MCA. Furthermore, since BBLA did not take any additional acquisition time, we rec- ommend its use as an adjunct to VWI and TOF-MRA in the evaluation of MCA stenosis during a single session of intracranial VWI. In addition to its accuracy as source images of VWI in the grading of stenosis, BBLA also showed the capacity to pres- ent combined information from both VWI and MRA. By selectively suppress- ing the signal from blood flow, 3D black-blood MRA has been used for the assessment of vessel lumens with medium-to-large diameters, such as the FIG 3. DSA (A) and CTA (B) depict severe stenosis of the left M1 MCA (arrow). C, TOF-MRA carotid artery.15,16 With technical devel- overestimates this severe stenosis as an occlusion (arrow). Both the 3D vessel wall image (D) and opment, the application of 3D black- black-blood luminal angiography (E) depict this severe stenosis correctly. blood MRA has been extended to arter- ies of smaller diameters such as small However, compared with CTA and BBLA, the estimated stenosis intracranial arteries.17,18 Despite the improved resolution in lu- length was significantly larger with source images of VWI (P Ͻ minal imaging, 3D black-blood MRA cannot provide sufficient .001 and P ϭ .010, respectively). information on the vessel wall. However, 2D/3D high-resolution 1830 Bai Oct 2018 www.ajnr.org VWI is a sequence specifically tailored to the intracranial vessel MCA stenosis. Accordingly, invasive DSA is now largely reserved walls by sampling the cross-sections of the target vessels. Another for use before interventional treatment rather than for diagnosis application of VWI is that it can evaluate the luminal stenosis of alone. Third, our study was based on a single vascular pathology the MCA with high accuracy, as already reported by some earlier (ie, MCA stenosis). Although the MCA is the most frequently studies.4,5,19 Liu et al4 reported a significant correlation between involved intracranial artery in stroke, the value of BBLA in other 2D VWI and DSA for the detection of MCA stenosis (Ͼ50%) and diseased vessel territories warrants further investigation. Fourth, occlusion, but stenosis length was not evaluated in that study. owing to technical restrictions, 3D-Cube proton-density VWI Until now, no studies have yet reported the use of VWI for was acquired with an anisotropic resolution, which may impair evaluating MCA stenosis length, which is also a critical parameter the demonstration of stenosis on reformatted BBLA. for potential interventional treatment. It is well-known from pre- vious studies of CTA and MRA that multiple postprocessing tech- CONCLUSIONS niques can provide an overview of the entire target vessel in a BBLA is comparable with CTA in the evaluation of stenosis degree much more straightforward manner than the cross-sectional and the length of the MCA. It is better than TOF-MRA in detect- 20,21 source images. Similarly, source VWI provides only the 2D ing severe stenosis and occlusion and more accurate than VWI in cross-sectional profile of a vessel, making it challenging to form a measuring the stenosis length. BBLA can be produced as a deriv- full picture of the diseased vessel based on individual source im- ative from VWI without extra acquisition time; therefore, it could ages. Without a direct measurement of the stenosis length, how- be implemented as an adjunct to VWI and TOF-MRA. ever, an intuitive estimation from VWI source images is inaccu- rate. As shown in this study, this estimation was significantly greater than the results obtained from CTA or BBLA. In contrast, REFERENCES BBLA shows the entire MCA lumen from different perspectives, 1. Weimar C, Goertler M, Harms L, et al. Distribution and outcome of whereby both the degree and the length of the stenosis can be symptomatic stenoses and occlusions in patients with acute cere- evaluated easily and accurately. More important, BBLA can be bral ischemia. Arch Neurol 2006;63:1287–91 CrossRef Medline presented like conventional TOF-MRA, CTA, and DSA, with 2. Nguyen-Huynh MN, Wintermark M, English J, et al. How accurate is CT angiography in evaluating intracranial atherosclerotic disease? which our clinical colleagues are already familiar. Stroke 2008;39:1184–88 CrossRef Medline 3D TOF-MRA is conventionally used for detecting luminal 3. Willinek WA, Born M, Simon B, et al. Time-of-flight MR angiography: narrowing and serves as a localizer for subsequent VWI. However, comparison of 3.0-T imaging and 1.5-T imaging—initial experience. its imaging quality depends on the prominent inflow effect of Radiology 2003;229:913–20 CrossRef Medline blood spins. It has intravascular signal loss at the stenosis, where 4. Liu Q, Huang J, Degnan AJ, et al. Comparison of high-resolution MRI with CT angiography and digital subtraction angiography for blood flow is slow and turbulent, especially in small intracranial the evaluation of middle cerebral artery atherosclerotic steno-oc- arteries. Therefore, to accurately define the stenosis, especially clusive disease. Int J Cardiovasc Imaging 2013;29:1491–98 CrossRef more severe ones, additional gadolinium-enhanced MRA or CTA Medline is sometimes required.14 In this study, TOF-MRA was found to be 5. Lee NJ, Chung MS, Jung SC, et al. Comparison of high-resolution less accurate than BBLA in detecting and differentiating severe MR imaging and digital subtraction angiography for the character- ization and diagnosis of intracranial artery disease. AJNR Am J Neu- stenosis and occlusion. We recorded 18 cases of overestimation of roradiol 2016;37:2245–50 CrossRef Medline the stenosis when TOF-MRA was used, compared with 6 when 6. Park JE, Jung SC, Lee SH, et al. Comparison of 3D magnetic reso- BBLA was used. Furthermore, BBLA was superior to TOF-MRA nance imaging and digital subtraction angiography for intracranial in the demonstration of arteries or collaterals distal to the severe artery stenosis. Eur Radiol 2017;27:4737–46 CrossRef Medline stenosis or occlusion, which is crucial for clinical intervention. 7. Mandell DM, Mossa-Basha M, Qiao Y, et al; Vessel Wall Imaging Study Group of the American Society of Neuroradiology. Intracra- Nevertheless, BBLA still overestimated stenoses in 6 cases, while nial vessel wall MRI: principles and expert consensus recommen- source images of VWI overestimated stenoses in 4 cases in this dations of the American Society of Neuroradiology. AJNR Am J study. The possible contributing factors were the suboptimal sup- Neuroradiol 2017;38:218–29 CrossRef Medline pression of the blood signal and the anisotropic resolution in our 8. Kim YJ, Lee DH, Kwon JY, et al. High resolution MRI difference reconstructed luminal images. between moyamoya disease and intracranial atherosclerosis. Eur J Neurol 2013;20:1311–18 CrossRef Medline This study has several limitations. First, a small number of 9. Debette S, Compter A, Labeyrie MA, et al. Epidemiology, pathophys- patients were enrolled, which may limit the statistical significance. iology, diagnosis, and management of intracranial artery dissec- Second, CTA rather than DSA was performed for comparison. tion. Lancet Neurol 2015;14:640–54 CrossRef Medline However, CTA has been well-established as a highly accurate ap- 10. van der Kolk AG, Hendrikse J, Brundel M, et al. Multi-sequence proach in identifying intracranial atherosclerotic stenosis. Using whole-brain intracranial vessel wall imaging at 7.0 Tesla. Eur Radiol 2013;23:2996–3004 CrossRef Medline 16–detector row CT, a prior study reported that CTA performed 11. Samuels OB, Joseph GJ, Lynn MJ, et al. A standardized method for Ն very well compared with DSA for the detection of 50% intra- measuring intracranial arterial stenosis. AJNR Am J Neuroradiol cranial stenosis with 97.1% sensitivity and 99.5% specificity.2 In 2000;21:643–46 Medline another earlier study with a 4–detector row technique, Bash 12. Huang J, Degnan AJ, Liu Q, et al. Comparison of NASCET and et al14 found that CTA had a higher sensitivity and positive pre- WASID criteria for the measurement of intracranial stenosis us- ing digital subtraction and computed tomography angiography dictive value than TOF-MRA for the detection of intracranial ste- of the middle cerebral artery. J Neuroradiol 2012;39:342–45 nosis and occlusion. Compared with DSA, which was available in CrossRef Medline 12 patients in our study, CTA provided identical information on 13. Jeon JS, Sheen SH, Hwang GJ, et al. Feasibility of intravenous flat AJNR Am J Neuroradiol 39:1827–32 Oct 2018 www.ajnr.org 1831 panel detector CT angiography for intracranial arterial stenosis. a role in the assessment of intracerebral aneurysms? Eur Radiol AJNR Am J Neuroradiol 2013;34:129–34 CrossRef Medline 2009;19:184–92 CrossRef Medline 14. Bash S, Villablanca J, Jahan R, et al. Intracranial vascular stenosis 18. Alexander AL, Buswell HR, Sun Y, et al. Intracranial black-blood MR and occlusive disease: evaluation with CT angiography, MR angiog- angiography with high-resolution 3D fast spin echo. Magn Reson raphy, and digital subtraction angiography. AJNR Am J Neuroradiol Med 1998;40:298–310 CrossRef Medline 2005;26:1012–21 Medline 19. Ryu CW, Jahng GH, Kim EJ, et al. High resolution wall and lumen 15. Lv P, Lin J, Guo D, et al. Detection of carotid artery stenosis: a com- MRI of the middle cerebral arteries at 3 Tesla. Cerebrovasc Dis 2009; parison between 2 unenhanced MRAs and dual-source CTA. AJNR 27:433–42 CrossRef Medline Am J Neuroradiol 2014;35:2360–65 CrossRef Medline 20. Lell MM, Anders K, Uder M, et al. New techniques in CT angiogra- 16. Zhao H, Wang J, Liu X, et al. Assessment of carotid artery atheroscle- phy. Radiographics 2006;26(Suppl 1):S45–62 CrossRef Medline rotic disease by using three-dimensional fast black-blood MRimaging: 21. Ota H, Takase K, Rikimaru H, et al. Quantitative vascular measure- comparison with DSA. Radiology 2015;274:508–16 CrossRef Medline ments in arterial occlusive disease. Radiographics 2005;25:1141–58 17. Stivaros SM, Harris JN, Adams W, et al. Does black blood MRA have CrossRef Medline

1832 Bai Oct 2018 www.ajnr.org ORIGINAL RESEARCH ADULT BRAIN

Clinical Evaluation of Highly Accelerated Compressed Sensing Time-of-Flight MR Angiography for Intracranial Arterial Stenosis

X S.s. Lu, X M. Qi, X X. Zhang, X X.h. Mu, X M. Schmidt, X Y. Sun, X C. Forman, X P. Speier, and X X.n. Hong

ABSTRACT BACKGROUND AND PURPOSE: Time-of-flight MR angiography is the preferred imaging technique to assess intracranial arterial stenosis but is limited by a relatively long acquisition time. Compressed sensing provides an innovative approach in undersampling k-space to minimize the data-acquisition time. We aimed to evaluate the diagnostic accuracy of compressed sensing TOF for detecting intracranial arterial stenosis by comparison with conventional parallel imaging TOF-MRA.

MATERIALS AND METHODS: Compressed sensing TOF and parallel imaging TOF were performed in 22 patients with intracranial arterial stenosis. The MRA scan times were 2 minutes and 31 seconds and 4 minutes and 48 seconds for compressed sensing TOF and parallel imaging TOF, respectively. The reconstructed resolutions were 0.4 ϫ 0.4 ϫ 0.4 and 0.4 ϫ 0.4 ϫ 0.6 mm3 for compressed sensing TOF and parallel imaging TOF, respectively. The diagnostic quality of the images and visibility of the stenoses were independently ranked by 2 neuroradiologists blinded to the type of method and were compared using the Wilcoxon signed rank test. Concordance was calculated with the Cohen ␬. Edge sharpness of the arteries and the luminal stenosis ratio were analyzed and compared using a paired-sample t test.

RESULTS: The interrater agreement was good to excellent. Compressed sensing TOF resulted in image quality comparable with that of parallel imaging TOF but boosted confidence in diagnosing arterial stenoses (P ϭ .025). The edge sharpness of the intracranial arteries for compressed sensing TOF was significantly higher than that for parallel imaging TOF (P Ͻ .001). The luminal stenosis ratio on compressed sensing TOF showed no significant difference compared with that on parallel imaging TOF.

CONCLUSIONS: Compressed sensing TOF both remarkably reduced the scan time and provided adequate image quality for the diagnosis of intracranial arterial stenosis.

ABBREVIATIONS: CS ϭ compressed sensing; GRAPPA ϭ generalized autocalibrating partially parallel acquisition; PI ϭ parallel imaging

erebrovascular disease is a major cause of morbidity and limitations of CTA include exposure to radiation, the use of io- Cmortality worldwide and can arise from several intracranial dinated contrast agents, and impaired accuracy in the presence of vessel wall pathologies, such as atherosclerosis, dissection, and vascular calcifications.4,5 Time-of-flight MRA is widely used and vasculitis.1,2 Studies have found that intracranial arterial stenosis is a noninvasive technique for intracranial vascular evaluation 3 is highly prevalent in fatal stroke. The imaging of intracranial that requires no exogenous contrast agent. Because conventional vessels is an important tool for the clinical evaluation of cerebro- TOF-MRA is a rather slow imaging technique, pursuing high spa- vascular disease. tial resolution is challenging. The spatial coverage is often com- The preferred imaging techniques used to assess intracranial promised to achieve a half-millimeter resolution and good signal- arterial stenosis in clinical practice include CTA and MRA. The to-noise ratio, while keeping the scan time clinically acceptable.6 Furthermore, relatively long acquisition times may lead to motion artifacts, which disrupt the detection of vascular lesions. Further Received May 17, 2018; accepted after revision July 15. From the Department of Radiology (S.s.L., M.Q., X.Z., X.h.M., X.n.H.), The First Affili- limitations of conventional TOF-MRA include a loss of signal ated Hospital of Nanjing Medical University, Nanjing, Jiangsu Province, China; Sie- intensity related to turbulent and slow flow.7 mens Healthcare GmbH (M.S., C.F., P.S.), Erlangen, Germany; and MR Collaboration NE Asia (Y.S.), Siemens Healthcare, Shanghai, China. Parallel imaging (PI) is routinely used as a method for k-space Please address correspondence to Shan shan Lu, MD, PhD, Department of Radiol- undersampling during the acquisition of TOF-MRA. However, ogy, The First Affiliated Hospital of Nanjing Medical University, No. 300 the acceleration factor of PI is often only 2- or 3-fold because of Guangzhou Rd, Gulou District, Nanjing, Jiangsu Province, China, 210029; e-mail: [email protected] the rapid increase in noise or aliasing at higher acceleration fac- http://dx.doi.org/10.3174/ajnr.A5786 tors.8-10 Compressed sensing (CS) provides an innovative ap-

AJNR Am J Neuroradiol 39:1833–38 Oct 2018 www.ajnr.org 1833 Table 1: MR imaging parameters for CS-TOF and PI-TOF Data were reconstructed using 10 iterations of the Modified Parameters CS-TOF PI-TOF Fast Iterative Shrinkage-Thresholding Algorithm according to a FOV (mm2) 220 ϫ 200 220 ϫ 200 previous report.13 Maximum-intensity-projection images of the TR/TE (ms) 21/3.49 21/3.49 axial, coronal, and sagittal views were reconstructed. Flip angle 18° 18° Matrix 368 ϫ 334 368 ϫ 334 Slice thickness (mm) 0.4 0.6 Imaging Evaluation No. of slabs 4 4 Two senior neuroradiologists (M.Q. and S.s.L., with 10 and 6 Slices per slab 60 40 years’ experience, respectively), who were blinded to both the Slice oversampling 20% 20% patients’ information and the type of reconstruction undertaken, Phase partial Fourier factor None 6/8 independently assessed the diagnostic quality of the images. All Slice partial Fourier factor None None Acceleration factor 10.3 GRAPPA 2 MIP images of PI-TOF and CS-TOF were presented to the 2 neu- Reconstructed voxel 0.4 ϫ 0.4 ϫ 0.4 0.4 ϫ 0.4 ϫ 0.6 roradiologists in random order. The diagnostic quality of the im- size (mm3) ages was graded on an ordinal scale from 0 to 3, with 0 indicating completely blurred arteries and severe artifacts; 1 indicating par- proach to undersampling k-space through the exploitation of the tially obscured arteries and moderate artifacts; 2 indicating good underlying sparsity in the appropriate transform domain,11,12 and clear arteries and slight artifacts; and 3 indicating excellent promising higher acceleration. This has been demonstrated in arteries and no artifacts. For visualization of the arterial stenosis, various MR imaging applications, such as TOF-MRA,11,13 con- the readers first identified a stenosis on the MIP images, and then trast-enhanced MRA,14 cardiac cine imaging,15 pediatric MR im- declared their level of confidence using a 3-point scoring scheme aging, and dynamic contrast-enhanced imaging.16,17 as follows: grade 2, definite stenosis and sufficiently recognized, The reduction of acquisition time using CS-TOF is beneficial, high confidence; grade 1, probable stenosis, moderately confi- but its influence in the diagnosis of intracranial arterial stenosis is dent; grade 0, low confidence. The location of the arterial stenosis not well-known. We therefore aimed to evaluate the diagnostic was also recorded. For any discrepancy between the 2 readers, accuracy of CS-TOF for intracranial arterial stenosis, comparing another senior neuroradiologist (X.n.H. with 20 years’ experi- it with conventional PI-TOF. ence) re-evaluated the images and assisted in reaching a consensus agreement. The consensus scores were used for the subsequent MATERIALS AND METHODS analyses. Patient Selection From December 2017 to January 2018, CS-TOF and PI-TOF were The degree of luminal stenosis was calculated using the follow- ϭ Ϫ performed in 85 patients who underwent TOF-MRA examina- ing formula: stenosis ratio (1 Narrow Lumen Diameter/ ϫ tions in our center. Twenty-two consecutive patients were retro- Reference Lumen Diameter) 100%, according to the Warfarin- 18 spectively recruited for data analysis according to the following Aspirin Symptomatic Intracranial Disease criterion. The reference inclusion criteria: They had at least 1 segment of intracranial ar- lumen was defined as the neighboring segment of normal appearance terial stenosis confirmed on digital subtraction angiography or on proximal to the stenotic site. The measurement results of the luminal either of the TOF-MRA sequences. Patients with no positive vas- stenosis ratio from the 2 neuroradiologists were averaged for subse- cular findings (n ϭ 49) or with other vascular lesions such as quent analysis. aneurysms (n ϭ 11) or obvious motion artifacts (n ϭ 3) on both After that, the MIP images were presented to the 2 neuroradi- CS-TOF and PI-TOF were excluded from further analysis. The ologists in randomized order for a side-by-side comparison the study was approved by the institutional review board for human next day. The readers were blinded to the type of reconstruction. studies. Written informed consent was obtained from all the They viewed the MIP images of CS-TOF and PI-TOF simultane- patients. ously and ranked the images in order of diagnostic quality pref- erence (CS-TOF better than PI-TOF, or equivalent, or PI-TOF MR Angiography Parameters better than CS-TOF) on the basis of the criteria that included All the images were obtained with a Magnetom Skyra 3T MR delineation of the cerebral arteries and recognition of arterial imaging scanner (Siemens, Erlangen, Germany) using a 20-chan- stenosis. nel head-neck coil. A research sequence and reconstruction pro- totype provided by Siemens were used for CS-TOF. All imaging Arterial Sharpness Evaluation parameters for CS-TOF and PI-TOF are listed in Table 1. The The edge sharpness of the intracranial arteries was calculated by a main differences of the 2 image protocols were as follows: The perceptual image sharpness metric called the Perceptual Sharp- acceleration factor of CS-TOF was set at 10.3, while the general- ness Index based on MIP images, according to a previous report,19 ized autocalibrating partially parallel acquisition (GRAPPA) fac- using Matlab (2013b; MathWorks, Natick, Massachusetts). The tor of PI-TOF was set at 2. In addition, the phase partial Fourier Perceptual Sharpness Index represents the perceived sharpness in factor was used for PI-TOF and set at 6/8. For CS-TOF, the ac- an image. Briefly, the method estimates the sharpness on the basis quired voxel size was 0.6 ϫ 0.6 ϫ 0.6 mm3 and reconstructed to of a statistical analysis of local edge gradients. It is a no-reference 0.4 ϫ 0.4 ϫ 0.4 mm3. For PI-TOF, the acquired voxel size was metric and takes properties of the human visual system into ac- 0.6 ϫ 0.6 ϫ 1.2 mm3 and reconstructed to 0.4 ϫ 0.4 ϫ 0.6 mm3. count. Based on perceptual properties, a relationship between the The total acquisition time was 4 minutes and 48 seconds and 2 extracted statistical features and the metric score is established to minutes and 31 seconds for PI-TOF and CS-TOF, respectively. form a Perceptual Sharpness Index. 1834 Lu Oct 2018 www.ajnr.org Statistical Analysis mally distributed or the Wilcoxon signed rank test as appropriate. The interrater reliability was performed using the Cohen ␬ anal- All the statistical analyses were performed using SPSS (Version ysis for grading the diagnostic quality and visualization of arterial 16.0; IBM, Armonk, New York). The P value was 2-sided, and P Ͻ stenosis. The intraclass correlation coefficients were calculated for .05 was considered statistically significant. measurement of the luminal stenosis ratio. Reliabilities Ͻ0.4 were characterized as poor; 0.4–0.6, fair; 0.6–0.8, good; and those RESULTS Ͼ0.8 were deemed excellent. Continuous data were summarized Patient Characteristics as mean Ϯ SD, and they were assessed for normality by the Kol- Among the 22 patients (11 men, 11 women; 28–86 years of age), mogorov-Smirnov test before further comparison. Categoric data 12 had multiple segments of intracranial arterial stenosis and 3 were recorded as counts and percentages. The diagnostic quality were eventually diagnosed with Moyamoya disease. No patient and visualization of arterial stenoses on PI-TOF and CS-TOF with a single stenosis seen only on 1 of the TOF sequences was were compared using the Wilcoxon signed rank test. The luminal found. In total, 48 arterial segments with stenosis were diagnosed. stenosis ratio and the edge sharpness of the intracranial arteries The demographic details of the patients in this study are listed in were compared using a paired-sample t test if the data were nor- Table 2.

Imaging Evaluation ؍ Table 2: Patient demographics (N 22) The interrater agreement for diagnostic-quality grading was 0.776 ؎ Characteristics Mean SD or Number (%) and 0.753 for CS-TOF and PI-TOF, respectively. The senior neu- Male 11 (50.0%) roradiologist had to decide on a consensus for 1 of 22 (4.5%) cases Age (yr) 61.8 Ϯ 16.8 of CS-TOF and 2 of 22 (9.1%) cases of PI-TOF. The diagnostic Stenosis location (R/L) 48 Intracranial internal carotid artery 6/5 quality of most CS-TOF (90.9%) and PI-TOF (77.3%) images was Middle cerebral artery 6/13 graded as excellent. The CS-TOF images provided comparable Anterior cerebral artery 5/2 diagnostic quality with the PI-TOF images (P ϭ .046). The mean Posterior cerebral artery 3/4 edge sharpness of the intracranial arteries for CS-TOF was Basilar artery 1 0.358 Ϯ 0.038, significantly higher than that for PI-TOF (0.267 Ϯ Intracranial vertebral artery 2/1 0.042) (P Ͻ .001). Note:—R indicates right; L, left; SD, standard deviation. The interrater agreement for visual- ization of arterial stenosis was 1.000 and 0.778 for CS-TOF and PI-TOF, respec- tively. The senior neuroradiologist had to decide on a consensus for 2 of 48 (4.2%) segments on PI-TOF. Forty- eight (100%) and 43 (89.6%) segments of arterial stenosis were sufficiently rec- ognized on CS-TOF and PI-TOF, respec- tively. Seven stenosed segments were re- corded as probable stenosis on PI-TOF (grade 1), of which 5 were in the intra- cranial internal carotid artery, and 2 in the proximal M2 segment of the middle cerebral artery. Two segments of the in- tracranial ICA with suspicious stenosis on PI-TOF (Fig 1B) were eventually considered normal after comparison with CS-TOF in the same patients (Fig 1D) by the senior neuroradiologist. These segments were excluded from fur- ther analysis. The other 5 segments were recorded as grade 2 (definite stenosis) on CS-TOF. CS-TOF resulted in more confidence in diagnosing intracranial arterial stenosis than PI-TOF (P ϭ .025). FIG 1. Source images and coronal view of MIP images in a 42-year-old patient. The speckled noise The intraclass correlation coefficient in the center can be seen on the source image of conventional PI-TOF (A), whereas some artifacts with a curved stripe pattern can be seen on the source image of CS-TOF (C). These artifacts are for luminal stenosis measurement was eliminated on the MIP images and have little effect on the visualization of the stenosis. An 0.980 and 0.979 for CS-TOF and PI- obvious stenosis located in the M1 segment of the left middle cerebral artery is sufficiently TOF, respectively. The mean luminal visualized on both PI-TOF (B) and CS-TOF (D)(arrowheads). The edge sharpness of vessels on CS-TOF (D) is higher than that on PI-TOF (B)(short arrows). The image quality of the right stenosis ratio measured on CS-TOF intracranial internal carotid artery (long arrow, B) is improved on CS-TOF (D). (57.9% Ϯ 30.5%) showed no significant AJNR Am J Neuroradiol 39:1833–38 Oct 2018 www.ajnr.org 1835 Table 3: Comparison between CS-TOF and PI-TOF for evaluating intracranial arterial stenosis P Variables CS-TOF PI-TOF Value Diagnostic qualitya .046 Grade 3 20 (90.9%) 17 (77.3%) Grade 2 2 (9.1%) 4 (18.2%) Grade 1 0 (0.0%) 1 (4.5%) Grade 0 0 (0.0%) 0 (0.0%) Stenosis visualizationb .025 Grade 2 48 (100.0%) 43 (89.6%) Grade 1 0 (0.0%) 5 (10.4%) Grade 0 0 (0.0%) 0 (0.0%) FIG 4. Bar plots of the 2 readers’ preferences. CS-TOF is considered Ϯ Ϯ Luminal stenosis 57.9% 30.5% 58.9% 31.0% .241 not inferior to PI-TOF in all cases. In 50.0% and 45.5% of patients, the ratio (mean Ϯ SD) diagnostic quality of CS-TOF is considered better than that of PI-TOF Edge sharpness of 0.358 Ϯ 0.038 0.267 Ϯ 0.042 Ͻ.001 by each of the 2 readers. artery (mean Ϯ SD) a The diagnostic quality of the CS-TOF and PI-TOF images was graded on an Side-by-Side Comparison ordinal scale from 0 to 3, with 0 indicating completely blurred arteries and severe The interrater agreement for side-by-side comparisons of the CS- artifacts; 1 indicating partially obscured arteries and moderate artifacts; 2 indicat- ing good and clear arteries and slight artifacts; and 3 indicating excellent arteries TOF and PI-TOF was 0.727. Figure 4 shows bar plots summariz- and no artifacts. ing the 2 readers’ preference. CS-TOF was considered not inferior b The visualization of arterial stenosis was graded as follows: grade 2, definite stenosis to PI-TOF in all the cases. In nearly half of the cases (50.0% for and sufficiently recognized, high confidence; grade 1, probable stenosis, moderately confident; grade 0, low confidence. reader 1 and 45.5% for reader 2), the diagnostic quality of CS-TOF was better than that of PI-TOF. For the other cases, the diagnostic quality of CS-TOF and PI-TOF was considered equivalent.

DISCUSSION In the present study, we evaluated the diagnostic accuracy of CS- TOF for intracranial arterial stenosis by comparison with conven- tional PI-TOF. Our results showed that CS-TOF resulted in more clear visualization of arteries and more confidence in diagnosing FIG 2. MIP images of a 68-year-old patient. Mild stenosis located in the proximal M1 segment of left middle cerebral artery can be suffi- intracranial arterial stenosis than PI-TOF, even in a reduced ac- ciently visualized on both PI-TOF (A) and CS-TOF (B)(arrowheads). quisition time of 2 minutes and 31 seconds. The edge sharpness of vessels on CS-TOF is higher than that on PI- 3D TOF MRA is an effective and widely used tool to noninva- TOF (arrows). sively evaluate and follow-up patients with cerebrovascular dis- ease.20,21 However, to maintain a reasonable spatial resolution and provide detailed depictions of the vessels, the acquisition time is relatively long for conventional PI-TOF. Reduction of the scan time is clinically significant because motion artifacts would be decreased in a shortened scan time, which would particularly ben- efit patients with acute ischemic stroke. CS is a novel technique that uses random undersam- pling.11,12,22 TOF-MRA is well-suited to CS because high-signal vessels are sparse in space.14,23-25 Previously, sparse TOF has been investigated in patients with cerebral aneurysms by Fushimi et al.26 They reported that cerebral aneurysms were visible with equivalent clarity in sparse TOF and PI-TOF. The measured neck height and width of aneurysms were not significantly different when either method was used. Most recently, the same group studied the reliability of CS-TOF in the evaluation of Moyamoya FIG 3. The degree of each luminal stenosis measured on CS-TOF and disease.13 The group found that CS-TOF could improve the visu- PI-TOF, respectively. alization of small collaterals in the same amount of time or pro- duce the same results in a shorter acquisition time compared with difference from that measured on PI-TOF (58.9% Ϯ 31.0%) (P ϭ PI-TOF. In our study, variable degrees of intracranial arterial ste- .241). Table 3 shows detailed comparison results between CS- nosis (from mild stenosis to occlusion) were included. An accel- TOF and PI-TOF for the evaluation of intracranial arterial steno- eration factor of 10.3 for CS-TOF significantly shortened the ac- sis. Representative cases are shown in Figs 1 and 2. The degree of quisition time, while providing comparable image quality and each luminal stenosis measured on CS-TOF and PI-TOF is shown more clear visualization of arteries than PI-TOF. The possibly in Fig 3. stenosed segments recorded on PI-TOF included the intracranial 1836 Lu Oct 2018 www.ajnr.org ICA and proximal M2 segment of the MCA, of which 2 segments would confer to obtain a consensus. However, such a dilemma were considered artifactual narrowing caused by signal loss due to was very rare in our study (only 2 segments). Second, the in-plane turbulent and slow flow according to previous studies.27 The di- resolution between PI-TOF and CS-TOF was identical (0.4 ϫ 0.4 agnostic confidence of stenosis in these segments was improved mm), while the slice thickness was not. Keeping the same voxel on CS-TOF. The luminal stenosis ratio based on PI-TOF and size (0.4 ϫ 0.4 ϫ 0.4 mm) as CS-TOF would lead to a scan time of CS-TOF was not significantly different; this finding suggests the around 11 minutes for PI-TOF. We therefore increased the slice absence of either over- or underestimation of the luminal stenosis thickness while keeping the slab thickness constant in the PI-TOF grade due to irregular undersampling with CS-TOF. protocols to achieve an acceptable acquisition time of around 5 We found that CS-TOF provided higher edge sharpness of the minutes. Third, the number of iterations was fixed at 10 in the intracranial arteries than PI-TOF and enhanced the contrast of current study, consistent with a previous report by Yamamoto et the vessels. In previous studies, the apparent contrast-to-back- al.24 Fushimi et al11 reported that CS-TOF with 10 iterations ground deviation of the sparse TOF images was reported to be could provide adequate image quality for the clinical diagnosis of stable at various acceleration factors and was significantly higher cerebral aneurysms. Vessel sharpness and the visibility of small than with PI-TOF.11,26 Our results were consistent with these vessels can be improved with an increasing number of itera- findings. The high apparent contrast-to-background deviations tions,14 resulting in a largely extended reconstruction time. The in CS-TOF may contribute to the better visualization of stenosis reconstruction time would be 4 minutes 50 seconds if the itera- in segments of the intracranial ICA and proximal M2 segment of tions were increased to 20 in our study, which may affect the the MCA, as well as small branches, which are more prone to clinical examination workflow. turbulent or slow flow. Curved stripe pattern artifacts associated with sparse under- CONCLUSIONS sampling could be observed on all the source images of CS-TOF. The image quality of highly accelerated CS-TOF is comparable Such artifacts are considered ghost artifacts originating from with that of PI-TOF, while CS-TOF has the obvious benefit of the skull boundaries.13 However, these artifacts are hardly notice- reducing imaging time and boosts confidence in diagnosing in- able on the MIP images of CS-TOF because the MIP connects the tracranial arterial stenosis. high-intensity dots of the blood vessels in 3D. Each point in the MIP represents the highest intensity experienced in that location ACKNOWLEDGMENTS on any partition within the imaging volume. Besides, the artifacts We sincerely thank Long-quan Dai from Nanjing University of introduce an additional modulation of the already inhomoge- Sciences and Technology for his kind and warm help during the neous background signal. The additional modulation is spatially analysis of edge sharpness of the intracranial arteries. We also slowly varying and of small amplitude compared with the normal thank Min-lin Zhou from the National Clinical Research Center signal variations in the background. Therefore, the performance of Kidney Diseases, Jinling Hospital, for reviewing all the of vessel segmentation algorithms performing well on the artifact- statistics. free image should not be degraded significantly by the artifacts. Disclosures: Christoph Forman—UNRELATED: Employment: Siemens. Peter Speier— Moreover, the speckled noise on conventional PI-TOF in the cen- UNRELATED: Employment: Siemens, Michaela Schmidt—UNRELATED: Employment: tral parts of the source images was reduced in CS-TOF. The total Siemens. Comments: full-time employee; Stock/Stock Options: Siemens. reconstruction time after scanning was approximately 1 minute 40 seconds for CS-TOF because we used a graphic processing unit REFERENCES in the current study, shortened to clinically acceptable times, 1. Writing Group Members, Mozaffarian D, Benjamin EJ, Go AS, et al; compared with 10 minutes in a previous report.13 American Heart Association Statistics Committee, Stroke Statistics There are several limitations to our study. First, DSA is con- Subcommittee. Heart disease and stroke statistics: 2016 update—a report from the American Heart Association. Circulation 2016;133: sidered the criterion standard. However, only 4 patients had DSA e38–360 CrossRef Medline within 4 months before or after the MRA examination. DSA is 2. Arenillas JF. Intracranial atherosclerosis: current concepts. Stroke invasive, and when MRA can depict reliable arterial stenosis, DSA 2011;42:S20–23 CrossRef Medline is not considered essential in clinical practice. Previous studies 3. Mazighi M, Labreuche J, Gongora-Rivera F, et al. Autopsy prevalence of intracranial atherosclerosis in patients with fatal stroke. Stroke have verified a good correlation between TOF-MRA and DSA for 2008;39:1142–47 CrossRef Medline detecting arterial stenosis, though the stenosis ratio may be 4. Bui TD, Gelfand D, Whipple S, et al. Comparison of CT and catheter 28 overestimated on TOF-MRA. Patients with mild (Ͻ50%) or arteriography for evaluation of peripheral arterial disease. Vasc En- moderate intracranial arterial stenosis (50%–69%) account dovascular Surg 2005;39:481–90 CrossRef Medline for a large proportion of all those with cerebrovascular dis- 5. Schernthaner R, Stadler A, Lomoschitz F, et al. Multidetector CT angiography in the assessment of peripheral arterial occlusive eases. However, such patients usually do not undergo DSA disease: accuracy in detecting the severity, number, and length of examinations. Because TOF-MRA is a good noninvasive screen- stenoses. Eur Radiol 2008;18:665–71 CrossRef Medline ing tool, we considered that including patients with different ste- 6. Patel MR, Klufas RA, Kim D, et al. MR angiography of the carotid nosis ratios varying from mild stenosis to occlusion in our study bifurcation: artifacts and limitations. AJR Am J Roentgenol 1994; would be more clinically significant. The evaluation of all the 162:1431–37 CrossRef Medline 7. Kaufmann TJ, Huston J 3rd, Cloft HJ, et al. A prospective trial of 3T images was performed by 3 experienced neuroradiologists to and 1.5T time-of-flight and contrast-enhanced MR angiography in avoid any false-positive or -negative judgment. In the case of ste- the follow-up of coiled intracranial aneurysms. AJNR Am J Neuro- nosis seen only on 1 of the TOF sequences, the 3 neuroradiologists radiol 2010;31:912–18 CrossRef Medline AJNR Am J Neuroradiol 39:1833–38 Oct 2018 www.ajnr.org 1837 8. Griswold MA, Jakob PM, Heidemann RM, et al. Generalized auto- measuring intracranial arterial stenosis. AJNR Am J Neuroradiol calibrating partially parallel acquisitions (GRAPPA). Magn Reson 2000;21:643–46 Medline Med 2002;47:1202–10 CrossRef Medline 19. Feichtenhofer CF, Fassold H, Schallauer P. A perceptual image 9. Bhagat YA, Emery DJ, Naik S, et al. Comparison of generalized au- sharpness metric based on local edge gradient analysis. IEEE Signal tocalibrating partially parallel acquisitions and modified sensitiv- Process Lett 2013;20:379–82 CrossRef ity encoding for diffusion tensor imaging. AJNR Am J Neuroradiol 20. Fujita N, Hirabuki N, Fujii K, et al. MR imaging of middle cerebral 2007;28:293–98 Medline artery stenosis and occlusion: value of MR angiography. AJNR Am J 10. Preibisch C, Wallenhorst T, Heidemann R, et al. Comparison of par- Neuroradiol 1994;15:335–41 Medline allel acquisition techniques generalized autocalibrating partially 21. Oelerich M, Lentschig MG, Zunker P, et al. Intracranial vascular parallel acquisitions (GRAPPA) and modified sensitivity encoding stenosis and occlusion: comparison of 3D time-of-flight and 3D (mSENSE) in functional MRI (fMRI) at 3T. J Magn Reson Imaging phase-contrast MR angiography. Neuroradiology 1998;40:567–73 2008;27:590–98 CrossRef Medline CrossRef Medline 11. Fushimi Y, Fujimoto K, Okada T, et al. Compressed sensing 3-di- 22. Konar AS, Aiholli S, Shashikala HC, et al. Application of region of mensional time-of-flight magnetic resonance angiography for ce- interest compressed sensing to accelerate magnetic resonance an- rebral aneurysms: optimization and evaluation. Invest Radiol 2016; giography. Conf Proc IEEE Eng Med Biol Soc 2014;2014:2428–31 51:228–35 CrossRef Medline CrossRef Medline 12. Lustig M, Donoho D, Pauly JM. Sparse MRI: the application of com- 23. Rapacchi S, Han F, Natsuaki Y, et al. High spatial and temporal res- pressed sensing for rapid MR imaging. Magn Reson Med 2007;58: olution dynamic contrast-enhanced magnetic resonance angiogra- 1182–95 CrossRef Medline phy using compressed sensing with magnitude image subtraction. 13. Yamamoto T, Okada T, Fushimi Y, et al. Magnetic resonance angiog- Magn Reson Med 2014;71:1771–83 CrossRef Medline raphy with compressed sensing: an evaluation of moyamoya dis- 24. Yamamoto T, Fujimoto K, Okada T, et al. Time-of-flight magnetic ease. PLoS One 2018;13:e0189493 CrossRef Medline resonance angiography with sparse undersampling and iterative 14. Stalder AF, Schmidt M, Quick HH, et al. Highly undersampled con- trast-enhanced MRA with iterative reconstruction: integration in a reconstruction: comparison with conventional parallel imaging for clinical setting. Magn Reson Med 2015;74:1652–60 CrossRef Medline accelerated imaging. Invest Radiol 2016;51:372–78 CrossRef Medline 15. Akc¸akaya M, Rayatzadeh H, Basha TA, et al. Accelerated late gado- 25. Hutter J, Grimm R, Forman C, et al. Highly undersampled periph- linium enhancement cardiac MR imaging with isotropic spatial res- eral time-of-flight magnetic resonance angiography: optimized olution using compressed sensing: initial experience. Radiology data acquisition and iterative image reconstruction. MAGMA 2015; 2012;264:691–99 CrossRef Medline 28:437–46 CrossRef Medline 16. Chandarana H, Block TK, Ream J, et al. Estimating liver perfusion 26. Fushimi Y, Okada T, Kikuchi T, et al. Clinical evaluation of time-of- from free-breathing continuously acquired dynamic gadolinium- flight MR angiography with sparse undersampling and iterative re- ethoxybenzyl-diethylenetriamine pentaacetic acid-enhanced ac- construction for cerebral aneurysms. NMR Biomed 2017;30 quisition with compressed sensing reconstruction. Invest Radiol CrossRef Medline 2015;50:88–94 CrossRef Medline 27. Anderson CM, Saloner D, Tsuruda JS, et al. Artifacts in maximum- 17. Vasanawala SS, Alley MT, Hargreaves BA, et al. Improved pediatric intensity-projection display of MR angiograms. AJR Am J Roent- MR imaging with compressed sensing. Radiology 2010;256:607–16 genol 1990;154:623–29 CrossRef Medline CrossRef Medline 28. Wilms G, Bosmans H, Demaerel P, et al. Magnetic resonance angiog- 18. Samuels OB, Joseph GJ, Lynn MJ, et al. A standardized method for raphy of the intracranial vessels. Eur J Radiol 2001;38:10–8

1838 Lu Oct 2018 www.ajnr.org ORIGINAL RESEARCH ADULT BRAIN

Cerebrovascular Reactivity during Prolonged Breath-Hold in Experienced Freedivers

X V.C. Keil, X L. Eichhorn, X H.J.M.M. Mutsaerts, X F. Tra¨ber, X W. Block, X B. Ma¨dler, X K. van de Ven, X J.C.W. Siero, X B.J. MacIntosh, X J. Petr, X R. Fimmers, X H.H. Schild, and X E. Hattingen

ABSTRACT BACKGROUND AND PURPOSE: Experienced freedivers can endure prolonged breath-holds despite severe hypoxemia and are therefore ideal subjects to study apnea-induced cerebrovascular reactivity. This multiparametric study investigated CBF, the spatial coefficient of variation as a correlate of arterial transit time and brain metabolism, dynamics during prolonged apnea.

MATERIALS AND METHODS: Fifteen male freedivers (age range, 20–64 years; cumulative previous prolonged breath-holds Ͼ2 minutes and 30 seconds: 4–79,200) underwent repetitive 3T pseudocontinuous arterial spin-labeling and 31P-/1H-MR spectroscopy before, during, and after a 5-minute breath-hold (split into early and late phases) and gave temporally matching venous blood gas samples. Correlation of temporal and regional cerebrovascular reactivity to blood gases and cumulative previous breath-holds of Ͼ2 minutes and 30 seconds in a lifetime was assessed.

RESULTS: The spatial coefficient of variation of CBF (by arterial spin-labeling) decreased during the early breath-hold phase (Ϫ30.0%, P ϭ .002), whereas CBF remained almost stable during this phase and increased in the late phase (ϩ51.8%, P ϭ .001). CBF differed between the anterior and the posterior circulation during all phases (eg, during late breath-hold: MCA, 57.3 Ϯ 14.2 versus posterior cerebral artery, 42.7 Ϯ 10.8 mL/100 g/min; P ϭ .001). There was an association between breath-hold experience and lower CBF (1000 previous breath-holds reduced WM CBF by 0.6 mL/100 g/min; 95% CI, 0.15–1.1 mL/100 g/min; P ϭ .01). While breath-hold caused peripheral lactate rise (ϩ18.5%) and hypoxemia (oxygen saturation, Ϫ24.0%), cerebral lactate and adenosine diphosphate remained within physiologic ranges despite early signs of oxidative stress [Ϫ6.4% phosphocreatine / (adenosine triphosphate ϩ adenosine diphosphate); P ϭ .02].

CONCLUSIONS: This study revealed that the cerebral energy metabolism of trained freedivers withstands severe hypoxic hypercarbia in prolonged breath-hold due to a complex cerebrovascular hemodynamic response.

ABBREVIATIONS: ATP ϭ adenosine triphosphate; ASL ϭ arterial spin-labeling; ASL-sCoV ϭ spatial coefficient of variation of CBF (by ASL); ATT ϭ arterial transit ϭ ϭ ϭ ϭ ϭ ϭ time; CVR cerebrovascular reactivity; HR heart rate; Pi inorganic phosphates; PCr phosphocreatine; pO2 partial pressure of oxygen; SpO2 oxygen saturation

reedivers acquire the ability to voluntarily breath-hold for sev- regulation and sufficiency of cerebrovascular reactivity (CVR) to Feral minutes: One breath-hold can be extended up to the world prevent ischemic brain damage during prolonged breath-hold are record of 11 minutes and 30 seconds.1 However, animal studies unclear. Such knowledge may hold clues to explain patterns of have suggested that threshold times for hypoxia-induced neuro- brain damage in adverse diving outcomes and may provide sug- nal cell death can be as low as 2 minutes and 30 seconds.2 The gestions for targeted therapies. It can also be useful to deduce CVR in medical conditions of repetitive hypoxemia such as obstructive sleep apnea syndrome.3 Received December 29, 2017; accepted after revision July 19, 2018. From the Departments of Radiology (V.C.K., F.T., W.B., H.H.S., E.H.) and Anesthesiol- Previous MR imaging breath-hold experiments involved hy- ogy and Intensive Care Medicine (L.E.), and Institut für Medizinische Biometrie, Informatik und Epidemiologie (R.F.), University Hospital Bonn, Bonn, Germany; De- partment of Radiology (H.J.M.M.M.), Academic Medical Center, Amsterdam, the Paper previously presented, in part, at: 99th German Röntgen Congress, May 27, Netherlands; Sunnybrook Research Institute (H.J.M.M.M., B.J.M.), University of To- 2017 (No. 4735), Leipzig, Germany; and Joint Annual Meeting of the International ronto, Toronto, Ontario, Canada; Department of Radiology (H.J.M.M.M., J.C.W.S.), Society for Magnetic Resonance in Medicine and the European Society for Mag- University Medical Center Utrecht, Utrecht, the Netherlands; Philips GmbH (B.M), netic Resonance in Medicine and Biology, June 16–21, 2018; Paris, France. Bonn, Germany; Philips Healthcare (K.v.d.V.), Best, the Netherlands; Spinoza Centre Please address correspondence to Vera Catharina Keil, MD. Department of Radiol- for Neuroimaging (J.C.W.S.), Amsterdam, the Netherlands; and Helmholtz Center ogy, University Hospital Bonn, Sigmund-Freud-Str 25, 53127 Bonn, Germany; e-mail: Dresden-Rossendorf, Institute for Radiopharmaceutic Cancer Research (J.P.), PET [email protected] Center, Dresden-Rossendorf, Germany. Indicates article with supplemental on-line table. Vera C. Keil, Lars Eichorn, and Henri J.M.M. Mutsaerts contributed equally to this work. http://dx.doi.org/10.3174/ajnr.A5790

AJNR Am J Neuroradiol 39:1839–47 Oct 2018 www.ajnr.org 1839 Anthropometric data of freediver volunteersa Longest Start of Estimated Estimated Breath-Hold Sport (yr Breath-Holds Sessions Cumulative Lifetime Age Participant until MRI (min) before MRI) >2 min 30 s per Session per Month Breath-Holds >2 min 30 s (yr) 1 10.6 11.3 30 0.4 1632 31 2 5.5 1.5 7 8 1008 36 3 5.4 2.0 1 1 24 20 4 5.4 11.0 ϳ08 433 5 5.0 2.0 0.125 8 24 40 6 5.5 10.0 ϳ04 438 7 5.0 8.0 5 8 3840 61 8 5.5 1.5 0.5 4 36 34 9 5.0 7.0 6 16 8064 47 10 6.2 9.0 6 7 4536 26 11 5.3 27.5 15 16 79,200 64 12 6.5 10.2 12 28 11,712 51 13 5.5 11.3 10 16 21,760 35 14 5.5 1.6 18 10 3420 32 15 5.0 5.0 0.083 4 20 50 Median 5.5 8.0 6 8 2526 36.0 95% CI 5.3–5.5 2.0–11.0 0.14–12.0 4.0–9.96 30–6300 32–50 Minimum 5.0 1.5 0 0.41 4 20 Maximum 10.6 27.5 30 28 79,200 64 a Participants were interviewed concerning the duration (start of sport) and intensity (freediving sessions per month) and the estimated frequency of breath-holding for Ͼ2 minutes and 30 seconds per session. The latter could be near zero in freedivers specializing in short high-frequency breath-holds. Some participants breath-held for Ͼ2 minutes and 30 seconds only as a qualification for this study; hence, their cumulative lifetime breath-holds of Ͼ2 minutes and 30 seconds before MRI were 4, while others very frequently underwent prolonged breath-hold. percarbic air composition or the use of relatively short breath- The goal of the present study in experienced freedivers was to holds.4 There are, however, crucial differences between experi- explore the following: 1) acute CVR during a prolonged breath- mental setups applying altered gas composition and those using hold challenge with ASL, and 2) the sufficiency of CVR to main- voluntary breath-hold. In the latter, the human body can be ex- tain aerobic cerebral energy metabolism as measured by 31P- and amined as a closed system of internal metabolic gas exchange, 1H-MR spectroscopy. In addition, this study investigated whether while altered-ventilation gas studies are an “open system” ap- pre-existing experience with prolonged breath-holds has an influ- proach with ventilation at variable frequencies as a complicating ence on the CVR during the breath-hold experiment. factor.5 Quite different from short breath-holds, prolonged breath-hold for several minutes is furthermore a unique mental MATERIALS AND METHODS challenge demanding an emotional preparation, for which Participants freedivers need to train and which may have an independent effect Fifteen experienced male freedivers (median age, 36.0 years; on CVR.6 The putative effect of experience from previous pro- 95% CI, 32.0–50.0 years) participated (Table). Inclusion longed breath-holds on CBF must be mentioned in this context criteria were adult age and an ability to breath-hold for Ͼ4 and has not been explored. minutes without stress symptoms (ie, tachycardia, oxygen sat-

Arterial spin-labeling (ASL) perfusion MR imaging allows an uration [SpO2] below 60%, delirium). Exclusion criteria were in vivo assessment of absolute CBF both regionally and on a ves- pre-existing cardiac or neurologic disorders and current smok- sel-selective level. ASL studies already successfully identified ing. The participant’s age and cumulative number of breath- chronic CVR alterations due to obstructive sleep apnea as well as holds longer than 2 minutes and 30 seconds during the volun- acute alterations during very short breath-holds.5,7 A corollary teer’s lifetime (estimated by interview report) were registered ASL measure is the spatial coefficient of variation (ASL-sCoV) of as potential influential factors. the CBF image itself, which was recently identified as a correlate for arterial transit time (ATT) and is a useful alternative when Study Design direct ATT estimations are not feasible.8 Thus, ASL seems a Preparatory evaluations involved questionnaires regarding claus- promising technique to study CVR in prolonged breath-hold. trophobia, noise tolerance. or issues with restraints. All partici- Blood gas alterations trigger a cerebrovascular but also a pe- pants were instructed to refrain from meals and caffeine for at ripheral vascular reaction in the body. During breath-hold, for least 2 hours before MR imaging to reduce perfusion confound- 10 example, the CBF increases and hence faster O2 supply facilitates ers. Preparation (eg, meditation, test placement on the MR im- the maintenance of an aerobic brain metabolism under hypox- aging table) was allowed. MR imaging–electrocardiography and emia—that is, to prevent the depletion of adenosine triphosphate finger oximetry guaranteed continuous monitoring of SpO2 and

(ATP) and the accumulation of lactate from anaerobic glycolysis. heart rate (HR). Arterial CO2 measurements were not permitted The sufficiency of CVR can be studied indirectly by 31P-MR spec- by the hospital ethics committee, but venous blood gas was repet- troscopy, which provides estimates of ATP and related metabo- itively analyzed. Continuity of breath-hold and consciousness lites,9 as well as by 1H-MR spectroscopy, which may show lactate were visually monitored (L.E.) to guarantee safety and correct if the metabolite is elevated beyond physiologic ranges. measurements. 1840 Keil Oct 2018 www.ajnr.org The MR imaging protocol consisted of a 3D T1-weighted ac- ␴͑CBF ͒ ϭ GM ϫ ASL-sCoVROI ␮͑ ͒ 100%. quisition followed by 5 ASL scans. Participants were instructed to CBFGM breathe with normal frequency to obtain the baseline CBF (base- Partial volume effects within the GM and WM ROIs were cor- line phase). Hyperventilation to reduce blood CO2 or increase O2 15 concentrations was a forbidden confounder. The participants rected using the method of Asllani et al. gave an acoustic signal when starting the 5-minute breath-hold, when 2 consecutive ASL scans were acquired (early and late MR Spectroscopy Proton-decoupled 31P spectra were acquired with a dual-tunable breath-hold phases). After the second breath-hold scan, the par- 31P/1H birdcage transmit/receive head coil (1024 data points; ticipant was instructed to breathe again at normal frequency (ap- 3-kHz sampling; TR, 4 seconds; 4 signal averages) as a continuous proximately 16/min.). Without delay, 2 normal-breathing ASL time-series with 16-second duration for each spectrum, compris- scans (early and late recovery phases) were obtained. A subset of ing a total of 40 scans before, during, and directly after breath- participants was available for 31P- and 1H-MR spectroscopy (n ϭ hold in a 25-mm-thick axial slice on the level of the basal ganglia. 11 and 8 participants, respectively) during separate sessions to 31P signals were processed by the AMARES algorithm of the determine relative brain metabolites and, again, venous blood JAVA-MRUI software (http://www.jmrui.eu/features/quantitation/) gas. The study was approved by the University Hospital Bonn, quantifying 15 peaks in each spectrum arising from phosphocre- Germany ethics committee. All participants provided written in- atine (PCr), inorganic phosphates (P ), ATP, and phosphomono- formed consent. i and diesters.16 The cerebral pH was determined from the fre- quency separation between P and PCr. Five consecutive spectra MR Imaging Sequences i sets within the 31P time-series were averaged for 2 minutes and 30 All imaging was performed on a 3T Ingenia MR imaging scanner seconds, which represent the metabolic baseline, first and second (Philips Healthcare, Best, the Netherlands). The T1-weighted halves of breath-hold, as well as early and late metabolic recovery MPRAGE sequence was 1 ϫ 1 ϫ 1mm3. The 5 identical pseudo- phases. This step was also to improve the signal-to-noise ratio. continuous ASL sequences were acquired with a 3D gradient- and Thus, the time course of pH and of the ratios P /PCr, P /␤-ATP, spin-echo readout (5 segments; acquisition voxel size, 3.75 ϫ i i PCr/␤-ATP, as well as the ratios of all 31P metabolites relative to 3.75 ϫ 6mm3; FOV, 240 ϫ 240 ϫ 96 mm3; TE/TR, 8/4.28 s; (ATP ϩ adenosine diphosphate), calculated from the mean of the labeling duration, 1.8 s; radiofrequency labeling pulse duration/ ␥- and ␣-ATP peaks, were obtained in the same 2 minutes and 30 interval, 0.7/1.41 ms; postlabeling delay, 2 s; 4 background sup- seconds intervals as the ASL sequences. ␥- and ␣-ATP peaks in- pression pulses; scan time, 2 minutes and 30 seconds per scan clude the signals from nucleotide diphosphates and thus are ex- phase). Each sequence consisted of 2 M0 images and 2 control- pected as near-constant. label pairs. A labeling distance of 13 cm (middle slice of the ASL In a consecutive single-volume 1H-MR spectroscopy breath- stack to labeling plane) was chosen. hold experiment, ratios of the 1H-MR signals of N-acetylaspar- tate, total creatine, choline, and lactate were determined from an Image Processing 8-mL volume in the left basal ganglia (point-resolved spectros- Image processing was performed with ExploreASL (www.Ex- copy sequence localized spectra: TR 2 s; TE 0.14 s; 128 signal ploreASL.com), a toolbox based on SPM12 routines (http:// averages). Recording periods were split to match the ASL-MR www.fil.ion.ucl.ac.uk/spm/software/spm12), which was initiated imaging and 31P-MR spectroscopy phases of the breath-hold through the European Union–funded European Cooperation in experiment. Science and Technology (COST) Action ASL In Dementia (http://s434060124.online.de/aslindementiacms/), aiming at Blood Gas Analysis harmonizing ASL image processing for single- and multicenter We drew 10 mL of venous blood before breath-hold, after 2 min- 11 ASL studies. Image-processing steps were the following: auto- utes and 30 seconds of breath-hold, at first breath after breath- mated segmentation of 3D T1-weighted images using the Com- hold, and after 2 minutes and 30 seconds of recovery. Immediate putational Anatomy Toolbox (CAT12 toolbox) rigid-body regis- analysis (RAPIDLab 1265; Siemens, Erlangen, Germany) in- tration of CBF to the gray matter partial volume map and spatial volved partial pressure of CO2 and oxygen (pO2), glucose, and normalization into a common space using the Diffeomorphic An- lactate levels. With the exception of partial pressure of CO2, these atomical Registration Through Exponentiated Lie Algebra Tool- parameters were shown to correlate well with arterial values.17 box (DARTEL, part of SPM).12 M0 images were masked, itera- tively smoothed, and extrapolated outside the mask, and CBF was Statistics quantified using a single-compartment quantification model.13 Overall CBF variability was analyzed by a mixed linear model with The ROIs analyzed were the total (cortical) GM and the total the participant as a random factor, while differences between sin- white matter. The GM was subsegmented into vascular territories gle time points were analyzed by a paired t test, which considers ⌬ of the anterior, middle, and posterior cerebral arteries. The vas- CBF differences compared with baseline ( CBF/CBF0)asthe cular territories were delineated in the common space on the expression of CVR without division by a fixed rate of stimulus 14 18 Montreal Neurological Institute atlas according to Tatu et al. such as per unit CO2. The same tests were applied for the partial The ASL-sCoV was defined as the SD of CBF divided by the volume–corrected ASL-sCoV. MR spectroscopy parameters were mean CBF, within the total cortical GM (gray matter probabil- analyzed in a mixed linear model and with the Pearson R correla- ity Ͼ 0.7)8: tion. The GM CBF in the vascular territories was analyzed sepa- AJNR Am J Neuroradiol 39:1839–47 Oct 2018 www.ajnr.org 1841 phase CBF was indeed lower than baseline CBF (mean, Ϫ22.4%; P ϭ .01). Baseline-to-late recovery phase CBF differences were not significant (P ϭ .55 for the total GM ROI). Return to baseline CBF or below occurred during the early recovery phase with a mild secondary CBF increase during the second 2 minutes and 30 seconds of recovery (P ϭ .03 in total GM ROI). The absolute CBF and CVR of the anterior circulation (ante- rior cerebral artery, MCA) were, at all times, higher than in the posterior circulation (P ϭ .001; On-line Table). The mean differ- ence between CBF of the MCA and the posterior cerebral artery increased steadily during breath-hold from 8.8 Ϯ 6.6 mL/100 g/min at baseline to 14.6 Ϯ 6.1 mL/100 g/min at late breath-hold (P ϭ .001). During all scan phases, between-participant CBF variability showed a narrower range in the GM than in the WM as well as a lower overall CVR (CVR variability in GM: 83.8%–152.8% from baseline; CVR in WM: 74.2%–231.5% increase from baseline CBF; P ϭ .001; On-line Table).

Spatial Coefficient of Variation The ASL-sCoV varied over the course of the experiment (P ϭ .001). In most cases (n ϭ 12/15), there was an ASL-sCoV decrease between baseline and the early breath-hold phase (mean decrease: Ϫ30.0% Ϯ 21.6%; P ϭ .002; Fig 2C). The ASL-sCoV remained reduced during the 5-minute breath-hold (P ϭ .81 for the differ- ence between early and late breath-hold) and rose again during recovery (P ϭ .01). There was no difference between the baseline FIG 1. Mean CBF during the breath-hold experiment. False color maps and recovery phases (P ϭ .29). of cohort mean cerebral blood flow during 5-minute breath-hold. The range of GM ASL-sCoV among participants was smaller Selected transverse brain sections of the cohort mean CBF before, Ϯ during, and after the 5-minute breath-hold challenge. The mean total during breath-hold than during normal breathing: mean, 2.8% GM CBF over all participants and over all phases was scaled to 60 1.0%; range, 1.9%–5.3% versus mean, 1.8% Ϯ 0.4%; range, mL/100 g/min. Likewise, the mean total WM CBF for all participants 1.5%–3.0%, for baseline versus early breath-hold phases, respec- over all phases was scaled to 20 mL/100 g/min for WM. tively. This range equates to a 35.7% decrease of ASL-sCoV vari- ability among participants (P ϭ .02). rately by a paired t test to identify differences between the anterior and posterior circulation. The influence of age and experience Physiologic Correlations with prolonged breath-holds estimated from cumulative breath- The dynamics of SpO2 and HR are presented in the On-line Table. hold events of Ͼ2 minutes and 30 seconds was tested for correla- SpO2 correlated with CVR, with an estimated increase of CBF of tion with CBF and ASL-sCoV using a mixed linear model with the 0.82 mL/100 g/min with each 1% SpO drop (95% CI, 0.6–1.1 participant as a random factor. Similarly, the relationships of the 2 mL/100 g/min; P ϭ .001; Fig 3A,-B). HR was not correlated with ASL parameters with SpO and HR were analyzed. Parametric 2 CVR (P ϭ .36; Fig-D). 3 Similarly,C, age was not identified as an testing occurred after testing for normal distribution. Generally, influential cofactor on CVR (P ϭ .32). absolute CBF differences (⌬ in mL/100 g/min) were used for de- If one took the entire group of 15 freedivers into account, scriptive statistics and statistical analyses, while ratios (in percent- previous prolonged breath-hold events were not correlated with ages) are mentioned for illustrative purposes. absolute CBF values (P ϭ .56). However, there were 2 outlier RESULTS participants (divers 11 and 13) with exceptionally extensive expe- Cerebrovascular Reactivity rience and comparatively high CBF. After their exclusion, a rela- All participants showed a significant CVR by an increase in CBF tionship between previous breath-hold experience and absolute until 5 minutes of breath-hold with a subsequent decline at recov- CBF was found with 1000 previous prolonged breath-holds, re- ery (P ϭ .001 for all vessel territories, GM, and WM; On-line ducing CBF in GM by 2.2 mL/100 g/min (95% CI, 0.7–3.7 mL/100 Table and Figs 1 and 2A,-B). While the CBF increase from base- g/min.; P ϭ .01; Fig 4) for the remaining 13 cases. This finding was line was substantial after 5 minutes (late breath-hold scan phase similar for CBF in WM: One thousand previous breath-holds re- for the total GM ROI: mean ⌬ CBF, 18.3 Ϯ 14.4 mL/100 g/min duced WM CBF by 0.6 mL/100 g/min; 95% CI, 0.15–1.1 mL/100 [ϩ51.8%]; P ϭ .001), it was only subtle within the first 2.5 min- g/min (P ϭ .01). Mean CVR, however, as defined by the ⌬ CBF/ ⌬ Ϯ utes of breath-hold (total GM ROI: mean CBF, 6.3 11.1 mL/ CBF0, was not correlated with previous experience with pro- 100 g/min [ϩ17.8%]; P ϭ .04). In 4 volunteers, early breath-hold- longed breath-holds (P ϭ .23). Similarly, ASL-sCoV was not cor- 1842 Keil Oct 2018 www.ajnr.org FIG 2. Cerebral blood flow and partial volume–corrected relative spatial coefficient of variation dynamics. Baseline scan with normal ventila- tion starts at 0 minutes, the second and third measurement points represent 2 minutes and 30 seconds and 5 minutes of breath-hold, and the fourth and fifth scan phases at 7 minutes and 30 seconds and 10 minutes represent the recovery phase. A, CBF in gray matter voxels only (n ϭ 15 participants, each represented by a color dot). B, The same for white matter only. C, Partial volume–corrected relative spatial coefficient of variation dynamics in gray matter. Y-axis: relative CBF spatial coefficient of variation defined as the ratio of the actual spatial coefficient of variation divided by the spatial coefficient of variation expected on the basis of anatomy (in arbitrary units [au]).

related with SpO2, HR, age, or diver experience with prolonged hold. Finally, this study identified indicators for an influence of breath-holds (P ϭ .45, .53, .90, and .69, respectively). earlier experience with prolonged breath-hold on absolute CBF during the breath-hold challenge but not on CVR itself. Physio- MR Spectroscopy and Blood Analyses logic responses can apparently withstand the extreme biochemi- 31 P-MR spectroscopy (Fig 5A) revealed minor fluctuations in cal challenge induced by a prolonged breath-hold of 5 minutes, pH (Fig 5B) and ATP metabolites within physiologic ranges which can be detected by ASL and MR spectroscopy. ϭ Ͼ during the course of breath-hold (P .07 for pH; P .05 for Multiple methods exist to assess CBF, including ASL, phase- ␣ ␤ ␥ dynamics in relative PCr; Pi; ATP- ,- ,- ; and phospho- contrast MR imaging, PET, and Doppler sonography. Maximum mono- and diester levels). There was a small-but-significant breath-hold studies are rare, and CBF evaluations were, until ϩ decrease of the PCr / (ATP adenosine diphosphate) ratio now, exclusively performed with Doppler sonography, which re- between baseline and late-phase breath-hold in the matched- vealed a continuous elevation of flow velocity in the MCA of pairs analysis of the individuals (Ϫ6.4%; P ϭ .02; Fig 5C). PCr 19,20 around 100%. The mean maximum CBF increase after 5 min- and ␤-ATP differences relative to their baseline values (⌬PCr utes of breath-hold observed in this ASL study was 51.8%, which and ⌬␤-ATP) were correlated with the differences in pH from is very close to values observed in ASL studies using hypercarbic baseline (R ϭ 0.53; P Ͻ .001 for ⌬PCr and R ϭ 0.45; P ϭ .003 gas inhalation or short breath-hold, but indeed in some cases for ⌬␤-ATP, respectively; Fig 5D). lower than values measured with ASL in a recent maximum 1H-MR spectroscopy never showed the CH doublet of lactate 3 breath-hold study (ϩ107%).5,21,22 Previous studies comparing at 1.34 ppm in any of the participants. No significant changes in CBF assessments with different methods similarly revealed sub- the levels of N-acetylaspartate, total creatine, or choline occurred stantial intermethod differences in absolute CBF but otherwise (P Ͼ .05 for all). confirmed a high correlation between methods. PET-estimated Venous blood samples showed a development of hypoxemia absolute CBF was, for example, consistently lower than phase- and hypercapnia during breath-hold (On-line Table, Fig 5E). A significant partial pressure of CO increase was noted only at late contrast MR imaging, while Doppler and ASL differed substan- 2 23,24 ϭ tially in relative CBF change in a drug-stimulation trial. These breath-hold (P .02), while pO2 had already dropped signifi- cantly after the early breath-hold phase (P ϭ .002). Both param- differences are not surprising and can be explained by different eters returned to baseline after breath-hold, while venous lactate influential factors acting on the respective flow parameters. While and glucose levels increased until the end of the experiment dur- ASL can measure absolute CBF, Doppler provides flow velocity in ing the recovery phase (P ϭ .001 and .01, respectively; Fig 5E). a local vessel segment as a surrogate parameter for CBF. CO2 is a strong vasodilating agent in cerebral tissue causing an increase in DISCUSSION CBF due to increased blood volume (CBV) based on the equation This study provides 3 key findings. First, despite individual vari- CBF ϭ CBV / MTT. The blood flow velocity rises despite vessel ability, freedivers show a relatively consistent and vessel-territory- dilation also due to reduced ATT. However, earlier Doppler stud- specific CVR during a breath-hold challenge, which is measurable ies revealed a strong neuromuscular response in proximal vessel with ASL. Second, CBF and the ATT correlate ASL-sCoV deliver segments of the anterior circulation, while more distal segments independent aspects of the cerebrovascular response to breath- and the posterior circulation seemed less responsive.25,26 This AJNR Am J Neuroradiol 39:1839–47 Oct 2018 www.ajnr.org 1843 conditions.27,28 A recent Doppler-mon- itored breath-hold study supports our findings of a lower CVR in the posterior cerebral artery and suggests a different sympathetic activation between the 2 territories as an additional explanation beyond the labeling aspect, which needs to be considered in ASL.29 An awareness of a vessel-selective CVR is, however, crucial when interpreting ASL measure- ments in focal ischemic lesions after prolonged clinical conditions of apnea. Another finding of this study is that CBF decreased during the first 2 minutes and 30 seconds of breath-hold in 4 of 15 participants. The counter-suggestive relatively higher CBF at baseline com- pared with later time points might be the effect of anticipation anxiety toward the upcoming breath-hold challenge.30 This well-known mental phenomenon among freedivers is currently not suffi- FIG 3. Relationship of SpO and heart rate with CBF. A, Mean GM CBF (colored dots) and mean 2 ciently addressed in sports physiologic WM CBF (triangles) are both strongly correlated with absolute SpO2. Color encoding for A and C: green for baseline, orange for 2 minutes and 30 seconds breath-hold, red for 5-minute breath- research. Indeed, our own results may hold, turquoise after 2 minutes and 30 seconds of recovery, and blue after 5 minutes of recovery. only suggest that some freedivers expe- The same WM time points are right below. B, ⌬CBF (here measured in GM) similarly correlates well ⌬ rienced an early sympathetic activation to SpO2 (all 75 single values). C and D, There is no strong correlation between heart rate and CBF ⌬ ⌬ in absolute values (C, mean) or CBF and SpO2 (D). before the breath-hold challenge. Due to the limited temporal resolution of the ASL sequence, which delivers a mean CBF over each of the 2 minutes and 30 seconds phases, we cannot readily assess how long the reduction of CBF persists and when exactly an elevation of the CBF above took place during the early breath-hold phase in these 4 volunteers. We further confirmed a difference between the cortical GM and the deep WM CVR, with the total GM CBF known to be at least 2 times higher than the WM CBF.31 Low WM signal is an obstacle for WM CBF assessment in ASL, despite the availability of background suppression.31 However, CVR differences between FIG 4. Scatterplot illustration of the relationship between cumula- GM and WM tissue can be interesting because they may help to tive breath-hold experience and CBF during the experiment. With the exception of 2 participants (divers 11 and 13, marked with arrows), better understand morphologic findings in brain diseases such as there is a lower CBF during the experiment with more previous pro- obstructive sleep apnea syndrome.32 We noted a larger CVR vari- longed breath-hold events (cumulative breath-hold, Ͼ2 minutes and ability and relative increase in WM than in GM, which could be 30 seconds). explained by the later arrival of blood in the relatively more distal may explain why Doppler-assessed breath-hold experiments WM vessels. WM CBF is difficult to measure at baseline due to identify a higher absolute and relative CVR, mostly in the M1 longer ATT. Direct ATT assessment was technically not possible segments of the MCA, than most of those applying ASL, which as part of our experimental setting due to temporal restrictions captures perfusion in most distal and nonmuscular vascular and a resulting mono-post-labeling delay ASL sequence. How- segments. ever, our ATT approximation based on ASL-sCoV confirmed a A heterogeneous CVR between vessels was also observed in the breath-hold-induced decrease in ATT, which might increase the present ASL study with lower CBF and CVR in the posterior ce- SNR of WM CBF, which inflates the measured ⌬CBF to a certain rebral artery territories. This finding can be explained by differ- extent. ences in labeling efficiency between vascular territories as well as For the interpretation of ASL-sCoV, it is viable to consider the the longer ATT in the posterior vascular territory. Microanatomic methodologic peculiarities of ASL. An ATT increase will cause the differences leading to a diverse autoregulation capacity between ASL signal to appear in larger vessels, resulting in vascular arti- the anterior and posterior circulation as an underlying reason for facts. The ASL signal can, at the same time, decrease in areas with a diverse CVR are, on the other hand, more controversially dis- higher baseline ATT as in, for example, the perfusion watershed. cussed even beyond the field of perfusion studies under extreme These 2 effects both increase ASL-sCoV.8 However, ASL-sCoV 1844 Keil Oct 2018 www.ajnr.org FIG 5. Energy metabolites and blood gases in brain and venous blood. A, Sample 31P-MR spectrum of 1 participant averaged over 2 minutes and 30 seconds of the late breath-hold phase (above), displayed together with AMARES-fitted spectral components (below). B, Brain pH as assessed by 31P-MR spectroscopy is near-constant during the entire breath-hold experiment. C, The PCr / (ATP ϩ adenosine diphosphate) ratio measured by 31P-MR spectroscopy slightly decreases during breath-hold. D, PCr and ␤-ATP differences (⌬) from individual baseline values correlate with the pH differences from baseline, indicating that small tendencies toward acidosis and ATP depletion occur during breath-hold. E, Venous blood analyses reveal significant hypoxemia and hypercapnia development (left axis in millimeters of mercury) during breath-hold with fast recovery. Blood glucose (in milligrams/deciliter, left axis) and lactate (in mmol/L, right axis) rose during breath-hold and did not return to baseline. All values are expressed as differences from baseline.

has a theoretic lower limit attained when all labeled blood has freedivers showed that cerebral O2 desaturation tends to occur arrived in the tissue and all vascular artifacts have already disap- within a mean of 175 Ϯ 50 seconds, but not before, which sup- peared. Further ATT decrease beyond this limit will have only ports our finding.34 minimal effect on the ASL-sCoV. Reaching this lower limit of In this study, CBF itself was not strongly correlated to HR. This ASL-sCoV during the early breath-hold could explain the low finding does not allow concluding that CBF in breath-hold is not ASL-sCoV variability in some participants. The ASL-sCoV de- modulated by cardiac causes. It is, however, beyond the scope of crease, however, appears to occur earlier than the CBF increase this study to evaluate cardiac cofactors to CVR such as heart and levels off during the second breath-hold phase, while CBF stroke volume or the diving reflex in detail. The factor age (an further increases. An earlier response to breath-hold leading to indirect measure of the cardiac and vascular influence on CVR) decreasing ATT before the CBF increase in the late breath-hold was assessed but was not associated with any of the flow parame- phase can be the explanation, denoting that the effect is mainly ters, which is not surprising considering that most volunteers in vascular and perfusion changes as detected by ASL follow later. this study were younger than 40 years of age. Also, a selection bias ASL-sCoV may be an earlier CVR marker of hypoxemia than of outstandingly healthy and well-trained freediver volunteers absolute CBF. must be considered a further contributing factor.

CVR is predominantly triggered by changes in blood CO2, A physiologic adaptation to breath-hold was another hypoth- while, for example, hypoxemia detected by peripheral chemore- esis to be tested in this study. A diminished CVR during hyper- ceptors is considered to play an independent-but-inferior role in carbia/hypoxemia was reported for patients with chronic obstruc- cerebral vasodilation.33 While Willie et al20 already reported that tive pulmonary disease as well as sleep apnea.35 In an attempt to

O2 metabolism has a crucial influence on the breath-hold capa- investigate a similar association between experience with hyper- bility of freedivers, stressing the role of O2 for breath-hold toler- carbic/hypoxemic states and CVR in freedivers, we estimated the 19 ance and CVR, Cross et al could not confirm an influence of O2 total amount of previous prolonged breath-holds of the partici- on cerebral autoregulation in their prolonged breath-hold study. pants before participation in the current breath-hold experiment Ͼ We identified falling O2 as an influential factor of CVR in this (defined as cumulative breath-holds of 2 minutes and 30 sec- study, which can usually not be observed in CBF studies applying onds in a lifetime). Our findings do not unconditionally corrob- hypercarbic-normoxic gas despite otherwise comparable CVR orate that breath-hold experience accounts for an adaptation between breath-hold and hypercarbic-normoxic gas studies.5,26 effect on CBF. First, CVR did not differ between more breath- Due to the design of this study, which could not rely on arterial hold-experienced freedivers and their less experienced counter-

CO2 measurements, it remains, however, impossible to discern parts. Second, by far, the 2 most prolonged breath-hold-experi- the relative contribution of hypercarbia and hypoxia to CBF in- enced freedivers, who additionally stated a high frequency of crease. CVR increased faster in the second half of the breath-hold longer breath-holds per training session, showed relatively high experiment in correlation with the secondarily faster SpO2 de- CBF values. However, for the remaining cohort, an association cline, which can be explained by pulmonary O2 stores that allow between experience and lower absolute CBF in all phases of the normal hemodynamic conditions during the first minutes of experiment could be observed, which may indicate that repetitive breath-hold. Cerebral near-infrared measurements in elite previous hypercarbia and hypoxemia have an acute cerebrovas- AJNR Am J Neuroradiol 39:1839–47 Oct 2018 www.ajnr.org 1845 cular effect during a breath-hold challenge. Due to the limited tabolism even during a prolonged breath-hold period of 5 min- cohort size, which also included freedivers normally specialized in utes and severely diminishing blood O2. Furthermore, ASL pa- shorter breath-holds, this interesting and also clinically relevant rameters, which are determined by blood flow and vessel diameter observation will need to undergo further critical evaluation in the alterations alike, serve as excellent candidate MR imaging param- future, favorably in a more homogeneous group regarding age eters to reveal this response, while 31P-MR spectroscopy revealed and freediving specialization. its utility to dynamically study acute changes in cerebral energy In prolonged breath-hold, the efficacy of the cerebrovascular metabolism. ASL may also provide evidence for long-term adap- but also the cardiac response to maintain a stable O2 supply to the tation of cerebral vasculature following repetitive hypoxia-hyper- brain despite decreasing availability is a crucial health aspect and capnia. Imaging and metabolic findings of the present freediver can be assessed spectroscopically regarding energy metabolism. study can be used to better understand CVR during hypoxia- Direct noninvasive in vivo measurements of brain energy metab- hypercapnia in critical care and sleep apnea conditions. olism during prolonged breath-hold are extremely rare.22,36 Cerebral lactate accumulation or acidosis was observed in ACKNOWLEDGMENTS none of our participants, suggesting a sufficient compensation The authors acknowledge Dipl.-Phys. Ju¨rgen Gieseke for his as- of limited O2 supply by recruitment of ATP stores and increas- sistance with the ASL sequence implementation, and we thank all ing CBF. However, we identified a substantial decrease in the freedivers for their outstanding effort to make this study possible. PCr / (ATP ϩ adenosine diphosphate) ratio during breath- hold. This can be interpreted as a compensatory PCr decrease Disclosures: Burkhard Ma¨dler—UNRELATED: Employment: Philips Healthcare, Com- ments: regular salary; Kim van de Ven—UNRELATED: Employment: Philips Health- to provide ATP by PCr hydrolysis as a consequence of declin- care. Comments: regular salary. ing O2 availability and reduced aerobic ATP production capac- ity in prolonged breath-hold. REFERENCES Rising peripheral venous lactate levels during the breath-hold 1. AIDA, International Association for the Development of Apnea. challenge in contrast to stable cerebral lactate stresses the shift Symbol of Freediving. SYMBhttps://www.aidainternational.org/ toward a preferred cerebral O2 supply in breath-hold, including a WorldRecords#recordsMan. Accessed June 28, 2018 CBF increase and, simultaneously, peripheral vasoconstriction. 2. Li PA, Kristian T, Shamloo M, et al. Effects of preischemic hypergly- The rising glucose levels are likely a consequence of adrenaline- cemia on brain damage incurred by rats subjected to 2.5 or 5 min- utes of forebrain ischemia. Stroke 1996;27:1592–601, discussion induced glucose mobilization and underline the exceptional met- 1601–02 Medline 37 abolic and mental challenge of prolonged breath-hold. The cor- 3. Peppard PE, Young T, Barnet JH, et al. Increased prevalence of sleep- relative venous pO2 analyses documented pathologically low O2 disordered breathing in adults. Am J Epidemiol 2013;177:1006–14 CrossRef Medline levels after 5 minutes of breath-hold (down to 60% SpO2) and corroborated similarities of this breath-hold experiment with 4. Urback AL, MacIntosh BJ, Goldstein BI. Cerebrovascular reactivity measured by functional magnetic resonance imaging during breath- clinical settings of hypoxemia. On the other hand, untrained per- hold challenge: a systematic review. Neurosci Biobehav Rev 2017;79: sons may encounter life-threatening consequences under these 27–47 CrossRef Medline circumstances, while freedivers face hypoxemia under voluntary 5. Tancredi FB, Hoge RD. Comparison of cerebral vascular reactivity and trained conditions. measures obtained using breath-holding and CO2 inhalation. Maximum breath-hold without contact with water is a partic- J Cereb Blood Flow Metab 2013;33:1066–74 CrossRef Medline 6. Pattinson KT, Governo RJ, MacIntosh BJ, et al. Opioids depress cor- ular challenge for freedivers because regular training sessions tical centers responsible for the volitional control of respiration. mostly include water immersion, which causes an augmented div- J Neurosci 2009;29:8177–86 CrossRef Medline ing reflex. Only very few participants fulfilled the inclusion crite- 7. Yadav SK, Kumar R, Macey PM, et al. Regional cerebral blood flow ria and were able to perform sufficient breath-hold in the noisy alterations in obstructive sleep apnea. Neurosci Lett 2013;555: MR imaging environment. Multiple candidates declined due to 159–64CrossRef Medline 8. Mutsaerts HJ, Petr J, Va´clavu˚L,etal.The spatial coefficient of vari- the lack of silence, or they would not volunteer for additional MR ation in arterial spin labeling cerebral blood flow images. J Cereb spectroscopy breath-hold experiments, thus the limited cohort Blood Flow Metab 2017;37:3184–92 CrossRef Medline size. Furthermore, participants did not tolerate a CO2 mask, 9. Erecin´ ska M, Silver IA. ATP and brain function. J Cereb Blood Flow which interfered with their meditative state. We therefore decided Metab 1989;9:2–19 CrossRef Medline 10. Clement P, Mutsaerts HJ, Vaclavu L, et al. Variability of physiologi- to analyze venous CO . However, while venous CO can be used 2 2 cal brain perfusion in healthy subjects: a systematic review of to confirm hypercapnia, it is, however, too variable to study the modifiers: considerations for multi-center ASL studies. J Cereb CO2 influence on CVR due to known substantial deviations from Blood Flow Metab 2017 Jan 1:271678X17702156. [Epub ahead of 38 arterial CO2 in the brain. For this reason, we can also only as- print] CrossRef Medline 11. Mutsaerts HJ, Lysvik E, Schrantee A, et al. ExploreASL: image pro- sume that all normoventilating volunteers started at normal CO2 cessing toolbox for multi-center arterial spin labeling population blood levels in the experiment. The unavailability of these other- analyses. In: Proceedings of the Congress and General Assembly of the wise valuable data limits the evaluation of the physiologic pro- European Society of Magnetic Resonance in Medicine and Biology, Bar- cesses behind our observations. celona, Spain. October 19–21, 2017 12. Ashburner J. A fast diffeomorphic image registration algorithm. Neuroimage 2007;38:95–113 CrossRef Medline CONCLUSIONS 13. Alsop DC, Detre JA, Golay X, et al. Recommended implementation This study revealed that experienced freedivers develop a CVR, of arterial spin-labeled perfusion MRI for clinical applications: a which is sufficient to maintain a physiologic cerebral energy me- consensus of the ISMRM perfusion study group and the European 1846 Keil Oct 2018 www.ajnr.org consortium for ASL in dementia. Magn Reson Med 2015;73:102–16 and vasoreactivity to CO2 measured by arterial spin labeling after 6 CrossRef Medline days at 4350m. Neuroimage 2013;72:272–79 CrossRef Medline 14. Tatu L, Moulin T, Bogousslavsky J, et al. Arterial territories of the 27. Nakagawa K, Serrador JM, Larose SL, et al. Autoregulation in the human brain: cerebral hemispheres. Neurology 1998;50:1699–708 posterior circulation is altered by the metabolic state of the visual CrossRef Medline cortex. Stroke 2009;40:2062–67 CrossRef Medline 15. Asllani I, Borogovac A, Brown TR. Regression algorithm correcting 28. Flu¨ck D, Beaudin AE, Steinback CD, et al. Effects of aging on the for partial volume effects in arterial spin labeling MRI. Magn Reson association between cerebrovascular responses to visual stimula- Med 2008;60:1362–71 CrossRef Medline tion, hypercapnia and arterial stiffness. Front Physiol 2014;5:49 16. Naressi A, Couturier C, Devos JM, et al. Java-based graphical user CrossRef Medline interface for the MRUI quantitation package. MAGMA 2001;12: 29. Bruce CD, Steinback CD, Chauhan UV, et al. Quantifying cerebro- 141–52 Medline vascular reactivity in anterior and posterior cerebral circulations 17. McCanny P, Bennett K, Staunton P, et al. Venous vs arterial blood during voluntary breath holding. Exp Physiol 2016;101:1517–27 gases in the assessment of patients presenting with an exacerbation CrossRef Medline of chronic obstructive pulmonary disease. Am J Emerg Med 2012;30: 30. Hasler G, Fromm S, Alvarez RP, et al. Cerebral blood flow in imme- 896–900 CrossRef Medline diate and sustained anxiety. J Neurosci 2007;27:6313–19 CrossRef 18. Vickers AJ. The use of percentage change from baseline as an out- Medline come in a controlled trial is statistically inefficient: a simulation 31. van Osch MJ, Teeuwisse WM, van Walderveen MA, et al. Can arterial study. BMC Med Res Methodol 2001;1:6 CrossRef Medline spin labeling detect white matter perfusion signal? Magn Reson Med 19. Cross TJ, Kavanagh JJ, Breskovic T, et al. Dynamic cerebral autoreg- 2009;62:165–73 CrossRef Medline ulation is acutely impaired during maximal apnoea in trained div- 32. Chen HL, Lin HC, Lu CH, et al. Systemic inflammation and altera- ers. PLoS One 2014;9:e87598 CrossRef Medline tions to cerebral blood flow in obstructive sleep apnea. J Sleep Res 20. Willie CK, Ainslie PN, Drvis I, et al. Regulation of brain blood flow 2017;26:789–98 CrossRef Medline and oxygen delivery in elite breath-hold divers. J Cereb Blood Flow 33. Ogoh S, Nakahara H, Ueda S, et al. Effects of acute hypoxia on cere- Metab 2015;35:66–73 CrossRef Medline brovascular responses to carbon dioxide. Exp Physiol 2014;99: 21. Zhou Y, Rodgers ZB, Kuo AH. Cerebrovascular reactivity measured 849–58 CrossRef Medline with arterial spin labeling and blood oxygen level dependent tech- 34. Eichhorn L, Erdfelder F, Kessler F, et al. Evaluation of near-infrared niques. Magn Reson Imaging 2015;33:566–76 CrossRef Medline spectroscopy under apnea-dependent hypoxia in humans. J Clin 22. Vestergaard MB, Larsson HB. Cerebral metabolism and vascular Monit Comput 2015;29:749–57 CrossRef Medline reactivity during breath-hold and hypoxic challenge in freediv- 35. Beaudin AE, Hartmann SE, Pun M, et al. Human cerebral blood flow ers and healthy controls. J Cereb Blood Flow Metab 2017 Jan control during hypoxia: focus on chronic pulmonary obstructive 1:271678X17737909. [Epub ahead of print] CrossRef Medline disease and obstructive sleep apnea. J Appl Physiol (1985) 2017;123: 23. Sorond FA, Hollenberg NK, Panych LP, et al. Brain blood flow and 1350–61 CrossRef Medline velocity: correlations between magnetic resonance imaging and 36. Bainbridge A, Tachtsidis I, Faulkner SD, et al. Brain mitochondrial transcranial Doppler sonography. J Ultrasound Med 2010;29: oxidative metabolism during and after cerebral hypoxia-ischemia 1017–22 CrossRef Medline studied by simultaneous phosphorus magnetic-resonance and 24. Vestergaard MB, Lindberg U, Aachmann-Andersen NJ, et al. Com- broadband near-infrared spectroscopy. Neuroimage 2014;102(Pt 1): parison of global cerebral blood flow measured by phase-contrast 173–83 CrossRef Medline mapping MRI with (15) O-H2 O positron emission tomography. J 37. Eichhorn L, Erdfelder F, Kessler F, et al. Influence of apnea-induced Magn Reson Imaging 2017;45:692–99 CrossRef Medline hypoxia on catecholamine release and cardiovascular dynamics. Int 25. Sato K, Sadamoto T, Hirasawa A, et al. Differential blood flow re- J Sports Med 2017;38:85–91 CrossRef Medline sponses to CO(2) in human internal and external carotid and ver- 38. Kelly AM. Review article: can venous blood gas analysis replace ar- tebral arteries. J Physiol 2012;590:3277–90 CrossRef Medline terial in emergency medical care. Emerg Med Australas 2010;22: 26. Villien M, Bouzat P, Rupp T, et al. Changes in cerebral blood flow 493–98 CrossRef Medline

AJNR Am J Neuroradiol 39:1839–47 Oct 2018 www.ajnr.org 1847 ORIGINAL RESEARCH INTERVENTIONAL

Reasons for Reperfusion Failures in Stent-Retriever-Based Thrombectomy: Registry Analysis and Proposal of a Classification System

X J. Kaesmacher, X J. Gralla, X P.J. Mosimann, X F. Zibold, X M.R. Heldner, X E. Piechowiak, X T. Dobrocky, X M. Arnold, X U. Fischer, and X P. Mordasini

ABSTRACT BACKGROUND AND PURPOSE: In 5%–10% of patients with acute ischemic stroke with an intention to treat with mechanical thrombec- tomy, no reperfusion can be achieved (Thrombolysis in Cerebral Infarction score ϭ 0/1). Purpose of this analysis was a systematic assessment of underlying reasons for reperfusion failures.

MATERIALS AND METHODS: An intention-to-treat single-center cohort (n ϭ 592) was re-evaluated for all patients in whom no reperfusion could be achieved (n ϭ 63). Baseline characteristics of patients were compared between patients with and without reperfusion failures. After qualitative review of all cases with reperfusion failures, a classification system was proposed and relative frequencies were reported. In a second step, occurrence of delayed recanalization at 24 hours after reperfusion failure and dependency on IV-tPA were evaluated.

RESULTS: In 63/592 patients with an intention to perform stent-retriever thrombectomy, no reperfusion was achieved (TICI 0/1, 10.6%, 95% CI, 8.2%–13.1%). Older patients (adjusted OR per yr ϭ 1.03; 95% CI, 1.01–1.05) and patients with M2 occlusion (adjusted OR ϭ 3.36; 95% CI, 1.82–6.21) were at higher risk for reperfusion failure. In most cases, no reperfusion was a consequence of technical difficulties (56/63, 88.9%). In one-third of these cases, reperfusion failures were due to the inability to reach the target occlusion (20/63, 31.7%), while “stent-retriever failure” occurred in 39.7% (25/63) of patients. Delayed recanalization was very rare (18.2%), without dependence on IV-tPA pretreatment status.

CONCLUSIONS: Reasons for reperfusion failure in stent-retriever thrombectomy are heterogeneous. The failure to establish intracranial or cervical access is almost as common as stent-retriever failure after establishing intracranial access. Systematic reporting standards of reasons may help to further estimate relative frequencies and thereby guide priorities for technical development and scientific effort.

he technical success of endovascular stroke therapy is one retrievers as the dominant technical platform for thrombec- Tof the most important modifiable predictors of therapy tomy; however, this may change with the final publication of A benefit in patients presenting with large-vessel-occlusion acute Comparison of Direct Aspiration Versus Stent Retriever as a First ischemic stroke.1,2 The American Stroke Association/Ameri- Approach (COMPASS) trial.3-5 Recently published large-cohort can Heart Association guidelines continue to support stent registries of patients treated with stent retrievers have shown that contemporary endovascular interventions in acute ischemic stroke are angiographically successful in up to 80%–90% of cases (Throm- Received March 22, 2018; accepted after revision June 25. bolysis in Cerebral Infarction 2b/3).6,7 While unsuccessful reperfusion From the University Institute of Diagnostic and Interventional Neuroradiology (J.K., J.G., P.J. Mosimann, F.Z., E.P., T.D., P. Mordasini) and Department of Neurology (J.K., (ՅTICI 2a) often results from incomplete retrieval due to distal M.R.H., M.A., U.F.), University Hospital Bern and University of Bern, Inselspital, Bern, Switzerland. embolization and/or clot fragmentation (TICI 2a), in some Urs Fischer and Pasquale Mordasini shared senior authorship. patients, no reperfusion (TICI 0/1) can be achieved (“reperfu- This work was supported by the Swiss Stroke Society, the Bangerter Foundation, sion failure”). Reasons for such reperfusion failures may range from and the Swiss Academy of Medical Sciences through the “Young Talents in Clinical difficulty establishing cervical or intracranial access to the inability to Research” program. Please address correspondence to Pasquale Mordasini, MD, MSc, University dislocate and retrieve the clot despite having reached the target loca- Institute of Diagnostic and Interventional Neuroradiology, University of Bern, tion and having established intracranial access.8,9 Further conceiv- Inselspital Bern, Freiburgstr 8, CH-3010, Switzerland; e-mail: [email protected] able explanations for failures to reestablish flow are underlying Indicates open access to non-subscribers at www.ajnr.org nonembolic vessel diseases (eg, vasculitis, intracranial atherosclero- Indicates article with supplemental on-line table. sis)10,11 or thrombi of nonthrombotic origin and extraordinary com- http://dx.doi.org/10.3174/ajnr.A5759 position, such as calcified or neoplastic thrombi.12,13

1848 Kaesmacher Oct 2018 www.ajnr.org The aim of this analysis was to provide estimates of the relative For all cases, categorization of reasons and the evaluation regard- frequencies of underlying causes of reperfusion failure in patients ing the primary intention of treating with a stent-retriever-based with acute ischemic stroke who underwent angiography with an thrombectomy were based on the following information: intention to perform stent-retriever-based thrombectomy. ● Review of the initial radiologic report with respect to a written decision and interdisciplinary consensus for stent-retriever-based MATERIALS AND METHODS thrombectomy. The prospective Bernese Stroke Registry was accessed to find all di- ● Review of materials used along with an angiographic report con- rectly admitted patients with acute ischemic stroke with an intention cerning catheter changes and a description of interventional to perform stent-retriever thrombectomy from January 2012 to July difficulties 2017 (n ϭ 592). For the classification of reasons for reperfusion fail- ● Review of all preinterventional and angiographic images. ure, all patients with postinterventional final Thrombolysis in Cere- bral Infarction grades 0 and 1 were included.14 Most important, pa- tients in whom no stent retriever was deployed (eg, because of access Patient Characteristics difficulties or vessel elongation) were also analyzed. The Bernese Baseline parameters, clinical outcomes, and information on re- Ϯ Stroke Registry was approved by the local ethics committee (Kan- canalization status at 24 6-hour follow-up are provided. Fol- tonal Ethics Committee Bern, amendment access number: 231/2014). low-up recanalization status was graded using the 4-step arterial occlusive lesion score14 on postinterventional intracranial vessel Classification of Reasons for Reperfusion Failures imaging, if available (44/63, 13 missing because only noncontrast After qualitative review of all cases with reperfusion failure, the CT was performed, 6 missing due to early death). underlying reasons for TICI 0/1 were classified into the follow- ing categories by a consensus of a neuroradiologist in training Endovascular Therapy (J.K., 3 years of experience) and an interventional neuroradi- Endovascular therapy was performed immediately after CT or ologist (P. Mordasini, 15 years of experience). MR imaging under the following conditions: 1) The diagnosis of ischemic stroke was established; 2) the NIHSS score on admission Technical Reasons: Target Occlusion Not Reached assessed by a neurologist was Ն4 points, isolated aphasia or hemi- (Category I) anopia or severe paresis of 1 hand was present, or neurologic deficits recurred; 3) CT or MR angiography showed occlusion of a ● IA: Intracranial target occlusion was not reached due to large intracranial artery; 4) hemorrhage was excluded; 5) neuro- marked cervical vessel tortuosity including twisted, logic deficits correlated with the vessel occlusion; and 6) no indi- looped, or kinked vessels.15 The proximal cervical vessels were successfully catheterized. vidual clinical or premorbid conditions or laboratory findings ● IB: Target occlusion was not reached owing to failed cathe- were contraindications. When the criteria for endovascular ther- terization of proximal supra-aortic vessels. The proximal apy were fulfilled, digital subtraction angiography was performed cervical vessels were not successfully catheterized owing to via a transfemoral approach using a biplane, high-resolution an- difficult aortic arch anatomy.16 giography system (Axiom Artis zee; Siemens, Erlangen, Germany) ● IC: Target occlusion was not reached owing to the inability to using iopamidol (Iopamiro 300; Bracco, Milan, Italy) as a contrast pass a cervical ICA occlusion (eg, unpassable tandem lesion). agent. The choice of access and retrieval technique was left to the discretion of the neurointerventionalist, taking access anatomy Technical Reasons: Target Occlusion Reached and occlusion pattern into account. For anterior circulation (Category II) strokes, a first-line recanalization technique consists of placing an 8F or 9F balloon-guiding catheter over a long exchange wire as ● IIA: Target occlusion was reached, but the operator was high into the cervical ICA as possible followed by stent-retriever unable to pass the thrombus with the microwire/microcath- thrombectomy under proximal balloon occlusion and manual eter. In these cases, no stent retriever is deployed. aspiration. In cases in which placement of a balloon-guiding cath- ● IIB: Target occlusion was reached, the stent retriever was de- eter was deemed suboptimal (eg, low position in the cervical ICA ployed, but no reperfusion occured after multiple retrievals or distal common carotid artery or highly tortuous cervical ves- (no clot retrieval or dislocation), thus, stent-retriever failure. sels) or stent-retriever thrombectomy through the balloon-guid- ● IIC: Initial reperfusion was achieved, followed by spontaneous or iatrogenic reocclusion (eg, intracranial stenosis, intracranial ing catheter alone was unsuccessful, a 5F or 6F intermediate cath- dissection, or perforation with subsequent vessel sacrifice). eter for concomitant distal aspiration during stent-retriever thrombectomy was used as a second-line technique. Stand-alone aspiration was only used as a third-line technique after failure of Nontechnical Reasons: Other (Category III) stent-retriever thrombectomy. During the study period, several ● Presumed futility. different stent retriever models were used, the mainstay consisting ● Adverse non-neurologic event with the need to stop me- of the Solitaire device (Covidien, Irvine, California). The decision chanical thrombectomy. of when to abandon the procedure and whether to administer ● Signs of contrast extravasation without perforation (early intra-arterial thrombolysis with urokinase was left to the discre- hemorrhagic transformation). tion of the neurointerventionalist. AJNR Am J Neuroradiol 39:1848–53 Oct 2018 www.ajnr.org 1849 FIGURE. Distribution of reasons for reperfusion failures. Relative frequencies of reasons underlying reperfusion failures. The most common reason was reperfusion failure after intracranial access was established (category II, 36/63). Here, the most common reason was the inability to retrieve the clot despite multiple attempts, stent-retriever failure (IIB). Other reasons in this category group included the inability to pass the clot with the microwire or microcatheter (IIA) or secondary spontaneous/iatrogenic reocclusion after initial reperfusion was established (IIC). Access failures were observed in 20 cases (category I). For further exact definitions of respective subcategories, see the proposed classification scheme outlined in the “Materials and Methods” section.

Statistical Analysis 95% CI, 1.82–6.21; and 1.03; 95% CI, 1.01–1.05, respectively, with Categoric group comparisons were performed applying the Fisher the Nagelkerke R2ϭ 0.074). The discriminative power of the multi- exact test. Normally distributed data are presented as mean Ϯ SD, variate logistic regression model derived from the included variables while non-normally distributed data are shown as median (inter- age, time to diagnosis, and M2 occlusions was weak (area under the quartile range). Comparison of continuous or ordinally scaled curve ϭ 0.665; 95% CI, 0.585–0.746). variables was performed using the Mann-Whitney U test or the Welch t test for independent samples (for non-normally and nor- Distribution of Reasons for TICI 0/1 mally distributed variables, respectively). Estimated 95% confi- In most cases, no reperfusion was a consequence of technical dif- dence intervals of prevalences were calculated using the normal ficulties (categories I and II: 56/63, 88.9%; 95% CI, 81.1%–96.6%, approximation interval (Wald interval). Variables with univariate Figure). One-third of reperfusion failures were due to the inability significant differences between groups were entered into a multi- to reach the target occlusion (category I: 20/63, 31.7%; 95% CI, variate logistic regression model. Results from multivariate lo- 20.3%–43.2%). Reasons for not reaching the target occlusion gistic regression analysis were presented as adjusted ORs and were subcategorized into cervical vessel tortuosity (the proximal respective 95% confidence intervals. Predicted probabilities cervical vessel was catheterized, 10/20), failed catheterization of were analyzed with receiver operator characteristic analysis proximal supra-aortic vessels because of difficult acrotic arch with calculation of the area under the curve to evaluate the anatomy (8/20), or, rarely, the inability to pass a proximal ICA discriminative power of the model. occlusion/stenosis in patients presenting with extracranial-intra- cranial tandem lesions (2/20). If the target occlusion was reached RESULTS (category II: 36/63, 57.1%; 95% CI, 44.9%–69.4%), reperfusion Study Cohort failure was due to the inability to pass the intracranial occlusion Five hundred ninety-two patients were included (mean age, with the microwire/microcatheter in 7/36 cases. When stent re- 72.2 Ϯ 14.3 years; 47.3% female). Of these patients, 93.1% (n ϭ trievers were deployed (29/36), the most common reason for fail- 551) were treated for anterior circulation occlusions. The median ure was the inability to retrieve/dislocate the thrombus after mul- symptom-onset to diagnosis interval (witnessed or last seen well) tiple attempts (stent-retriever failure, 25/36; median attempts, 3), was 126 minutes (interquartile range, 90–222 minutes), and pa- while in the remaining cases, iatrogenic perforation and subsequent tients presented with severe neurologic deficits (median admis- coiling were causative (4/36). Nontechnical reasons for TICI 0/1 (cat- sion NIHSS, 15; interquartile range, 9–20). Other baseline char- egory III: 7/63, 11.1%; 95% CI, 3.4%–18.9%) were an active consen- acteristics are shown in the On-line Table. In 63 of 592 patients sus decision to stop treatment after diagnostic angiography or an with an intention to treat with stent-retriever thrombectomy, no initial thrombectomy attempt (presumed futility, 5/7), a non-neu- reperfusion was achieved (TICI 0/1, 10.6%; 95% CI, 8.2%– rologic event with the need to stop endovascular therapy (1/7), 13.1%). Patients in whom no reperfusion was achieved were older and signs of contrast extravasation without perforation, inter- (mean age, 76.9 versus 71.6 years; P ϭ .004), were treated later preted as early hemorrhagic transformation (1/7). (median symptom-onset to diagnosis 155 versus 121 minutes, P ϭ .007), and had more M2 and posterior circulation occlusions (P Rescue and Patency at 24-Hour Follow-Up for overall difference Ͻ .001). Other baseline characteristics did not Follow-up imaging at 24 hours was available for 44/63 (69.8%) reveal significant differences (On-line Table). When we entered sig- patients without initial reperfusion. Of those patients, 8 (18.2%) nificant variables derived from univariate comparison into a multi- revealed substantial vessel recanalization at follow-up (arterial oc- variable logistic regression model, only M2 occlusions and increased clusive lesion score, 2/3), while all other patients (36/44, 81.8%) age were associated with reperfusion failure (adjusted OR ϭ 3.36; showed evidence of persistent occlusion without or with minimal 1850 Kaesmacher Oct 2018 www.ajnr.org recanalization (arterial occlusive lesion score, 0/1). Recanaliza- atherosclerosis, dissection, or vasculitis.10,11 According to current tion at follow-up tended to be observed less often if the reason for observational studies, thrombus length has not been shown to no reperfusion was a category II failure (11.5% versus 33.3%, P ϭ effect retrieval efficacy.28,29 Besides clot and occlusion site char- .176). If intra-arterial thrombolysis was applied as a rescue, a non- acteristics, stent-retriever dwelling time30 and respective device- significant trend toward higher rates of recanalization at fol- clot interaction,8 may serve as additional contributing factors. low-up was observed (33.3% versus 11.5%, P ϭ .125). No differ- The present analysis included only patients with an intention to ence in 24-hour recanalization in patients with reperfusion failure treat with stent-retriever-based thrombectomy. Recent random- was observed when comparing those treated with IV-tPA with ized-controlled trial results have suggested that similar technical those treated without IV-tPA (21.4% versus 16.7%, P ϭ .501). and clinical success rates could be achieved with aspiration tech- niques using large-bore catheters (COMPASS, International 31 DISCUSSION Stroke Conference 2018). A different mechanical retrieval ap- The presented study suggests that reasons for reperfusion failure in proach is likely to increase the probability of achieving reperfu- patients with an intention to perform stent-retriever-based throm- sion, because some occlusions, which are refractory to one tech- bectomy are not homogeneous and underline the notion that reper- nique may respond favorably to an other approach. Currently, fusion failure in contemporary stroke thrombectomy is a multifac- there is emerging evidence that the relative effectiveness of one torial problem.8 The failure to establish intracranial or cervical access approach over the other may depend on the localization of the is approximately as common as the inability to retrieve the clot once occlusion (eg, anterior-versus-posterior circulation)32 or the adequate intracranial positioning and access to the target occlusion shape of the proximal occlusion site.33 are established (true stent-retriever failure). The presented data high- light the need to distinguish among potential factors associated with Delayed Recanalization and Rescue Options reperfusion failure and advocate for technical development and sci- While reperfusion rates in intracranial vessel occlusions after IV- 34 entific effort equally focusing on stent-retriever efficacy as well as tPA may be as high as 60% after 7 hours, only a few patients had tools and alternative access routes to improve cervical and intracra- reperfusion at 24 hours after endovascular reperfusion failure, nial access. particularly after stent-retriever failure (ϳ10%). These results provide preliminary evidence that clots not responding to me- Access Failures and Alternative Access Routes chanical treatment also show low response rates to intravenous The two main domains of difficulty are the inability to reach the thrombolysis and spontaneous clot lysis, corroborating findings target occlusion and the failure to dislocate or retrieve the clot. A that they may represent clots of distinct composition (eg, promi- recent study has suggested that carotid tortuosity relates to tech- nent calcification).12 In the presented cohort, rescue administra- nical failure17; however, another study did not find a significant tion of intra-arterial thrombolytics showed a trend toward higher association between carotid elongation and technical success or rates of recanalization at follow-up, without reaching statistical procedure length.18 One reason for the discrepant results of these significance. Recently, low-dose tirofiban has been associated studies may be because vessel elongation predominantly causes with increased reperfusion rates in patients treated with mechan- category I reperfusion failures, while intracranial geometry and ical thrombectomy.35,36 However, the results are mainly derived thrombus properties may primarily relate to stent-retriever fail- from Asian cohorts, which have higher rates of underlying intra- ures (category IIB). Distribution of these reasons in such cohorts cranial atherosclerosis and higher rates of early re-occlusions, may thus severely affect the sensitivity and respective power of the thus severely limiting its transferability to other cohorts. Besides aforementioned analyses. Although no femoral access failure has medical rescue approaches, stent placement and Y-stent-retriever been observed in the presented consecutive cohort, brachial19 or ca- maneuvers may serve as mechanical rescue approaches if an in- rotid access20,21 may serve as a reliable rescue approach in cases of tracranial access was successfully established.37-39 A recent cohort femoral access failure or other category I failures (eg, in the presence study has suggested that the higher recanalization rates achievable of vessel tortuosity or the inability to adequately catheterize the prox- with rescue permanent intracranial stent placement translated imal supra-aortic vessels). Until the date of this analysis, we had not into clinical benefit without increasing the risk of symptomatic used alternative access routes in a standardized fashion if the femoral intracerebral hemorrhage. Although further studies in other pop- artery was successfully cannulated but a category I failure had oc- ulations are needed to confirm these findings, the results promote curred. One consequence of this analysis is the institutional introduc- permanent stent placement as a feasible rescue technique in cases tion of a more widely applied and standardized use of alternative of stent-retriever failures.40 access routes. We now consider these alternative routes not only when femoral access cannot be achieved (failed femoral artery can- Identifying Patients with a High Risk of Reperfusion nulation) but also in all cases of category I failures. Failure Preinterventional stratification of patients with a high chance Retrieval Failures of mechanical reperfusion failure is desirable. This is impor- Several factors influencing stent-retriever failure have been pro- tant not only because in these cases, inclusion into currently posed. So far, there is some preliminary evidence that MCA geo- enrolling randomized-controlled trials (Multicenter Random- metric anatomy,22,23 along with thrombus composition12,24-26 ized Clinical Trial of Endovascular Treatment for Acute Isch- and shape,27 may predict the inability to dislocate or retrieve the emic Stroke in the Netherlands–NO IV, [ISRCTN80619088]; clot. Other factors may include underlying vessel disease such as and Solitaire With the Intention For Thrombectomy Plus In- AJNR Am J Neuroradiol 39:1848–53 Oct 2018 www.ajnr.org 1851 travenous t-PA versus DIRECT Solitaire Stent-Retriever as technically and clinically equally effective approaches Thrombectomy in Acute Anterior Circulation Stroke [SWIFT (COMPASS/Combined Use of Contact Aspiration and the Stent DIRECT, NCT03192332, clinicaltrials.gov]) may raise poten- Retriever Technique Versus Stent Retriever Alone for Recanalisa- tial ethical considerations but also because primary alternative tion in Acute Cerebral Infarction [ASTER]).31 Although aspira- access routes might be considered beforehand. However, no tion attempts were implemented in the standard rescue approach reliable classification algorithm with high sensitivity and spec- at our center, overall distribution of reperfusion failure reasons ificity is currently available. In our cohort, patients with rep- may differ in centers that more commonly use aspiration tech- erfusion failure were older, treated later, and frequently had niques (ie, as first attempt) and other categories must be added. M2 occlusions. Combining those parameters, however, yielded low accuracy in correctly identifying patients with reperfusion CONCLUSIONS failures. The issue of more reperfusion failures in M2 occlu- Reasons for reperfusion failure in stent-retriever thrombectomy sions observed in the presented cohort deserves attention be- are heterogeneous. The failure to establish intracranial or cervical cause a recent meta-analysis has suggested that endovascular access is nearly as common as the inability to retrieve the clot therapy is successful in up to 85% of M2 occlusion.41 In this despite the clot having been passed and adequate intracranial posi- meta-analysis, recanalization rates were comparable between tioning having been established (true stent-retriever failure). System- patients with M1 and M2 occlusions.41 However, the author atic reporting standards of those reasons may help to elucidate acknowledged that the results can only be “interpreted in the relative frequencies and thereby guide priorities for technical context of patients with M2 occlusions that can be safely ac- development and scientific effort (eg, adequate subgroup analyses cessed by mechanical thrombectomy.”41 In the present analy- regarding predictive factors). The low rates of delayed recanalization sis, the initial intention for stent-retriever thrombectomy was after mechanical reperfusion failures underline the benefit of alter- crucial for inclusion, and patients were included even if no native access routes and the need for a systematic evaluation of other stent retriever was deployed at any time point during angiog- medical (intra-arterial thrombolytics, antiplatelets, and so forth), or, raphy. This criterion may differ from that in most of the ret- if feasible, mechanical (eg, stent placement) rescue approaches. rospective studies included into the aforementioned quantita- tive synopsis because the major inclusion criterion was Disclosures: Johannes Kaesmacher—RELATED: Grant: Young Talent in Clinical Re- search Grant (Swiss Academy of Medical Sciences and Bangerter Foundation). Jan treatment with a stent retriever, rather than intention to treat Gralla—RELATED: Consulting Fee or Honorarium: Medtronic, Comments: Global with stent retrievers.41 Furthermore, the overall frequency of Principal Investigator of the SWIFT Direct Trial, consultant for Medtronic, member of M2 occlusions treated with stent retrievers was small in our the Clinical Event Committee of the PROMISE Study (Penumbra)*; UNRELATED: Grants/Grants Pending: Swiss National Funds, Comments: grant for stroke outcome cohort, giving it scope for confounding due to the operator’s prediction in MRI,* Marcel Arnold—RELATED: Grant: Swiss Heart Foundation, Com- experience and associated learning curves. ments: research grant*; Consulting Fee or Honorarium: Covidien, Medtronic, Com- ments: honoraria for lectures; UNRELATED: Consultancy: Bayer HealthCare, Bristol- Myers Squibb, Boehringer Ingelheim, Daichi Sankyo, Pfizer, Amgen, Comments: Strengths and Limitations Scientific Advisory Board honoraria; Grants/Grants Pending: Swiss National Science While the present data allow a real-world intention-to-treat anal- Foundation, Comments: research grants.* Urs Fischer—UNRELATED: Consultancy: ysis of the relative frequencies of reasons for reperfusion failures, Medtronic, Stryker*; Grants/Grants Pending: Medtronic, Comments: research grant.* *Money paid to the institution. there are several limitations to this study: First, category IIB is defined as reperfusion failure after multiple attempts. Thus, there may be different thresholds for different interventionalists and REFERENCES centers to stop at different time points, and aggressiveness of res- 1. Goyal M, Menon BK, Van Zwam WH, et al; HERMES collaborators. cue approaches vary. Second, futility criteria based on a consensus Endovascular thrombectomy after large-vessel ischaemic stroke: a between the treating neurologists and neuroradiologists may meta-analysis of individual patient data from five randomised tri- Ͼ als. Lancet 2016;387:1723–31 CrossRef Medline change with time (eg, advanced time windows of 10 hours may 2. Manning NW, Chapot R, Meyers PM. Endovascular stroke no longer serve as a futility reason in the era of published Endo- management: key elements of success. Cerebrovasc Dis 2016;42: vascular Therapy Following Imaging Evaluation for Acute Isch- 170–77 CrossRef Medline emic Stroke 3 (DEFUSE 3)42 and Clinical Mismatch in the Triage 3. Leslie-Mazwi T, Chandra RV, Fraser JF, et al. AHA/ASA 2018 AIS of Wake Up and Late Presenting Strokes Undergoing Neuroin- guidelines: impact and opportunity for endovascular stroke care. 43 J Neurointerv Surg 2018 May 28. [Epub ahead of print] CrossRef Medline tervention with Trevo (DAWN) trial results. Third, intracranial 4. Powers WJ, Rabinstein AA, Ackerson T, et al; American Heart Asso- access stability and torqueing forces on the stent retriever may ciation Stroke Council. 2018 Guidelines for the Early Management impact the vessel and clot geometry and the respective clot inter- of Patients with Acute Ischemic Stroke: A Guideline for Healthcare action of the device.8 Thus, a classification into the true stent- Professionals from the American Heart Association/American Stroke Association. Stroke 2018;49:e46–110 CrossRef Medline retriever failure category may neglect the partial contribution of 5. Turk AS, Siddiqui AH, Mocco J. A comparison of direct aspiration ver- unstable intracranial access, potentiating difficulties to retrieve sus stent retriever as a first approach (‘COMPASS’): protocol. J Neuro- the clot or impeding the device efficacy.8 Fourth, standardized use interv Surg 2018 Feb 20. [Epub ahead of print] CrossRef Medline of more aggressive rescue approaches for access failures may have 6. Froehler MT, Saver JL, Zaidat OO, et al; on behalf of the STRATIS averted some reperfusion failures in other centers. Last, distal as- Investigators. Interhospital transfer prior to thrombectomy is asso- ciated with delayed treatment and worse outcome in the STRATIS piration techniques with large-bore catheters have gained popu- Registry. Circulation 2017;136:2311–21 larity and have shown increasing effectiveness during the past 7. Zaidat OO, Castonguay AC, Nogueira RG, et al. TREVO stent-re- years. Recent randomized-controlled trial results endorsed them triever mechanical thrombectomy for acute ischemic stroke sec- 1852 Kaesmacher Oct 2018 www.ajnr.org ondary to large vessel occlusion registry. J Neurointerv Surg 2018;10: 27. Zhu L, Liebeskind DS, Jahan R, et al. Thrombus branching and vessel 516–24 CrossRef Medline curvature are important determinants of middle cerebral artery 8. Yoo AJ, Andersson T. Thrombectomy in acute ischemic stroke: trunk recanalization with Merci thrombectomy devices. Stroke challenges to procedural success. J Stroke 2017;19:121–30 CrossRef 2012;43:787–92 CrossRef Medline Medline 28. Seker F, Pfaff J, Wolf M, et al. Impact of thrombus length on recan- 9. Ribo M, Flores A, Rubiera M, et al. Interhospital transfer prior to alization and clinical outcome following mechanical thrombec- thrombectomy is associated with delayed treatment and worse out- tomy in acute ischemic stroke. J Neurointerv Surg 2017;9:937–39 come in the STRATIS Registry. J Neurointerv Surg 2013;5(Suppl 1): CrossRef Medline i70–73 CrossRef Medline 29. Weisstanner C, Gratz PP, Schroth G, et al. Thrombus imaging in 10. Jia B, Feng L, Liebeskind DS, et al. Mechanical thrombectomy and acute stroke: correlation of thrombus length on susceptibility- rescue therapy for intracranial large artery occlusion with underly- weighted imaging with endovascular reperfusion success. Eur Ra- ing atherosclerosis. J Neurointerv Surg 2017 Dec 4. [Epub ahead of diol 2014;24:1735–41 CrossRef Medline print] CrossRef Medline 30. Kannath SK, Rajan JE, Sylaja PN, et al. Dwell time of Stentriever 11. Lee JS, Hong JM, Lee KS, et al. Primary stent retrieval for acute influences complete revascularization and first-pass TICI 3 revas- intracranial large artery occlusion due to atherosclerotic disease. J cularization in acute large vessel occlusive stroke. World Neurosurg Stroke 2016;18:96–101 CrossRef Medline 2018;110:169–73 CrossRef Medline 12. Dobrocky T, Piechowiak E, Cianfoni A, et al. Thrombectomy of cal- 31. Lapergue B, Blanc R, Gory B, et al ASTER Trial Investigators. Effect of cified emboli in stroke: does histology of thrombi influence the ef- endovascular contact aspiration vs stent retriever on revasculariza- fectiveness of thrombectomy? J Neurointerv Surg 2018;10:345–50 tion in patients with acute ischemic stroke and large vessel CrossRef Medline occlusion: the ASTER randomized clinical trial. JAMA 2017;318: 13. Judge C, Mello S, Bradley D, et al. A systematic review of the causes 443–52 CrossRef Medline and management of ischaemic stroke caused by nontissue emboli. 32. Gory B, Mazighi M, Blanc R, et al; Endovascular Treatment in Isch- Stroke Res Treat 2017;2017:7565702 CrossRef Medline emic Stroke (ETIS) Research Investigators. Mechanical thrombec- 14. Zaidat OO, Yoo AJ, Khatri P, et al; Cerebral Angiographic Revascu- tomy in basilar artery occlusion: influence of reperfusion on clini- larization Grading (CARG) Collaborators, STIR Revascularization cal outcome and impact of the first-line strategy (ADAPT vs stent working group, STIR Thrombolysis in Cerebral Infarction (TICI) retriever). J Neurosurg 2018 Jan 12:1–10. [Epub ahead of print] Task Force. Recommendations on angiographic revascularization CrossRef Medline grading standards for acute ischemic stroke: a consensus state- 33. Consoli A, Rosi A, Coskun O, et al. Thrombectomy for M1-middle ment. Stroke 2013;44:2650–63 CrossRef Medline cerebral artery occlusion angiographic aspect of the arterial occlu- 15. Han HC. Twisted blood vessels: symptoms, etiology and biome- sion and recanalization: a preliminary observation. 2018;49: chanical mechanisms. J Vasc Res 2012;49:185–97 CrossRef Medline 1286–89 CrossRef Medline 16. Burzotta F, Nerla R, Pirozzolo G, et al. Clinical and procedural im- 34. Menon BK, Najm M, Al-Ajlan F, et al. Abstract 186: IV tPA recana- pact of aortic arch anatomic variants in carotid stenting proce- lization rates by site of occlusion and time after tPA bolus: main dures. Catheter Cardiovasc Interv 2015;86:480–89 CrossRef Medline results of the Interrsect Multinational Multicenter Prospective 17. Kaymaz Z, Nikoubashman O, Brockmann M, et al. Influence of ca- Cohort Study. Stroke 2017;48:A186. http://stroke.ahajournals.org/ rotid tortuosity on internal carotid artery access time in the treat- content/48/Suppl_1/A186.abstract. Accessed January 4, 2018 ment of acute ischemic stroke. Interv Neuroradiol 2017;23:583–88 35. Goh D, Jin S, Jeong HW, et al. Mechanical Solitaire thrombectomy CrossRef Medline with low-dose booster tirofiban injection. Neurointervention 2016; 18. Yilmaz U, Mu¨hl-Benninghaus R, Simgen A, et al. Carotid elongation 11:114–19 CrossRef Medline does not affect angiographic results of mechanical thrombectomy 36. Zhao W, Che R, Shang S, et al. Low-dose tirofiban improves func- in acute stroke. Clin Neuroradiol 2016;26:183–87 CrossRef Medline tional outcome in acute ischemic stroke patients treated with endo- 19. Snelling BM, Sur S, Shah SS, et al. Transradial access: lessons learned vascular thrombectomy. Stroke 2017;48:3289–94 CrossRef Medline from cardiology. J Neurointerv Surg 2018;10:487–92 CrossRef Medline 37. Brekenfeld C, Schroth G, Mattle HP, et al. Stent placement in acute 20. Mokin M, Snyder KV, Levy EI, et al. Direct carotid artery puncture cerebral artery occlusion: use of a self-expandable intracranial stent access for endovascular treatment of acute ischemic stroke: techni- for acute stroke treatment. Stroke 2009;40:847–52 CrossRef Medline cal aspects, advantages, and limitations. J Neurointerv Surg 2015;7: 38. Baek JH, Kim BM, Kim DJ, et al. Stenting as a rescue treatment after 108–13 CrossRef Medline failure of mechanical thrombectomy for anterior circulation large 21. Blanc R, Piotin M, Mounayer C, et al. Direct cervical arterial access artery occlusion. Stroke 2016;47:2360–63 CrossRef Medline for intracranial endovascular treatment. Neuroradiology 2006;48: 39. Crosa R, Spiotta AM, Negrotto M, et al. “Y-stent retriever”: a new 925–29 CrossRef Medline rescue technique for refractory large-vessel occlusions? J Neurosurg 22. Schwaiger BJ, Gersing AS, Zimmer C, et al. The curved MCA: influ- 2018;128:1349–53 CrossRef Medline ence of vessel anatomy on recanalization results of mechanical 40. Chang Y, Kim BM, Bang OY, et al. Rescue stenting for failed me- thrombectomy after acute ischemic stroke. AJNR Am J Neuroradiol chanical thrombectomy in acute ischemic stroke: a multicenter ex- 2015;36:971–76 CrossRef Medline perience. Stroke 2018;49:958–64 CrossRef Medline 23. Yamamoto S, Yamagami H, Todo K, et al. Correlation of middle 41. Saber H, Narayanan S, Palla M, et al. Mechanical thrombectomy for cerebral artery tortuosity with successful recanalization using the acute ischemic stroke with occlusion of the M2 segment of the mid- Merci retrieval system with or without adjunctive treatments. Neu- dle cerebral artery: a meta-analysis. J Neurointerv Surg 2018;10: rol Med Chir (Tokyo) 2014;54:113–19 CrossRef Medline 620–24 CrossRef Medline 24. Sporns PB, Hanning U, Schwindt W, et al. Ischemic stroke: histolog- 42. Albers GW, Marks MP, Kemp S, et al; DEFUSE 3 Investigators. ical thrombus composition and pre-interventional CT attenuation Thrombectomy for stroke at 6 to 16 hours with selection by perfu- are associated with intervention time and rate of secondary embo- sion imaging. N Engl J Med 2018;378:708–18 CrossRef Medline lism. Cerebrovasc Dis 2017:344–50 CrossRef Medline 43. Nogueira RG, Jadhav AP, Haussen DC, et al; DAWN Trial Investigators. 25. Kaesmacher J, Boeckh-Behrens T, Simon S, et al. Risk of thrombus Thrombectomy 6 to 24 hours after stroke with a mismatch between fragmentation during endovascular stroke treatment. AJNR Am J deficit and infarct. N Engl J Med 2018;378:11–21 CrossRef Medline Neuroradiol 2017;38:991–98 CrossRef Medline 44. Adams HP, Bendixen BH, Kappelle LJ, et al. Classification of subtype 26. Gunning GM, McArdle K, Mirza M, et al. Clot friction variation with of acute ischemic stroke. Definitions for use in a multicenter clini- fibrin content; implications for resistance to thrombectomy. J Neu- cal trial. TOAST. Trial of Org 10172 in Acute Stroke Treatment. rointerv Surg 2018;10:34–38 CrossRef Medline Stroke. 1993;24:35–41 CrossRef

AJNR Am J Neuroradiol 39:1848–53 Oct 2018 www.ajnr.org 1853 ORIGINAL RESEARCH INTERVENTIONAL

Thrombus Permeability on Dynamic CTA Predicts Good Outcome after Reperfusion Therapy

X Z. Chen, X F. Shi, X X. Gong, X R. Zhang, X W. Zhong, X R. Zhang, X Y. Zhou, and X M. Lou

ABSTRACT BACKGROUND AND PURPOSE: Thrombus permeability assessed on conventional CTA is associated with neurologic outcome in patients with acute ischemic stroke. We aimed to investigate whether dynamic CTA can improve the accuracy of thrombus permeability assess- ment and its predictive value for outcome.

MATERIALS AND METHODS: We reviewed consecutive patients with acute ischemic stroke who had occlusion of the M1 segment of the middle artery cerebral artery and underwent pretreatment perfusion CT. Thrombus permeability, determined by thrombus attenuation

increase (TAI), was assessed on 26-phase dynamic CTA derived from perfusion CT. TAImax was defined as the maximum TAI among phases; Յ TAIpeak, as TAI of peak arterial phase; TAIcon, as TAI on phase 13. Good outcome was defined as a 3-month mRS score of 2.

RESULTS: One hundred four patients were enrolled in the final analysis. The median TAImax, TAIpeak, and TAIcon were 30.1 HU (interquartile range, 13.0–50.2 HU), 9.5 HU (interquartile range, Ϫ1.6–28.7 HU), and 6.6 HU (interquartile range, Ϫ5.1–24.4 HU), respectively. Multivariable ϭ ϭ ϭ ϭ regression analyses showed that TAImax (OR 1.027; 95% CI, 1.007–1.048; P .008), TAIpeak (OR 1.029; 95% CI, 1.005–1.054; P .020), and ϭ ϭ TAIcon (OR 1.026; 95% CI, 1.002–1.051; P .037) were independently associated with good outcome. The areas under the ROC curve

of TAImax, TAIpeak, and TAIcon in predicting good outcome were 0.734, 0.701, and 0.658, respectively. CONCLUSIONS: Thrombus permeability assessed on dynamic CTA could be a better predictor of outcome after reperfusion therapy than that assessed on conventional single-phase CTA.

ABBREVIATIONS: AIS ϭ acute ischemic stroke; AUC ϭ area under the curve; IQR ϭ interquartile range; PH ϭ parenchymal hemorrhage; ONT ϭ onset to ϭ ϭ ϭ ϭ intravenous thrombolysis; ROC receiver operating characteristic; TAI thrombus attenuation increase; TAIcon TAI on phase 13; TAImax the maximum TAI among ϭ phases; TAIpeak TAI of peak arterial phase

erebral large-artery occlusion accounts for about one-third of cause improved neurologic outcome is the goal for treatment. Cacute ischemic strokes (AISs), which may cause severe dis- Thrombus characteristics on admission imaging, such as clot ability and high mortality rates.1-6 Effort has been made on neu- length, density, and location, may have the potential to predict the roimaging to predict the outcome of acute ischemic stroke be- outcome of patients with AIS under different treatments.7 In patients with AIS, the thrombus is commonly regarded as a cork where the artery is completely occluded.8 The permeability Received May 17, 2018; accepted after revision July 17. of thrombi has been well-characterized in preclinical studies.9-11 From the Department of Neurology (Z.C., F.S., X.G., R.Z., W.Z., R.Z., Y.Z., M.L.), Sec- ond Affiliated Hospital of Zhejiang University, School of Medicine, Hangzhou, Chi- It has been suggested that thrombus permeability might be related na; and Zhejiang University Brain Research Institute (M.L.), Hangzhou, Zhejiang, to the physical porosity of thrombus and might reflect the ability China. of soluble molecules to move within the gaps among adjacent Z. Chen and F. Shi contributed equally to this work. platelets, fibrin filaments, and red blood cells.8,9,12,13 Preclinical This study was supported by the National Key Research and Development Pro- gram of China (2016YFC1301500), the National Natural Science Foundation of China studies have demonstrated that high permeability of the throm- (81622017, 81471170, and 81601017), Fundamental Research Funds for the Central Uni- versities (2017XZZX002-09), and Zhejiang Provincial Natural Science bus within the occluded artery allowed residual blood to flow Foundation (LQ16H090003). through the thrombus,9,10 which may have a positive effect on Please address correspondence to Min Lou, MD, PhD, Department of Neurology, neurologic outcome after acute ischemia due to the compensating the Second Affiliated Hospital of Zhejiang University, School of Medicine, No. 88 Jiefang Rd, Hangzhou, China, 310009; e-mail: [email protected] oxygenation of brain tissue distal to the occluded artery. Recently, Indicates open access to non-subscribers at www.ajnr.org thrombus attenuation increase (TAI) was used to assess the Indicates article with supplemental on-line table. thrombus permeability on conventional single-phase CT angiog- http://dx.doi.org/10.3174/ajnr.A5785 raphy and noncontrast CT. A pervious thrombus with a high TAI

1854 Chen Oct 2018 www.ajnr.org was found to be associated with better neurologic outcome after injection was used at a flow rate of 6 mL/s, followed by a 20-mL reperfusion therapy in patients with AIS.8,12 saline chaser at 6 mL/s. The enhancement of arteries on conventional single-phase CTA is affected by the acquisition time because the contrast filling Imaging Analysis of the occlusion site is delayed compared with that in the normal The Arterial Occlusive Lesion Scale was used to assess recanaliza- condition.14 Therefore, the permeability of the thrombi may be tion on 24-hour MRA or CTA (grade 0: complete occlusion of the underestimated if single-phase CTA is acquired before contrast target artery; grade 1: incomplete occlusion or partial local recan- completely penetrates the thrombus. Compared with single- alization at the target artery with no distal flow; grade 2: incom- phase CTA, dynamically acquired CTA provides a broad temporal plete occlusion or partial local recanalization at the target artery coverage that spans from unenhanced through to the arterial and, with any distal flow; grade 3: complete recanalization and resto- subsequently, venous phases. In addition, previous studies have ration of the target artery with any distal flow).15 Recanalization demonstrated that dynamic CTA can better characterize the in- and no recanalization were defined as arterial occlusive lesion tracranial thrombus burden than single-phase CTA.14 grades 2–3 and 0–1, respectively. Hemorrhagic transformation We thus hypothesized that dynamic CTA could overcome the was assessed on 24-hour SWI or NCCT according to the second timing limitation and improve the assessment accuracy of throm- European-Australasian Acute Stroke Study (ECASS II), including bus permeability. We then assessed thrombus permeability on hemorrhagic infarction and parenchymal hemorrhage (PH).16 26-phase dynamic CTA derived from CT perfusion and aimed to Previously validated thresholds were applied to measure the base- investigate its predictive value for neurologic outcome in patients line hypoperfusion volume (time-to-maximum Ͼ 6 seconds)17 with AIS after reperfusion therapy. and infarct core volume (relative cerebral blood flow of Ͻ30%) on CTP.18 MATERIALS AND METHODS Patient Inclusion Thrombus Permeability Assessment We retrospectively reviewed our prospectively collected data base Thrombus permeability assessments were conducted on com- for consecutive patients with AIS who received intravenous mercial software (MIStar; Apollo Medical Imaging Technology, thrombolysis with or without endovascular thrombectomy from Melbourne, Australia). The 26-phase dynamic CTA with 4.5- May 2009 to February 2017. Then, we enrolled patients who ful- mm-thick maximum intensity projection was reconstructed from filled the following criteria: 1) They had a diagnosis of AIS con- CTP source images. The proximal artery of the contralateral firmed by diffusion-weighted imaging or CT at 24 hours after hemisphere was selected to generate the arterial input function symptom onset, 2) underwent CTP within 8 hours after stroke curve. The measurements of TAI were performed as described 8,12,19 onset, 3) had occlusion of the M1 segment of the middle cerebral previously. Three ROIs with a radius of 1 mm were placed artery without involvement of the internal carotid artery, 4) un- on the thrombus by 2 experienced neuroradiologists blinded to derwent CTA or time-of-flight MR angiography at 24 hours after the patients’ information, with rater discrepancies settled by con- treatment, and 5) underwent follow-up NCCT or susceptibility- sensus discussion; the mean attenuation on each phase was calcu- weighted imaging at 24 hours after treatment. We excluded pa- lated. The ROIs might partially overlap each other in case of a tients who had poor image quality due to motion artifacts or small thrombus. The mean attenuation of the thrombus on phase incomplete consecutive acquisitions. 1 was set as the reference value. The TAI of phase 2 to phase 26 was defined as the increase of the mean attenuation of the thrombus Ethics Statement from phase 1 to each phase, respectively. Three parameters of TAI,

Ethical approval was obtained from the human ethics committee including TAImax, TAIpeak, and TAIcon, were used for the final of our center. The clinical investigation was conducted according analyses (Fig 1). TAImax was defined as the maximal TAI among to the principles expressed in the Declaration of Helsinki. Written 25 phases (phases 2–26). TAIpeak was defined as the TAI of the informed consent was obtained from all patients. arterial peak phase according to the arterial input function curve. The mean time between contrast injection and peak concentra- Imaging Parameters tion on the normal side of the middle cerebral artery was 25 sec-

CTP was performed on a 64-slice CT scanner (Somatom Defini- onds (phase 13) in current study; thus, TAIcon was simulated as tion Flash; Siemens, Erlangen, Germany), including an NCCT TAI on phase 13 of dynamic CTA. scan (120 kV, 320 mA, contiguous 5-mm axial slices, 7-second acquisition time) and volume CTP (100 mm in the z-axis, 4-sec- Clinical Data ond delay after the start of contrast material bolus injection, 74.5- We reviewed demographic, clinical, and radiologic data, includ- second total imaging duration, 80 kV, 120 mA, 1.5-mm slice ing age; sex; prior antiplatelet use; risk factors such as smoking, thickness, 32 ϫ 1.2 mm collimation). Volume CTP consisted of hypertension, diabetes mellitus, hyperlipidemia, smoking, and 26 consecutive spiral acquisitions of the brain. All 26 scans were atrial fibrillation; time interval from stroke onset to intravenous divided into 4 parts: 1) two scans with 3-second cycle time, 2) thrombolysis (ONT); National Institutes of Health Stroke Scale fifteen scans with 1.5-second cycle time, 3) four scans with 3-sec- score on admission; baseline hypoperfusion volume and infarct ond cycle time, and 4) five scans with 6-second cycle time. Axial core volume on admission; hemorrhagic transformation and slice coverage was 150 mm. A 60-mL bolus of contrast medium recanalization after intravenous thrombolysis; and modified (iopamidol, Imeron; Bracco Sine, Shanghai, China) with a single Rankin Scale score after 3 months. Patients were dichotomized AJNR Am J Neuroradiol 39:1854–59 Oct 2018 www.ajnr.org 1855 continuous variables, as appropriate. The Spearman rank corre- lation test was used to test the association of TAI with clinical and imaging variables. The strength of these associations was com- pared using the area under the receiver operating characteristic (ROC) curve. The ROC-derived optimal cutoff was determined at the maximal Youden Index. Variables with a P value Ͻ .1 in uni- variate analyses were enrolled in the multivariable regression model. A P value Ͻ .05 was considered statistically significant. All statistical analyses were performed with SPSS, Version 22.0 (IBM, Armonk, New York).

RESULTS Overall Characteristics One hundred four patients with AIS with MCA-M1 were enrolled in the final analysis. The mean age was 68 Ϯ 14 years, the mean NIHSS score was 14 Ϯ 6 on admission, and 61 (58.7%) patients were male. The median ONT was 215 (IQR, 140–295) minutes.

The median TAImax, TAIpeak and TAIcon were 30.1 HU (IQR, 13.0–50.2 HU), 9.5 HU (IQR, Ϫ1.6–28.7 HU), and 6.6 HU (IQR, Ϫ5.1–24.4 HU), respectively.

Association of TAI with Radiologic and Neurologic Outcome As shown in Table 1, 44 (42.3%) patients achieved good outcome.

Patients with good outcome had higher TAImax, TAIpeak, and Ͻ Ͻ TAIcon than those with poor outcome (P .001, P .001, and P ϭ .006, respectively). Patients with good outcome were younger (P ϭ .005) and had a lower baseline NIHSS score (P Ͻ .001), lower baseline infarct core volume (P ϭ .001), lower baseline hypoperfusion volume (P ϭ .004), and higher recanalization rates (84% versus 64%, P ϭ .023) compared with those with poor outcome. After we adjusted for age, baseline NIHSS score, atrial fibrilla- ϭ tion, ONT, baseline infarct core volume, TAImax (OR 1.027; ϭ ϭ 95% CI, 1.007–1.048; P .008), TAIpeak (OR 1.029; 95% CI, ϭ ϭ 1.005–1.054; P .020), and TAIcon (OR 1.026; 95% CI, 1.002– 1.051; P ϭ .037) were independently associated with good out-

come, respectively. Patients with PH had lower TAImax and

TAIpeak than those without PH, while TAIcon did not show a sig- ϭ FIG 1. Illustration of thrombus attenuation increase assessment with nificant difference between patients with and without PH (P 3 ROIs (black circles) measuring the average thrombus attenuations .308). on a left M1 segment of the middle cerebral artery on motion-cor- The ROC curves of TAI , TAI , and TAI in predicting rected dynamic CTA. The arterial input function (AIF) curve for all 26 max peak con phases was generated (A) and the thrombus permeability was as- good outcome are shown in Fig 2, and the areas under the curve sessed on the CTA images with maximum intensity projection (MIP) in (AUCs) were 0.734, 0.701, and 0.658, respectively. The AUCs of all 26 phases (B), at the mean time delay following intravenous con- each single-phase CTA derived from CTP are shown in On-line trast injection in the current study (phase 13) (C), at the arterial peak phase (phase 18) (D), and at the phase with the maximum value (phase Table. The optimal cutoffs were 30.9, 9.0, and 14.6 HU for 22) (E) of the mean attenuation increase. TAImax, TAIpeak and TAIcon, respectively. Patients were then di- chotomized into a pervious thrombus group and an impervious into good (mRS score of Յ2) and poor outcome (mRS score thrombus group according to the optimal cutoff values of TAImax, of Ͼ2) at 90 days. TAIpeak, and TAIcon, respectively. After we adjusted for age, base- line NIHSS score, atrial fibrillation, ONT, and baseline infarct Statistical Analysis core volume, patients with a pervious thrombus had a higher rate Mean with SD, medians with interquartile range (IQR), and per- of good outcome than those with an impervious thrombus when ϭ centages were used to describe the distribution of continuous and dichotomized by TAImax (OR 3.957; 95% CI, 1.463–10.705; ϭ ϭ ϭ categoric variables. The Fisher exact test was used to compare P .007) and TAIpeak (OR 2.887; 95% CI, 1.075–7.754; P categoric variables among groups, whereas the independent-sam- .035) (Table 2). However, when dichotomized by TAIcon, patients ples 2-tailed t test or the Mann-Whitney U test was used for the with a pervious thrombus were not independently associated with 1856 Chen Oct 2018 www.ajnr.org Table 1: Comparison of baseline characteristics according to clinical outcome Poor Outcome Good Outcome (mRS>2)(n = 60) (mRS≤2)(n = 44) P Value Age (yr) 74 (65–81) 66 (58–76) .005 Male (No.) (%) 33 (55.0%) 28 (63.6%) .424 Baseline NIHSS (mean) 16 Ϯ 511Ϯ 6 Ͻ.001 Onset-to-needle time (mean) (min) 230 (168–297) 207 (119–259) .059 Endovascular thrombectomy (No.) (%) 22 (36.7%) 16 (36.4%) 1.000 Prior antiplatelet usage (No.) (%) 12 (20.0%) 9 (20.5%) 1.000 Risk factors Smoking (No.) (%) 18 (30.0%) 13 (29.5%) 1.000 Hypertension (No.) (%) 35 (58.3%) 22 (50.0%) .431 Diabetes mellitus (No.) (%) 16 (26.7%) 6 (13.6%) .146 Hyperlipidemia (No.) (%) 22 (36.7%) 21 (47.7%) .315 History of stroke/TIA (No.) (%) 11 (18.3%) 5 (11.4%) .415 Atrial fibrillation (No.) (%) 38 (63.3%) 20 (45.5%) .076 Radiologic data Baseline infarct core volume (mean) (mL) 69.27 Ϯ 49.51 40.72 Ϯ 30.91 .001 Baseline hypoperfusion volume (mean) (mL) 134.28 Ϯ 61.07 98.22 Ϯ 60.43 .004 Ͻ TAImax (median) (IQR) (HU) 18.0 (10.0–38.7) 39.9 (26.1–73.3) .001 Ϫ Ͻ TAIpeak (median) (IQR) (HU) 4.6 ( 4.2–17.5) 18.7 (5.9–43.0) .001 Ϫ Ϫ TAIcon (median) (IQR) (HU) 2.2 ( 5.5–12.4) 10.4 ( 2.0–36.7) .006 Ͻ Thrombus perviousness by dichotomized TAImax (No.) (%) 18 (30.0%) 31 (70.5%) .001 Thrombus perviousness by dichotomized TAIpeak (No.) (%) 22 (36.7%) 31 (70.5%) .001 Thrombus perviousness by dichotomized TAIcon (No.) (%) 13 (21.7%) 21 (47.7%) .005

Table 2: Multivariable regression for good outcome P OR 95% CI Value Model 1 Age 0.966 0.926–1.008 .114 Baseline NIHSS 0.911 0.823–1.008 .071 Atrial fibrillation 0.872 0.298–2.555 .803 ONT 0.992 0.987–0.997 .004 Baseline infarct core volume 0.986 0.972–1.001 .061 Thrombus perviousness by 3.957 1.463–10.705 .007

dichotomized TAImax Model 2

FIG 2. ROC curve of thrombus attenuation increase for TAImax, Age 0.963 0.924–1.004 .074 TAIpeak, and TAIcon, respectively. Baseline NIHSS 0.912 0.826–1.006 .066 Atrial fibrillation 0.736 0.260–2.082 .564 ϭ ϭ ONT 0.992 0.987–0.997 .003 good outcome (OR 2.323; 95% CI, 0.815–6.621; P .115) Baseline infarct core volume 0.988 0.974–1.002 .093 (Table 2). Thrombus perviousness by 2.887 1.075–7.754 .035

dichotomized TAIpeak DISCUSSION Model 3 In the present study, we found that high thrombus permeability Age 0.965 0.926–1.004 .080 Baseline NIHSS 0.917 0.827–1.016 .098 on dynamic CTA is an independent predictor of good outcome in Atrial fibrillation 0.751 0.269–2.102 .586 patients with AIS. Most interesting, TAImax and TAIpeak derived ONT 0.992 0.987–0.997 .003 from CTP data are better imaging markers than TAIcon for pre- Baseline infarct core volume 0.986 0.972–1.000 .050 dicting neurologic outcome. Thrombus perviousness by 2.323 0.815–6.621 .115 dichotomized TAI Our finding that the permeable thrombus was strongly asso- con ciated with good outcome is consistent with previous studies us- ing conventional single-phase CTA.8,12 These studies also re- come. Usually, single-phase CTA is used to assess thrombus per- ported that the association between permeable thrombi and good meability by calculating the attenuation increase before and after outcome was independent of recanalization status.8 High throm- contrast penetrating the thrombus. The thrombus permeability bus permeability promotes anterograde filling of blood, and, on single-phase CTA might be underestimated due to timing therefore, less severely ischemic brain tissue. limitations, hemodynamic restriction, or even pseudo-occlu- 14,20 Most important, our study revealed that the AUCs of TAImax sion. It was reported that delayed phases after the arterial peak and TAIpeak for predicting good outcome (0.734 and 0.701) were phase on dynamic CTA could provide better thrombus depiction ϭ higher than those of TAIcon (AUC 0.658) and previously re- and prognostic information and thus might affect treatment de- ported values (AUC ϭ 0.67) using conventional single-phase cisions in the acute setting.14,21 A previous study using 3-phase CTA; these findings indicate that thrombus permeability assessed CTA found that arterial phase CTA was superior to venous phase on dynamic CTA might be a better biomarker for predicting out- CTA (8 seconds after arterial phase CTA) or delayed-phase CTA AJNR Am J Neuroradiol 39:1854–59 Oct 2018 www.ajnr.org 1857 (16 seconds after arterial phase CTA) to assess the TAI.22 How- sion therapy than that assessed on conventional single-phase ever, 3-phase CTA did not cover enough time points, and the CTA. optimal phase could be located between the arterial phase and venous phase. Thus, we believe dynamic CTA derived from CTP Disclosures: Zhicai Chen—RELATED: Grant: National Natural Science Foundation of China (81601017) and Zhejiang Provincial Natural Science Foundation of China was better for evaluating clot perviousness. Moreover, the char- (LQ16H090003).* Min Lou—RELATED: Grant: National Natural Science Foundation acteristics of the arterial input function curve were affected by of China (81622017 and 81471170), Comments: National Natural Science Foundation of cardiac function, vascular curvature, and vascular stenosis. Thus, China (81622017 and 81471170), the National Key Research and Development Program of China (2016YFC1301500), and Fundamental Research Funds for the Central Univer- parameters based on individual hemodynamic characteristics de- sities (2017XZZX002-09).* *Money paid to the institution. rived from CTP data can better represent thrombus permeability than conventional single-phase CTA. Furthermore, the evalua- REFERENCES tion of dynamic CTA has no additional x-ray exposure and con- 1. Gorelick PB, Wong KS, Bae HJ, et al. Large artery intracranial occlu- trast medium usage because it is based on CTP, which can also sive disease: a large worldwide burden but a relatively neglected provide information about infarct core and penumbra. frontier. Stroke 2008;39:2396–99 CrossRef Medline To the best of our knowledge, we are the first to have found 2. Saver JL, Jahan R, Levy EI, et al. Solitaire flow restoration device that low thrombus permeability was associated with a high rate of versus the Merci Retriever in patients with acute ischaemic stroke (SWIFT): a randomised, parallel-group, non-inferiority trial. Lan- PH after reperfusion therapy. It has been shown that compensat- cet 2012;380:1241–49 CrossRef Medline ing oxygenation to ischemic brain tissue could reduce hemor- 3. van Seeters T, Biessels GJ, Kappelle LJ, et al; Dutch Acute Stroke Study rhagic transformation after thrombolysis.23 Therefore, we postu- (DUST) investigators. The prognostic value of CT angiography and late that permeable thrombi may enhance tissue oxygenation CT perfusion in acute ischemic stroke. Cerebrovasc Dis 2015;40: 258–69CrossRef Medline distal to the occlusion due to increased passage of blood, which 4. Malhotra K, Gornbein J, Saver JL. Ischemic strokes due to large- reduces the rate of PH. Additionally, the negative correlation be- vessel occlusions contribute disproportionately to stroke-related tween TAI and baseline NIHSS score, infarct core volume, and dependence and death: a review. Front Neurol 2017;8:651 CrossRef hypoperfusion volume might imply less severity of ischemia in Medline patients with pervious thrombi, which could be related to lower 5. Lima FO, Silva GS, Furie KL, et al. Field assessment stroke triage for emergency destination: a simple and accurate prehospital rates of PH. scale to detect large vessel occlusion strokes. Stroke 2016;47: We did not find a significant association between thrombus 1997–2002 CrossRef Medline permeability and recanalization, inconsistent with previous stud- 6. Beumer D, Mulder MJ, Saiedie G, et al. Occurrence of intracranial ies,8,12 which may be due to the inclusion of patients who received large vessel occlusion in consecutive, non-referred patients with acute ischemic stroke. Neurovascular Imaging 2016;2:11–975 endovascular thrombectomy because endovascular thrombec- CrossRef tomy usually has higher recanalization with mechanical retrieval 7. Yoo AJ, Pulli B, Gonzalez RG. Imaging-based treatment selection devices. Another reason may be due to the assessment of recana- for intravenous and intra-arterial stroke therapies: a comprehen- lization at 24 hours after stroke in our study, while previous stud- sive review. Expert Rev Cardiovasc Ther 2011;9:857–76 CrossRef Medline ies underwent follow-up imaging at 3 days or even later.8,12 Be- 8. Santos EM, Marquering HA, den Blanken MD, et al; MR CLEAN sides, the sample size of the current study was relatively small, and Investigators. Thrombus permeability is associated with improved all patients were from a single center. Thus, further investigations functional outcome and recanalization in patients with ischemic in larger and multicenter cohorts are needed. stroke. Stroke 2016;47:732–41 CrossRef Medline This study has several limitations. First, this was a retrospec- 9. Voronov RS, Stalker TJ, Brass LF, et al. Simulation of intrathrombus fluid and solute transport using in vivo clot structures with single tive design and might have a potential risk of selection bias, platelet resolution. Ann Biomed Eng 2013;41:1297–307 CrossRef though the data were prospectively collected using the same Medline stroke registry and CTP protocol. Second, although the ROI 10. Laurens N, Koolwijk P, de Maat MP. Fibrin structure and wound placement was performed by 2 experts blinded to the patients’ healing. J Thromb Haemost 2006;4:932–39 CrossRef Medline 11. Brass LF, Wannemacher KM, Ma P, et al. Regulating thrombus information with rater discrepancies settled by consensus discus- growth and stability to achieve an optimal response to injury. J sion, variation still exists within a thrombus. Therefore, the mean Thromb Haemost 2011;9(Suppl 1):66–75 CrossRef Medline attenuation of the 3 ROIs selected may not represent the entire 12. Santos EM, Dankbaar JW, Treurniet KM, et al; DUST Investigators. thrombus. Third, the overlapping vessels, calcified clots, and vas- Permeable thrombi are associated with higher intravenous recom- binant tissue-type plasminogen activator treatment success in pa- cular calcification might influence the attenuation measure- tients with acute ischemic stroke. Stroke 2016;47:2058–65 CrossRef ments, though the ROIs were placed very carefully to avoid them. Medline Fourth, further research about blood flow status on DSA is needed 13. Welsh JD, Stalker TJ, Voronov R, et al. A systems approach to to explain the association between thrombus perviousness and hemostasis: 1. The interdependence of thrombus architecture and good outcome. Finally, the sample size was modest, and all the agonist movements in the gaps between platelets. Blood 2014;124: 1808–15 CrossRef Medline samples were from a single center. Confirmation and extension of 14. Fro¨lich AM, Schrader D, Klotz E, et al. 4D CT angiography more these findings in larger and multicenter cohorts are needed. closely defines intracranial thrombus burden than single-phase CT angiography. AJNR Am J Neuroradiol 2013;34:1908–13 CrossRef Medline CONCLUSIONS 15. Khatri P, Neff J, Broderick JP, et al; IMS-I Investigators. Revasculariza- Our study suggests that thrombus permeability assessed on dy- tion end points in stroke interventional trials: recanalization versus namic CTA could be a better predictor of outcome after reperfu- reperfusion in IMS-I. Stroke 2005;36:2400–03 CrossRef Medline 1858 Chen Oct 2018 www.ajnr.org 16. Hacke W, Kaste M, Fieschi C, et al. Randomised double-blind pla- 20. Marquering HA, Nederkoorn PJ, Beenen LF, et al. Carotid pseudo- cebo-controlled trial of thrombolytic therapy with intravenous al- occlusion on CTA in patients with acute ischemic stroke: a concern- teplase in acute ischaemic stroke (ECASS II). Lancet 1998;352: ing observation. Clin Neurol Neurosurg 2013;115:1591–94 CrossRef 1245–51 CrossRef Medline 17. Lin L, Bivard A, Levi CR, et al. Comparison of computed tomo- 21. Byrne D, Sugrue G, Stanley E, et al. Improved detection of anterior graphic and magnetic resonance perfusion measurements in acute circulation occlusions: the “delayed vessel sign” on multiphase CT ischemic stroke: back-to-back quantitative analysis. Stroke 2014;45: angiography. AJNR Am J Neuroradiol 2017;38:1911–16 CrossRef 1727–32 CrossRef Medline Medline 18. Campbell BC, Christensen S, Levi CR, et al. Cerebral blood flow is 22. Santos EM, d’Esterre CD, Treurniet KM, et al; PRove-IT investi- the optimal CT perfusion parameter for assessing infarct core. gators. Added value of multiphase CTA imaging for thrombus Stroke 2011;42:3435–40 CrossRef Medline perviousness assessment. Neuroradiology 2018;60:71–79 CrossRef 19. Santos EM, Yoo AJ, Beenen LF, et al; MR CLEAN investigators. Ob- Medline server variability of absolute and relative thrombus density mea- 23. Sun L, Zhou W, Mueller C, et al. Oxygen therapy reduces secondary surements in patients with acute ischemic stroke. Neuroradiology hemorrhage after thrombolysis in thromboembolic cerebral isch- 2016;58:133–39 CrossRef Medline emia. J Cereb Blood Flow Metab 2010;30:1651–60 CrossRef Medline

AJNR Am J Neuroradiol 39:1854–59 Oct 2018 www.ajnr.org 1859 ORIGINAL RESEARCH INTERVENTIONAL

Identification of Hostile Hemodynamics and Geometries of Cerebral Aneurysms: A Case-Control Study

X B.J. Chung, X F. Mut, X C.M. Putman, X F. Hamzei-Sichani, X W. Brinjikji, X D. Kallmes, X C.M. Jimenez, and X J.R. Cebral

ABSTRACT BACKGROUND AND PURPOSE: Hostile hemodynamic conditions and geometries are thought to predispose aneurysms for instability and rupture. This study compares stable, unstable, and ruptured aneurysms while controlling for location and patient characteristics.

MATERIALS AND METHODS: The hemodynamics and geometries of 165 stable, 65 unstable, and 554 ruptured aneurysms were compared. Hemodynamics was modeled using image-based computational fluid dynamics. Case-control pairs were selected matching aneurysm location, patient age, and sex. Paired Wilcoxon tests were used to compare hemodynamic and geometric variables among different aneurysm groups. The pairing was repeated 100 times, and the combined P values were calculated and adjusted for multiple testing.

RESULTS: Ruptured aneurysms had lower minimum wall shear stress (P ϭ .03), higher maximum wall shear stress (P ϭ .03), more concen- trated (P ϭ .03) and mean oscillatory shear stress (P ϭ .03), higher maximum velocity (P ϭ .03), and more complex flows (vortex core-line length, P ϭ .03) than stable aneurysms. Similarly, unstable aneurysms had more concentrated shear stress (P ϭ .04) and more complex flows (vortex core-line length, P ϭ .04) than stable aneurysms. Compared with stable aneurysms, ruptured aneurysms were larger (size ratio, aneurysm size/vessel size, P ϭ .03), more elongated (aspect ratio, P ϭ .03), and irregular (nonsphericity index, P ϭ .03). Similarly, unstable aneurysms were larger (size ratio, P ϭ .04), more elongated (aspect ratio, P ϭ .04), and irregular (bulge location, P ϭ .04; area-weighted Gaussian curvature; P ϭ .04) than stable aneurysms. No significant differences were found between unstable and ruptured aneurysms.

CONCLUSIONS: Unstable and ruptured aneurysms have more complex flows with concentrated wall shear stress and are larger, more elongated, and irregular than stable aneurysms, independent of aneurysm location and patient sex and age.

ABBREVIATIONS: AR ϭ aspect ratio; Asize ϭaneurysm maximum size; BL ϭ bulge location; CORELEN ϭ vortex core-line length, flow complexity; CP ϭ conicity parameter; GAA ϭ area-weighted Gaussian curvature; LSA ϭ percentage area under low WSS; max ϭ maximum; min ϭ minimum; NSI ϭ nonsphericity index; OSI ϭ oscillatory shear stress; SCI ϭ concentrated shear stress; SizeR ϭ size ratio (aneurysm size/vessel size); Vmax ϭ maximum velocity; VOR ϭ volume-to-ostium ratio; WSS ϭ wall shear stress

any factors appear to be related to aneurysm rupture risk: lines for intracranial aneurysm management recommend that in Mlocation, size, morphology, hemodynamics, perianeurys- addition to the size and location of the aneurysm and patient’s age mal environment, and patient history, among others.1 It is well- and health status, it may be reasonable to consider the morpho- understood that aneurysm rupture-risk estimation is based on the logic and hemodynamic characteristics of the aneurysm when dis- combination of such factors and is a complex process.2 The Amer- cussing its rupture risk.3 Thus, identification of “hostile” hemo- ican Heart Association and American Stroke Association guide- dynamic conditions and geometries that predispose aneurysms to instability and rupture is important to discriminate high- and low-risk aneurysms and guide the management strategy for each 4 Received October 16, 2017; accepted after revision June 27, 2018. individual patient. Additionally, understanding adverse condi- From the Bioengineering and Mechanical Engineering Departments (B.J.C., F.M., tions and their connections to the underlying mechanisms of an- J.R.C.), Volgenau School of Engineering George Mason University, Fairfax, Virginia; eurysm progression, instability, and rupture is valuable for de- Interventional Neuroradiology Unit (C.M.P.), Inova Fairfax Hospital, Falls Church, 5 Virginia; Department of Neurological Surgery (F.H.-S.), University of Massachusetts, signing new therapeutic strategies targeting those mechanisms. Worcester, Massachusetts; Department of Radiology (W.B., D.K.), Mayo Clinic, Several previous studies have analyzed possible associations Rochester, Minnesota; and Neurosurgery Department (C.M.J.), University of Antio- 6-9 quia, Medellin, Colombia. between hemodynamics and rupture and between aneurysm This work was partially supported by the National Institutes of Health–National Institute of Neurological Disorders and Stroke grant No. R21NS094780. Indicates open access to non-subscribers at www.ajnr.org Please address correspondence to Juan R. Cebral, PhD, Bioengineering Department, Indicates article with supplemental on-line tables. Volgenau School of Engineering, George Mason University, 4400 University Dr, MSN 2A1, Fairfax, VA 22030; e-mail: [email protected] http://dx.doi.org/10.3174/ajnr.A5764

1860 Chung Oct 2018 www.ajnr.org morphology and rupture.1,10 One of the most important limita- rigid, and blood viscosity was approximated as Newtonian. The tions of most cross-sectional studies comparing ruptured and un- unsteady incompressible Navier-Stokes equations were numeri- ruptured aneurysms is the inability to discriminate stable and cally solved with a finite-element code developed in-house.18 All unstable unruptured aneurysms.8,9 On the other hand, the most simulations were run in parallel shared-memory computers for 2 important limitation of most longitudinal studies is the small cardiac cycles. Data from the second cycle were used to quantita- sample size drawn from a highly selected population.11-13 Addi- tively characterize the aneurysm hemodynamic environment by tionally, only a few studies control for aneurysm location and/or computing several flow variables over the aneurysm cavity, the patient characteristics.14 These limitations may cause confusion aneurysm orifice, and the aneurysm surface.19 Additionally, sev- and apparent conflicts among the results of different studies.15 eral geometric variables were also computed to characterize the The goal of this study was to compare the hemodynamics and aneurysm morphology.20 A list of the variables considered is pro- geometries of stable, unstable, and ruptured aneurysms while vided in On-line Table 2. controlling for location and patient characteristics. Analysis MATERIALS AND METHODS Case-control studies are a common and efficient means of study- Data ing rare diseases with long latency periods, such as ruptured/ Our data base contains Ͼ2000 cerebral aneurysms imaged with unruptured aneurysms.21 Matching of cases and controls is fre- 3D rotational angiography and basic information including an- quently used to control the effects of known potential confound- eurysm status, location, size, patient age, and sex. The main con- ing variables. The analysis of matched data requires specific sta- tributors have been Inova Fairfax Hospital (Virginia), Mt. Sinai tistical methods. The nonparametric Wilcoxon signed rank test Medical Center (New York), and the Mayo Clinic (Minnesota). for measured outcomes with 1:1 matching was used in the follow- Age was divided into 3 groups: young (younger than 40 years), ing 4 case-control studies: 1) stable-versus-ruptured aneurysms, middle age (40–60 years), and old (older than 60 years). The 2) stable-versus-unstable aneurysms, 3) unstable-versus-rup- images and data have been anonymized, and the study has been tured aneurysms, and 4) unruptured-versus-ruptured aneu- approved by the institutional review board of George Mason Uni- rysms. In each study, case-control pairs were created and the versity. In the present study, a subset of 784 aneurysms was in- mean values of hemodynamic and geometric variables of the 2 cluded, subdivided into the following subgroups (the distribution groups were statistically compared using a paired Wilcoxon test. of aneurysms by location is presented in On-line Table 1): The matching of cases and controls in each study was performed 1) Stable Aneurysms. Untreated unruptured aneurysms followed as detailed below: longitudinally without noticeable enlargement, shape change, 1) Stable-versus-Ruptured Aneurysms. In this case, the controls new symptoms, or rupture for at least 1 year were considered (stable aneurysms) were fewer than the cases (ruptured aneu- stable. A total of 165 stable aneurysms in 85 patients were rysms); therefore, the matching was performed as follows: For considered. each stable aneurysm, a list of ruptured aneurysms with matching 2) Unstable Aneurysms. Unruptured aneurysms were considered location, sex, and age groups was created, and one of these match- unstable if they either enlarged during follow-up (as determined ing aneurysms was randomly selected. A total of 134 aneurysm by Sforza et al11) or presented with symptoms such as cranial pairs were thus created. nerve palsy indicative of aneurysm instability. A total of 65 2) Stable-versus-Unstable Aneurysms. In this study, the cases aneurysms in 55 patients were classified as unstable, with 53 (unstable aneurysms) were fewer than the controls (stable aneu- aneurysms classified as growing and 12 with instabilities re- rysms); therefore, for each unstable aneurysm, a stable one was lated to symptoms. randomly selected from the list of stable aneurysms at the same 3) Ruptured Aneurysms. The ruptured aneurysms group in- location with matching sex and age groups. Sixty-five aneurysm cluded all aneurysms confirmed as the source of subarachnoid pairs were created. hemorrhage. A total of 554 ruptured aneurysms in 501 patients 3) Unstable-versus-Ruptured Aneurysms. In this study, for each were considered. unstable aneurysm, a ruptured one was randomly selected from the list of ruptured aneurysms matching location, sex, and age. Models Sixty pairs were created. Computational fluid dynamics models of all 2000 aneurysms in our data base have been created from the corresponding 3D rota- 4) Unruptured-versus-Ruptured Aneurysms. In this study, the tional angiography images (at baseline for longitudinally followed controls (unruptured aneurysms) were fewer than the cases (rup- aneurysms) using previously described methods.16 These models tured aneurysms). Thus, for each unruptured aneurysm, a rup- include the patient-specific vascular geometry, but because pa- tured one was randomly selected from the list of unruptured an- tient-specific flow conditions were not available, typical pulsatile eurysms at the same location with matching sex and age groups. A flow conditions were derived from phase-contrast MR imaging total of 180 pairs were created. measurements in healthy subjects and scaled with the area of the Once the case-control pairs were created, the mean values of inflow vessel.17 Outflow boundary conditions consistent with the hemodynamic and geometric variables of the 2 groups were flow splits given by the principle of minimal work (Murray law) statistically compared using a paired Wilcoxon test. Because the were prescribed at the outlets. Vessel walls were approximated as pairing processes described above involve random selection of a AJNR Am J Neuroradiol 39:1860–66 Oct 2018 www.ajnr.org 1861 Values of hemodynamic and geometric variables in stable, unstable, and ruptured aneurysmsa Unruptured Unstable Variable Stable Symptoms Growing Ruptured WSSmin 0.8 Ϯ 1.9 0.3 Ϯ 0.5 1.0 Ϯ 4.5 0.4 Ϯ 1.3 WSSmax 223.7 Ϯ 187.5 235.0 Ϯ 101.8 253.7 Ϯ 170.2 377.5 Ϯ 948.8 WSSmean 21.4 Ϯ 18.5 16.6 Ϯ 13.8 22.0 Ϯ 22.1 23.2 Ϯ 28.9 WSSnorm 0.6 Ϯ 0.3 0.4 Ϯ 0.3 0.5 Ϯ 0.3 0.4 Ϯ 0.3 SCI 3.78 Ϯ 4.76 8.52 Ϯ 9.44 5.52 Ϯ 4.53 6.23 Ϯ 6.88 LSA 45.8 Ϯ 33.6 62.9 Ϯ 29.7 58.8 Ϯ 30.9 51.6 Ϯ 33.0 OSImax 0.259 Ϯ 0.137 0.358 Ϯ 0.089 0.303 Ϯ 0.125 0.317 Ϯ 0.114 OSImean 0.012 Ϯ 0.011 0.014 Ϯ 0.008 0.017 Ϯ 0.016 0.017 Ϯ 0.017 Q 0.746 Ϯ 0.805 1.360 Ϯ 1.368 1.081 Ϯ 0.957 0.730 Ϯ 0.807 ICI 0.716 Ϯ 0.763 1.193 Ϯ 1.309 1.088 Ϯ 0.941 0.791 Ϯ 0.764 Vmax 62.9 Ϯ 31.8 70.8 Ϯ 19.6 80.4 Ϯ 50.0 88.8 Ϯ 62.6 VE 9.7 Ϯ 6.4 8.7 Ϯ 5.7 10.1 Ϯ 6.1 9.8 Ϯ 8.1 SR 226.2 Ϯ 176.1 155.0 Ϯ 131.4 202.9 Ϯ 221.2 231.5 Ϯ 260.7 VO 307.7 Ϯ 241.1 218.0 Ϯ 183.7 278.9 Ϯ 279.0 317.4 Ϯ 344.4 CORELEN 1.753 Ϯ 3.313 2.726 Ϯ 1.827 2.504 Ϯ 3.043 2.378 Ϯ 2.593 PODENT 0.176 Ϯ 0.125 0.227 Ϯ 0.115 0.228 Ϯ 0.155 0.207 Ϯ 0.137 Asize 0.687 Ϯ 0.524 1.109 Ϯ 0.603 0.933 Ϯ 0.550 0.770 Ϯ 0.398 Nsize 0.482 Ϯ 0.309 0.690 Ϯ 0.461 0.577 Ϯ 0.311 0.429 Ϯ 0.172 SizeR 1.834 Ϯ 1.395 2.440 Ϯ 1.127 2.533 Ϯ 1.487 2.492 Ϯ 1.373 AR 0.879 Ϯ 0.607 1.308 Ϯ 0.828 1.200 Ϯ 0.831 1.304 Ϯ 0.691 VOR 0.894 Ϯ 3.355 1.220 Ϯ 1.310 1.460 Ϯ 2.906 0.919 Ϯ 1.559 EI 0.273 Ϯ 0.036 0.268 Ϯ 0.021 0.263 Ϯ 0.026 0.273 Ϯ 0.025 NSI 0.202 Ϯ 0.051 0.229 Ϯ 0.080 0.211 Ϯ 0.041 0.246 Ϯ 0.054 BF 1.167 Ϯ 0.465 1.414 Ϯ 0.478 1.377 Ϯ 0.574 1.322 Ϯ 0.441 BL 0.290 Ϯ 0.161 0.431 Ϯ 0.170 0.361 Ϯ 0.144 0.355 Ϯ 0.163 CR 0.762 Ϯ 0.144 0.793 Ϯ 0.083 0.816 Ϯ 0.118 0.785 Ϯ 0.100 CP 0.210 Ϯ 0.161 0.069 Ϯ 0.170 0.139 Ϯ 0.144 0.145 Ϯ 0.163 UI 0.238 Ϯ 0.144 0.207 Ϯ 0.083 0.184 Ϯ 0.118 0.215 Ϯ 0.100 GAA 16.339 Ϯ 17.905 5.861 Ϯ 4.444 7.893 Ϯ 8.034 12.763 Ϯ 13.717 Note:—Q indicates aneurysm flow rate; ICI, inflow concentration index; VE, mean aneurysm velocity; SR, mean aneurysm shear rate; VO, mean aneurysm velocity; PODENT, POD entropy, flow instability; POD, proper orthogonal decomposition; Nsize, neck maximum size; EI, ellipticity index; BF, bottleneck factor; CR, convexity ratio; UI, undulation index; WSSnorm, normalized wall shear stress. a Values are given as mean Ϯ SD. case-control pair from the list of available matches, a bootstrap- ability of these ratios over the 100 random pairings. The mean and ping or random sampling with a replacement approach was SDs of all variables over the 2 groups, along with the correspond- used22 to improve the robustness of the analysis. Specifically, the ing P values, are listed in On-line Table 3. pairing and tests were repeated 100 times, and the P values were These results indicate that compared with stable aneu- 23 combined with the Wilkinson method. A difference was then rysms, after adjustment for multiple testing, ruptured aneu- considered statistically significant if the combined P value was rysms tend to have hemodynamic environments characterized Ͻ .05. The combined P values were then adjusted for multiple by lower minimum wall shear stress (WSSmin, P ϭ .03), higher 24 testing using the false discovery rate method. It was verified that maximum WSS (WSSmax, P ϭ .03), more concentrated shear 100 pairings were enough by performing 200 pairings and verify- stress distributions (SCI, P ϭ .03), more mean oscillatory shear ing that the significance of the comparisons did not change. All stress (OSImean, P ϭ .03), higher maximum velocity (Vmax, statistical analyses were performed by using R scripts (R statistical P ϭ .03), and more complex flow patterns (vortex core-line and computing software; http://www.r-project.org). length [CORELEN], P ϭ .03). Additionally, ruptured aneu- rysms had geometries characterized by larger size (size ratio, RESULTS aneurysm size/vessel size [SizeR], P ϭ .03), more elongated The general hemodynamic and geometric characteristics of the shapes (aspect ratio [AR], P ϭ .03), and more irregular shapes aneurysm groups considered in this study are summarized in the (nonsphericity index [NSI], P ϭ .03). Table. This Table lists the mean values and SDs of each variable over each group. Stable-versus-Unstable Unruptured Aneurysms Stable-versus-Ruptured Aneurysms Comparisons of hemodynamic and geometric variables between The statistically significant differences (P Ͻ .05 before correcting stable and unstable unruptured aneurysms are presented in On- for multiple testing) between stable and ruptured aneurysms are line Table 4 and are summarized in Fig 2. These results indicate summarized in Fig 1. The bars represent the ratio of the mean that unstable aneurysms tend to have more SCI (P ϭ .04) values of the ruptured aneurysm group over the mean values of and more complex flow patterns (CORELEN, P ϭ .04) than stable the stable aneurysm group, and the error bars represent the vari- aneurysms. Unstable aneurysms also tend to have larger areas 1862 Chung Oct 2018 www.ajnr.org FIG 1. Ratios of the mean values of the ruptured aneurysm group over mean values of the stable aneurysm group. Only hemodynamic (dark gray) and geometric (light gray) variables that were significantly different between these groups (P Ͻ .05 before adjusting for multiple testing) are included.

FIG 2. Ratios of mean values of the unstable aneurysm group over mean values of the stable aneurysm group. Only hemodynamic (dark gray) and geometric (light gray) variables that were significantly different (P Ͻ .05 before adjusting for multiple testing) or marginally significant (P Ͻ .1 before adjusting for multiple testing; marked with a plus sign) are included. ICI indicates inflow concentration index; BF, bottleneck factor; CR, convexity ratio. under low WSS (percentage area under low WSS [LSA], P ϭ .10 ciations did not reach statistical significance. Geometrically, un- before adjustment) and more concentrated inflow jets (inflow stable aneurysms are larger (aneurysm maximum size [Asize], concentration index, P ϭ .07 before adjustment), but these asso- P ϭ .04; SizeR, P ϭ .04), more elongated (AR, P ϭ .04; volume- AJNR Am J Neuroradiol 39:1860–66 Oct 2018 www.ajnr.org 1863 FIG 3. Ratios of mean values of the ruptured aneurysm group over mean values of the unruptured aneurysm group. Only hemodynamic (dark gray) and geometric (light gray) variables that were significantly different between these groups (P Ͻ .05 before adjusting for multiple testing) are included. to-ostium ratio [VOR], P ϭ .04), and more irregular in shape DISCUSSION (bulge location [BL], P ϭ .04; conicity parameter [CP], P ϭ .04; The current study attempts to characterize hostile aneurysm he- area-weighted Gaussian curvature; [GAA], P ϭ .04) than stable modynamics and geometry that could predispose aneurysms to aneurysms. instability and rupture while controlling for location, sex, and age. This knowledge is important for improved understanding of the Unstable-versus-Ruptured Aneurysms mechanisms of wall degeneration and aneurysm progression. No significant differences between unstable unruptured aneu- The differences observed between stable and ruptured aneu- rysms and matched ruptured aneurysms were found, including rysm pairs allow us to identify hemodynamic and geometric char- both hemodynamic and geometric variables (see the results in acteristics that are more prevalent in ruptured aneurysms. Our On-line Table 5). results indicate that these hostile hemodynamic characteristics include extreme values of wall shear stress (low and high), con- Unruptured-versus-Ruptured Aneurysms centrated and oscillatory shear stress distributions, and complex Results of the comparison of all unruptured aneurysms included intrasaccular flow patterns. Adverse aneurysm geometries include in this study (stable and unstable combined) against matching large, elongated, and irregular shapes. These characteristics are ruptured aneurysms are presented in On-line Table 6 and sum- illustrated with 3 examples presented in Fig 4. Additionally, a marized in Fig 3. Again, ruptured aneurysms had more SCI (P ϭ simple correlation analysis was performed to identify variables .04) and more complex flow patterns (CORELEN, P ϭ .04) with correlated to one another. Groups of correlated variables (whose higher Vmax (P ϭ .04) compared with unruptured aneurysms. pair-wise regression coefficient was greater than 0.80) are pre- They also tended to have lower WSSmin (P ϭ .02 before adjust- sented in On-line Tables 7–9. Most interesting, no correlations ment) and larger WSSmax (P ϭ .02 before adjustment), as well were found between hemodynamic and geometric variables in as more maximum oscillatory shear stress (OSImax, P ϭ .02; this sample. OSImean, P ϭ .02 before adjustment), but these associations be- Differences between stable and unstable aneurysms were less came only marginally significant (P ϭ .06) after adjusting for pronounced but in the same general direction as those between multiple testing. Geometrically, ruptured aneurysms were more stable and ruptured aneurysms. Specifically, unstable aneurysms elongated (AR, P ϭ .04) and more irregular (NSI, P ϭ .04) and had more complex flow patterns with concentrated wall shear tended to be larger (SizeR, P ϭ .02 before adjustment), but the stress than stable aneurysms, and they were larger and more elon- latter association became marginally significant (P ϭ .06) after gated and irregular. These features suggest that unstable and rup- adjusting for multiple testing. tured aneurysms differ from stable aneurysms in similar ways. 1864 Chung Oct 2018 www.ajnr.org larger samples are needed to confirm this conjecture. Finally, when compar- ing all unruptured aneurysms (stable and unstable combined) against rup- tured aneurysms, differences similar to those observed between stable and rup- tured aneurysms were observed. This is likely due to the larger proportion of sta- ble aneurysms in the unruptured group. Several studies have compared stable and unstable unruptured aneurysms, but without controlling for location or patient characteristics. One such study did not find significant geometric or he- modynamic differences (149 stable, 20 unstable),26 while 2 other studies found differences in geometric (70 stable, 23 unstable)27 and hemodynamic (17 sta- ble, 16 unstable) characteristics.11 Our findings agree with these latter studies. Two other studies found larger LSAs in unstable aneurysms compared with sta- ble aneurysms after matching location and size (12 stable, 12 unstable)14 as well as sex and age (12 stable, 12 unstable).28 In general, our results show similar trends, but LSA did not reach statistical significance. In summary, the results of our study suggest that hostile hemodynamic envi- ronments (ie, more prevalent in rup- tured and unstable aneurysms than in stable aneurysms) are characterized by complex flows and concentrated wall shear stress distributions, while adverse geometries include larger and more elongated and irregular shapes. This finding leads us to propose that through geometric and flow analysis, there are ways to a priori identify unruptured an- eurysms that will evolve toward instabil- ity and will therefore be at a higher risk of rupture. Additionally, our study sug- gests that lumping stable and unstable aneurysms into a single group of unrup- tured aneurysms (as in most cross-sec- FIG 4. Examples of stable (left), unstable (middle), and ruptured (right) aneurysms at a single tional studies) still allows characteriza- location (posterior communicating artery). Visualizations show, from upper to lower: inflow jet, tion of adverse conditions likely because flow pattern, vortex core lines at 4 instants during the cardiac cycle, wall shear stress, and oscillatory shear index. of the higher prevalence of stable aneu- rysms in the unruptured population. Furthermore, no significant differences were found between un- Our study has several limitations. Computational fluid dy- stable and ruptured aneurysms. These observations are consistent namics models are based on several assumptions and approxima- with a study25 that found differences between symptomatic and tions.19 Our sample size and available data were not large enough asymptomatic posterior communicating artery aneurysms that to allow us to match and control for more patient characteristics were similar to the differences between symptomatic and rup- or to obtain more statistical significance. The unstable group con- tured aneurysms at this location. This finding suggests that hemo- tained growing and cerebral palsy cases, which should be studied dynamic environments and shapes of unstable aneurysms may separately with larger samples. This study identified several trends resemble those of ruptured aneurysms, but further studies with that suggest that hostile hemodynamics and geometries could AJNR Am J Neuroradiol 39:1860–66 Oct 2018 www.ajnr.org 1865 help identify aneurysms at higher risk of instability and rupture hemodynamic environment in ruptured and unruptured brain an- and should be considered in studies aiming at further under- eurysms. AJNR Am J Neuroradiol 2011;32:145–51 CrossRef Medline standing the underlying mechanisms that contribute to the evo- 10. Dhar S, Tremmel M, Mocco J, et al. Morphology parameters for intracranial aneurysm rupture risk assessment. Neurosurgery 2008; lution of a cerebral aneurysm from stable to unstable and ulti- 63:185–96; discussion 196–97 CrossRef Medline mately ruptured status. 11. Sforza DM, Kono K, Tateshima S, et al. Hemodynamics in growing and stable cerebral aneurysms. J Neurointerv Surg 2016;8:407–12 CONCLUSIONS CrossRef Medline 12. Tateshima S, Tanishita K, Omura H, et al. Intra-aneurysmal hemo- Unstable and ruptured aneurysms have more complex flows with dynamics during the growth of unruptured aneurysm: in vitro concentrated wall shear stress and are larger, more elongated, and study using longitudinal CT angiogram database. AJNR Am J Neu- irregular than stable aneurysms, independent of aneurysm loca- roradiol 2007;28:622–27 Medline tion, sex, and age. These adverse conditions could be used to iden- 13. Boussel L, Rayz V, McCulloch C, et al. Aneurysm growth occurs at tify unruptured aneurysms at higher risk of rupture and should be region of low wall shear stress: patient-specific correlation of he- modynamics and growth in a longitudinal study. Stroke 2008;39: taken into account in studies of the mechanisms responsible 2997–3002 CrossRef Medline for aneurysm wall degradation and progression to instability 14. Brinjikji W, Chung BJ, Jimenez C, et al. Hemodynamic differences and rupture. between unstable and stable unruptured aneurysms independent of size and location: a pilot study. J Neurointerv Surg 2017;9:376–80 Disclosures: Fernando Mut—RELATED: Grant: National Institutes of Health, Com- CrossRef Medline ments: research grant.* Waleed Brinjikji—UNRELATED: Consultancy: Johnson and 15. Kallmes DF. Point: CFD—computational fluid dynamics or con- Johnson; Stock/Stock Options: Marblehead Medical LLC; Travel/Accommoda- founding factor dissemination. AJNR Am J Neuroradiol 2012;33: tions/Meeting Expenses Unrelated to Activities Listed: Johnson and Johnson. Juan R. 396–96 CrossRef Medline Cebral—RELATED: Grant: National Institutes of Health *; UNRELATED: Grants/ 16. Cebral JR, Castro MA, Appanaboyina S, et al. Efficient pipeline Grants Pending: National Institutes of Health, Philips Healthcare.* *Money paid to for image-based patient-specific analysis of cerebral aneurysm the Institution. hemodynamics: technique and sensitivity. IEEE Trans Med Imaging 2005;24:457–67 CrossRef Medline 17. Cebral JR, Castro MA, Putman CM, et al. Flow-area relationship in REFERENCES internal carotid and vertebral arteries. Physiol Meas 2008;29:585–94 1. Chien A, Sayre J, Vin˜uela F. Comparative morphological analysis of CrossRef Medline the geometry of ruptured and unruptured aneurysms. Neurosurgery 18. Mut F, Aubry R, Lo¨hner R, et al. Fast numerical solutions of patient- 2011;69:349–56 CrossRef Medline specific blood flows in 3D arterial systems. Int J Num Meth Biomed 2. Greving JP, Wermer MJ, Brown RD Jr, et al. Development of the Eng 2010;26:73–85 CrossRef Medline PHASES score for prediction of risk of rupture of intracranial 19. Mut F, Lo¨hner R, Chien A, et al. Computational hemodynamics aneurysms: a pooled analysis of six prospective cohort studies. Lan- framework for the analysis of cerebral aneurysms. Int J Num Meth cet Neurol 2014;13:59–66 CrossRef Medline Biomed Eng 2011;27:822–39 CrossRef Medline 3. Connolly ES Jr, Rabinstein AA, Carhuapoma JR, et al; American 20. Raghavan ML, Ma B, Harbaugh RE. Quantified aneurysm shape and Heart Association Stroke Council, Council on Cardiovascular Radi- rupture risk. J Neurosurg 2005;102:355–62 CrossRef Medline ology and Intervention, Council on Cardiovascular Nursing, Council 21. Niven DJ, Berthiaume LR, Fick GH, et al. Matched case-control on Cardiovascular Surgery and Anesthesia, Council on Clinical Car- studies: a review of reported statistical methodology. Clin Epidemiol diology. Guidelines for the management of aneurysmal subarach- 2012;4:99–110 CrossRef Medline noid hemorrhage: a guideline for healthcare professionals from the 22. Efron B, Tibshirani RJ. An Introduction to the Bootstrap. Boca Raton: American Heart Association/American Stroke Association. Stroke Chapman and Hall; 1993 2012;43:1711–37 CrossRef Medline 23. Wilkinson B. A statistical consideration in psychological research. 4. Cebral J, Ollikainen E, Chung BJ, et al. Flow conditions in the intra- Psychol Bull 1951;48:156–58 CrossRef Medline cranial aneurysm lumen are associated with inflammation and de- 24. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a generative changes of the aneurysm wall. AJNR Am J Neuroradiol practical and powerful approach to multiple testing. J R Stat Soc 2017;38:119–26 CrossRef Medline Series B 1995;57:289–300 5. Fro¨sen J, Tulamo R, Paetau A, et al. Saccular intracranial aneurysm: 25. Lv N, Yu Y, Xu J, et al. Hemodynamic and morphological charac- pathology and mechanisms. Acta Neuropathol 2012;123:773–86 teristics of unruptured posterior communicating artery aneu- CrossRef Medline rysms with oculomotor nerve palsy. J Neurosurg 2016;125: 6. Cebral JR, Castro MA, Burgess JE, et al. Characterization of cerebral 264–68CrossRef Medline aneurysms for assessing risk of rupture by using patient-specific 26. Ramachandran M, Retarekar R, Raghavan ML, et al. Assessment of computational hemodynamics models. AJNR Am J Neuroradiol image-derived risk factors for natural course of unruptured cere- 2005;26:2550–59 Medline bral aneurysms. J Neurosurg 2016;124:288–95 CrossRef Medline 7. Shojima M, Oshima M, Takagi K, et al. Magnitude and role of wall 27. Chien A, Xu M, Yokota H, et al. Nonsphericity index and size ratio shear stress on cerebral aneurysm: computational fluid dynamic identify morphologic differences between growing and stable an- study of 20 middle cerebral artery aneurysms. Stroke 2004;35: eurysms in a longitudinal study of 93 cases. AJNR Am J Neuroradiol 2500–05 CrossRef Medline 2018 Jan 25. [Epub ahead of print] CrossRef Medline 8. Xiang J, Natarajan SK, Tremmel M, et al. Hemodynamic-morpho- 28. Skodvin TØ, Evju Ø, Helland CA, et al. Rupture prediction of intra- logic discriminants for intracranial aneurysm rupture. Stroke 2011; cranial aneurysms: a nationwide matched case-control study of he- 42:144–52 CrossRef Medline modynamics at the time of diagnosis. J Neurosurg 2017 Nov 3:1–7. 9. Cebral JR, Mut F, Weir J, et al. Quantitative characterization of the [Epub ahead of print] CrossRef Medline

1866 Chung Oct 2018 www.ajnr.org BRIEF/TECHNICAL REPORT INTERVENTIONAL

Integrating 3D Rotational Angiography into Gamma Knife Planning

X H. Hasegawa, X S. Hanakita, X M. Shin, X M. Kawashima, X T. Kin, X W. Takahashi, X M. Shojima, X A.K. Nomoto, X S. Aoki, and X N. Saito

ABSTRACT SUMMARY: 3D rotational angiography provides remarkable spatial resolution for cerebrovascular disorders; however, it cannot be integrated directly into gamma knife planning due to the discrepancy of DICOM “tag” information, and most physicians still cannot benefit from 3D rotational angiography. Here, we describe a simple and easy technique to enable the integration of 3D rotational angiography.

3D rotational angiography ؍ ABBREVIATIONS: GKRS ϭ gamma knife stereotactic radiosurgery; 3DRA

amma knife stereotactic radiosurgery (GKRS) is an image- grated directly into GKRS planning because the planning software Gguided radiation therapy characterized by its high geometric (Leksell GammaPlan; Elekta Instruments, Stockholm, Sweden) accuracy; thus, no treatment margin is usually required when cir- does not accept 3DRA. Moreover, very few studies have described cumscribing the target. This feature, in conjunction with its sharp the effectiveness of 3DRA on GKRS planning, though they do not dose fall-off, enables high-dose irradiation in a single session; state a detailed integration method.19-21 Thus, most physicians however, successful radiosurgery is highly dependent on the qual- still cannot benefit from 3DRA. Here, we show a very simple ity of radiographic images used. method to integrate 3DRA into GKRS planning. GKRS has been accepted as one of the standard therapeutic modalities for small-to-medium arteriovenous malformations.1-9 Techniques Currently, biplanar DSA and CT angiography or MR imaging or Before the day of treatment, we usually perform MR imaging both are commonly used in most institutions.10-15 Recently, ad- (mainly time-of-flight MR angiography, supplemented by T2 and vances in modern endovascular suites and newer generation flat gadolinium-enhanced T1 images) for preplanning. As in the usual panel detectors with C-arm systems have enabled acquisition of preparations for GKRS, the Leksell frame (Elekta Instruments) is 3D rotational angiography (3DRA), providing remarkable spatial set on the patient’s head with the patient under sedation with local resolution for cerebrovascular disorders. Indeed, 3DRA is begin- anesthesia. Then, the patient is transferred to the angiographic ning to be used in the treatment planning of other modalities of suite (Allura Xper FD20/10; Philips Healthcare, Best, Nether- stereotactic radiation therapy, including CyberKnife (Accuray, lands). Along with conventional DSA, 3DRA is acquired using the Sunnyvale, California), XKnife (Integra LifeSciences, Plainsboro, programmed acquisition protocol (3DRA mode). The amount of New Jersey), and Trilogy (Varian Medical Systems, Palo Alto, contrast medium used and preinjection delay are individually de- California), contributing to improved accuracy of the treatment termined by neuroendovascular surgeons; briefly, 1–3 mL/s of planning.16-18 One remaining issue is that 3DRA cannot be inte- contrast medium is continuously injected with a 1.5- to 2.0-sec- ond preinjection delay during the rotation of the C-arm. Because 3DRA could be easily coregistered to stereotactic CT if 3DRA Received April 4, 2018; accepted after revision June 12. contained enough bony tissue, a large-sized detector is generally From the Departments of Neurosurgery (H.H., S.H., M. Shin, M.K., T.K., M. Shojima, N.S.) and Radiology (W.T., A.K.N., S.A.), University of Tokyo Hospital, Tokyo, Japan. preferred for precise image coregistration (Fig 1A). On the con- This study was supported by Japan Society for the Promotion of Science (JSPS) trary, the smallest 8-inch detector could be used when the nidus is KAKENHI grant (No. JP17K16628). located near the skull base (basal ganglia, posterior fossa, and so Please address correspondence to Hirotaka Hasegawa, MD, Department of Neuro- surgery, The University of Tokyo Hospital, 7-3-one Hongo, Bunkyo-ku, Tokyo forth) because acquired images spontaneously contain a large 113-8655, Japan; e-mail: [email protected] portion of bony tissues (Fig 1B). The obtained volume dataset is Indicates open access to non-subscribers at www.ajnr.org automatically transferred to the preinstalled workstation (Xtra- Indicates article with supplemental on-line photo. Vision; Philips Healthcare), with which further reconstruction is http://dx.doi.org/10.3174/ajnr.A5763 performed with a 2563-or5123-resolution voxel matrix in a

AJNR Am J Neuroradiol 39:1867–70 Oct 2018 www.ajnr.org 1867 slow C-arm rotation and an 8-inch detector, which might be better for very small arteriovenous malforma- tions. However, use of the high-resolu- tion XperCT mode might raise a con- cern about coregistration. Notably, images containing a large portion of the cranium as well as skull base bone are important for precise coregistration; the use of a smaller detector contributes to further increased spatial resolution but also loses images of the surrounding cranium and skull base, leading to dif- ficulty in coregistration. We usually perform stereotactic CT after finishing angiography so that the contrast me- FIG 1. 3D rotational angiography images obtained with a large (A) and an 8-inch detector dium remaining in the blood vessels (B) are coregistered to each stereotactic CT image on treatment-planning software (GammaPlan). can provide additional information of vascular anatomies, which may en- planned cube-shaped FOV with preset side lengths of 34.96, hance the quality of image coregistration. Further research is 52.18, 69.92, or 104.36 mm. We mostly use a 69.92-mm side- desirable to examine the coregistration accuracy to ensure the 3 length cube with a 256 -resolution voxel matrix to maintain quality of the prescription of the therapeutic radiation dose. the balance between spatial resolution and contrast resolution; Moreover, when a nidus receives blood flow from Ͼ2 ves- thus, spatial resolution is roughly calculated as 0.27 mm. Then, sels, the precise nidus contour is shown as the summation of the DICOM “tag” technique is corrected from XA into CT parts of the nidus obtained by cannulation in each vessel. Thus, in our software so that the GammaPlan can recognize the it is quite important to perform 3- or 4-vessel angiography and 3DRA properly as a CT-like image. Once the 3DRA is installed judge the involvement to avoid underestimation of the whole in the GammaPlan, image coregistration to stereotactic CT nidus angioarchitecture. will be automated using the preinstalled coregistration func- Although 3DRA provides superior resolution for angio- tion. A step-by-step instruction manual to integrate 3DRA is architectures of vascular lesions, we recommend creating ra- shown in On-line Figure. Detailed case illustrations are shown diosurgical plans by meticulously comparing all the available in Fig 2. imaging modalities. Particularly, MR imaging exhibits excel- DISCUSSION lent contrast resolution and provides a better understanding of With the above-described simple contrivance, 3DRA can readily the surrounding functional brain anatomies, and DSA enables be used for GKRS planning. The strength of this technique is its surgeons to instinctively recognize a spatial expanse of the ni- accessibility. GammaPlan cannot directly accept 3DRA because of dus. Notably, surgeons should manipulate the DICOM header the discrepancy of DICOM tag information; accordingly, we must at their own risk because carelessly manipulating DICOM address this issue. No changes are required in the geometric or could spoil the quality of the images. Meticulous care must be patient information; thus, the image quality itself is intact. Be- taken not to change important information in the DICOM cause 3DRA is a CT-like image, being characteristic of the finest header other than technique because it could alter spatial rela- spatial resolution for high-contrast objects with poor contrast res- tionships. Given the above reasons, 3DRA should not be used olution and thus having many features common to CT,22,23 ste- as a main technique but as reference information during radio- reotactic CT would be the best reference image for the coregistra- surgical planning. tion.18 By means of high-definition images, physicians can not only reduce unwanted waste radiation to the surrounding brain CONCLUSIONS tissues but also enable a safe prescription of high radiosurgical We describe a simple, easy-to-access technique to enable inte- doses, enough to obliterate the nidus, which might theoretically gration of 3DRA into GKRS planning, which could provide the lead to improvement in the obliteration rate as well as a decrease highest resolution for angioarchitectures of vascular lesions. in radiation-induced adverse events. However, this article is only The present method remains preliminary; thus, the created a technical report, and the actual clinical outcomes should be treatment plans should be validated in comparison with the further examined. conventional planning method. Further research is desirable to Although 3DRA mode is preferred in our institution, contrast- assess the effect of this technique on the actual radiosurgical enhanced conebeam CT with a small targeted FOV (high-resolu- outcomes. tion XperCT mode; Philips Healthcare) is also available in our angiographic suite. This acquisition mode provides superior spa- Disclosures: Hirotaka Hasegawa—RELATED: Grant: JSPS KAKENHI, Comments: grant tial resolution with a long acquisition time (20 seconds), using a No. JP17K16628. 1868 Hasegawa Oct 2018 www.ajnr.org FIG 2. A 34-year-old woman with an unruptured, small, left medial frontal arteriovenous malformation, once treated with gamma knife (A). The small remnant nidus persisted at 3.5 years from the initial radiosurgery (B,ared arrowhead shows the remnant nidus). The actual treatment planning of the secondary treatment (C) shows that 3D rotational angiography (left column) successfully depicts the faint remnant, while both time-of-flight (middle column) and gadolinium-enhanced T1 images (right column) fail to depict it. The nidus was finally obliterated (D)at1.5 years from the secondary treatment. Yellow lines show the prescription isodose lines.

REFERENCES 9. Starke RM, Ding D, Kano H, et al. International multicenter cohort 1. Cohen-Gadol AA, Pollock BE. Radiosurgery for arteriovenous mal- study of pediatric brain arteriovenous malformations, part 2: out- formations in children. J Neurosurg 2006;104:388–91 Medline comes after stereotactic radiosurgery. J Neurosurg Pediatr 2017;19: 2. Lunsford LD, Kondziolka D, Flickinger JC, et al. Stereotactic radio- 136–48CrossRef Medline surgery for arteriovenous malformations of the brain. J Neurosurg 10. Blatt DR, Friedman WA, Bova FJ. Modifications based on computed 1991;75:512–24 CrossRef Medline tomographic imaging in planning the radiosurgical treatment of 3. Karlsson B, Lindquist C, Steiner L. Prediction of obliteration after arteriovenous malformations. Neurosurgery 1993;33:588–95; dis- gamma knife surgery for cerebral arteriovenous malformations. cussion 595–96 Medline Neurosurgery 1997;40:425–30; discussion 430–31 Medline 11. Ding D, Xu Z, Shih HH, et al. Stereotactic radiosurgery for partially 4. Schwartz M, Sixel K, Young C, et al. Prediction of obliteration of resected cerebral arteriovenous malformations. World Neurosurg arteriovenous malformations after radiosurgery: the obliteration 2016;85:263–72 CrossRef Medline prediction index. Can J Neurol Sci 1997;24:106–09 CrossRef Medline 12. Hanakita S, Koga T, Shin M, et al. The long-term outcomes of radio- 5. Flickinger JC, Kondziolka D, Maitz AH, et al. An analysis of the surgery for arteriovenous malformations in pediatric and adoles- dose-response for arteriovenous malformation radiosurgery cent populations. J Neurosurg Pediatr 2015;16:222–31 CrossRef and other factors affecting obliteration. Radiother Oncol 2002;63: Medline 347–54 CrossRef Medline 13. Kano H, Lunsford LD, Flickinger JC, et al. Stereotactic radiosurgery 6. Fleetwood IG, Steinberg GK. Arteriovenous malformations. Lancet for arteriovenous malformations, part 1: management of Spetzler- 2002;359:863–73 CrossRef Medline Martin Grade I and II arteriovenous malformations. J Neurosurg 7. Shin M, Kawamoto S, Kurita H, et al. Retrospective analysis of a 2012;116:11–20 CrossRef Medline 10-year experience of stereotactic radio surgery for arteriovenous 14. Kondziolka D, Lunsford LD, Kanal E, et al. Stereotactic magnetic malformations in children and adolescents. J Neurosurg 2002;97: resonance angiography for targeting in arteriovenous malforma- 779–84CrossRef Medline tion radiosurgery. Neurosurgery 1994;35:585–90; discussion 590–91 8. Starke RM, Kano H, Ding D, et al. Stereotactic radiosurgery for CrossRef Medline cerebral arteriovenous malformations: evaluation of long-term 15. Lunsford LD, Niranjan A, Kano H, et al. The technical evolution of outcomes in a multicenter cohort. J Neurosurg 2017;126:36–44 gamma knife radiosurgery for arteriovenous malformations. Prog CrossRef Medline Neurol Surg 2013;27:22–34 CrossRef Medline AJNR Am J Neuroradiol 39:1867–70 Oct 2018 www.ajnr.org 1869 16. van der Bom IM, Gounis MJ, Ding L, et al. Target delineation for 20. Kang J, Huang J, Gailloud P, et al. Planning evaluation of C-arm radiosurgery of a small brain arteriovenous malformation using cone beam CT angiography for target delineation in stereotactic high-resolution contrast-enhanced cone beam CT. BMJ Case Rep radiation surgery of brain arteriovenous malformations. Int J Ra- 2013;2013 CrossRef Medline diat Oncol Biol Phys 2014;90:430–37 CrossRef Medline 17. Colombo F, Cavedon C, Francescon P, et al. Three-dimensional an- 21. Chen KK, Guo WY, Yang HC, et al. Application of time-resolved 3D giography for radiosurgical treatment planning for arteriovenous digital subtraction angiography to plan cerebral arteriovenous malformations. J Neurosurg 2003;98:536–43 CrossRef Medline malformation radiosurgery. AJNR Am J Neuroradiol 2017;38: 18. Hristov D, Liu L, Adler JR, et al. Technique for targeting arterio- 740–46CrossRef Medline venous malformations using frameless image-guided robotic ra- 22. Miracle AC, Mukherji SK. Conebeam CT of the head and neck, part diosurgery. Int J Radiat Oncol Biol Phys 2011;79:1232–40 CrossRef 1: physical principles. AJNR Am J Neuroradiol 2009;30:1088–95 Medline CrossRef Medline 19. Safain MG, Rahal JP, Raval A, et al. Use of cone-beam computed 23. Kanayama S, Hara T, Hamada Y, et al. Potential of 80-kV high-res- tomography angiography in planning for gamma knife radiosurgery olution cone-beam CT imaging combined with an optimized pro- for arteriovenous malformations: a case series and early report. Neu- tocol for neurological surgery. Neuroradiology 2015;57:155–62 rosurgery 2014;74:682–95; discussion 695–96 CrossRef Medline CrossRef Medline

1870 Hasegawa Oct 2018 www.ajnr.org ORIGINAL RESEARCH INTERVENTIONAL

Quantification of Blood Velocity with 4D Digital Subtraction Angiography Using the Shifted Least-Squares Method

X Y. Wu, X G. Shaughnessy, X C.A. Hoffman, X E.L. Oberstar, X S. Schafer, X T. Schubert, X K.L. Ruedinger, X B.J. Davis, X C.A. Mistretta, X C.M. Strother, and X M.A. Speidel

ABSTRACT BACKGROUND AND PURPOSE: 4D-DSA provides time-resolved 3D-DSA volumes with high temporal and spatial resolutions. The pur- pose of this study is to investigate a shifted least squares method to estimate the blood velocity from the 4D DSA images. Quantitative validation was performed using a flow phantom with an ultrasonic flow probe as ground truth. Quantification of blood velocity in human internal carotid arteries was compared with measurements generated from 3D phase-contrast MR imaging.

MATERIALS AND METHODS: The centerlines of selected vascular segments and the time concentration curves of each voxel along the centerlines were determined from the 4D-DSA dataset. The temporal shift required to achieve a minimum difference between any point and other points along the centerline of a segment was calculated. The temporal shift as a function of centerline point position was fit to a straight line to generate the velocity. The proposed shifted least-squares method was first validated using a flow phantom study. Blood velocities were also estimated in the 14 ICAs of human subjects who had both 4D-DSA and phase-contrast MR imaging studies. Linear regression and correlation analysis were performed on both the phantom study and clinical study, respectively.

RESULTS: Mean velocities of the flow phantom calculated from 4D-DSA matched very well with ultrasonic flow probe measurements with 11% relative root mean square error. Mean blood velocities of ICAs calculated from 4D-DSA correlated well with phase-contrast MR imaging measurements with Pearson correlation coefficient r ϭ 0.835.

CONCLUSIONS: The availability of 4D-DSA provides the opportunity to use the shifted least-squares method to estimate velocity in vessels within a 3D volume.

ABBREVIATIONS: PC ϭ phase-contrast; SBR ϭ sideband ratio; TCC ϭ time concentration curve; VIPR ϭ vastly undersampled isotropic projection reconstruction

D digital subtraction angiography and a 3D rotational angiog- flow in arteries from 2D-DSA; these are divided into 2 major 2raphy acquisition followed by 3D-DSA reconstruction are the classes: bolus-tracking and computational methods.4-11 A thorough standards for vascular morphology assessment. There is an in- review of these techniques can be found in Shpilfoygel et al.12 How- creasing demand for hemodynamic information, including blood ever, 2D-DSA based methods are challenged by vessel overlap and flow rate and velocity for diagnosis, treatment planning, and eval- vessels being aligned with the primary x-ray beam direction. 1-3 uation. Many algorithms have been proposed to estimate blood Recently, a new reconstruction method13-16 has been applied to a single 3D rotational angiography acquisition to generate time-resolved 3D-DSA at frame rates up to 30 frames/second (4D Received February 6, 2018; accepted after revision June 11. DSA). The availability of the geometric and temporal data in a From the Departments of Medical Physics (Y.W., G.S., C.A.H., C.A.M., M.A.S.), Bio- medical Engineering (E.L.O., K.L.R., B.J.D.), Radiology (C.A.M., C.M.S., T.S.), and Medi- 4D-DSA reconstruction provides the opportunity to estimate ve- cine (M.A.S.), University of Wisconsin, Madison, Wisconsin; Siemens Healthineers locity and flow more accurately than has previously been possible (S.S.), USA; and Department of Radiology and Nuclear Medicine (T.S.), Basel Univer- sity Hospital, Basel, Switzerland. with 2D DSA. In this article, a shifted least-squares based tech- This work was supported by the National Institutes of Health Center for Scientific nique was used with the 4D-DSA data to estimate blood velocity. Review 2R01HL116567 and Siemens Healthineers. The proposed algorithm was first validated using flow phantom The concepts and information presented in this article are based on research and are not commercially available. studies, in which the velocity could be documented using an ul- Please address correspondence to Yijing Wu, MD, Department of Medical Physics, trasonic flow probe. Blood velocity was also determined in 4D- University of Wisconsin, Madison, 1125 WIMR, 1111 Highland Ave, Madison, WI 53705; e-mail: [email protected] DSA datasets from 14 ICAs of volunteers who also underwent 3D Indicates open access to non-subscribers at www.ajnr.org fast phase-contrast with vastly undersampled isotropic projection http://dx.doi.org/10.3174/ajnr.A5793 reconstruction (PC VIPR) studies.17-19

AJNR Am J Neuroradiol 39:1871–77 Oct 2018 www.ajnr.org 1871 ͉ ͉2 MATERIALS AND METHODS PSf SBR ϭ 0 , 4D DSA Acquisition and Data Preparation 1 ͸͉PS ͉2 Noncontrast (mask) and contrast-enhanced (fill) rotational ac- N fi quisitions were performed using a conventional flat panel detec- tor angiographic system (Artis zee; Siemens, Erlangen, Germany) where the denominator summation includes the neighboring fre- with the following settings: 70 kV(peak), 0.36 ␮Gy/frame, 304 quency points and excludes the fundamental frequency f0. From images, 260° arc. Contrast injection (iohexol, Omnipaque 300; this SBR map, a low-pass filter was applied and a threshold of GE Healthcare, Piscataway, New Jersey) into the subject was one-fourth of the median SBR was enforced to generate a mask. started at the beginning of the fill run (3 mL/second, 7-second This mask excludes low pulsatile signals from consideration in the injection duration). The rotational acquisition protocol yields x- velocity calculation. ray projection images, which are used to automatically recon- struct both 3D-DSA and 4D-DSA with an isotropic spatial reso- Phantom Study lution of 0.46 ϫ 0.46 ϫ 0.46 mm3. To validate the proposed method for velocity estimation, we used The centerline of each vascular segment was determined from a flow phantom containing a loop of plastic tubing to perform the static 3D volume using an efficient 3D parallel-thinning algo- 4D-DSA studies. Pulsatile water flow with different mean flow 20 rates was generated using a perfusion pump with a controllable rithm. The contrast waveform map, which is the time concen- flow setting (S3; Storkcert, Freiburg, Germany). A 5F catheter was tration curves (TCCs) along the centerline path (curvilinear introduced into the tubing for injection of iodinated contrast me- length z), was extracted from the 4D-DSA and stored for each dium (Omnipaque 300 diluted with 25% water) with a power vascular segment for velocity calculation. injector. 4D-DSA-derived flow was compared with flow measure- Shifted Least-Squares Algorithm ments made with an ultrasonic flow probe (400-Series Multi- In an arterial contrast injection, there is a temporal oscillation in Channel Flowmeter; Transonic, Ithaca, New York) attached to iodine concentration that arises from the mixing of contrast me- the tubing 5 inches away from the catheter tip. dium, which is injected at a fixed rate, and nonopacified blood, Rotational projection images were acquired using the mask which flows at a variable rate driven by the cardiac cycle. This and iodine-filled scans with 12-second acquisition over a 260° arc. temporal variation in contrast, referred to as pulsatility, appears at The frame rate was 30 frames per second. The 4D-DSA volume points downstream of the injection with a time delay. The time data were reconstructed using the vendor’s prototype software delay is related to the distance downstream and the blood velocity. (Siemens Healthineers, Forchheim, Germany). Therefore, velocity can be estimated by measuring the distance A time-resolved flow profile was recorded during each fill scan along the vessel centerline and the time delay. using the ultrasonic probe. The reference flow measurement was For any 2 points, i and j, along the centerline, the shifted least- taken as the average of the flow profile over the same time window that was used for 4D-DSA-derived flow. squares difference between their TCCs, ci(t) and cj(t), can be cal- culated as Five different flow settings were selected to provide average flow rates ranging from 285 to 1140 mL/min (corresponding to the mean velocity of 15–60 cm/s for the tubing with a 0.64-cm 1 T ␧͑␶ ͒ ϭ ͸͓ ͑ Ϫ ␶ ͒ Ϫ ͑ ͔͒2 diameter). Three 4D-DSA scans and reference flow measure- ij ͱ ci t ij cj t . T t ϭ 1 ments were made for each flow setting.

␶0 ␶0 ʦ The value ij, ij [0, T] that minimizes the shifted least-squares Clinical Study ␧ ␶ difference, ( ij), is regarded as the time of bolus transport be- Nine subjects who had both DSA and PC VIPR examinations tween these 2 points. The spatial distance z between the 2 points is were selected from a data base generated under University of Wis- calculated from the 3D path length along the vessel centerline. consin review board approval. Because 5 of 9 subjects had bilateral ␶0 The time-shift as a function of the spatial distance z can then be studies, a total of 14 ICAs were available for evaluation. Scanning fit to a linear relation: parameters for postcontrast PC VIPR were the following: FOV ϭ 22 ϫ 22 ϫ 22 cm2, TR/TE ϭ 12.5/4.8 ms, velocity-encoded gra- ␶0 ϭ ␣ ϫ z ϩ b, dient-echo imaging ϭ 80ϳ100 cm/s, bandwidth ϭ 83.3 kHz, where the slope a is the inverse of the velocity v. readout matrix ϭ 320 points per projections, spatial resolution for the complex difference image ϭ 0.68 ϫ 0.68 ϫ 0.68 mm3. The Optimizing Waveform Selection complex difference images from the PC VIPR scan were seg- It has been found that the shifted least-squares algorithm provides mented for vessel content using Materialise Mimics (Materialise the most reliable velocity calculation in vessel segments where NV, Leuven, Belgium). The segmented MR imaging data were pulsatility is strong and consistent. To automatically select the then registered to the 3D-DSA using a correlated point registra- waveform regions with strong and consistent pulsatility, we gen- tion with manual point placement, followed by an affine rigid erated a sideband ratio (SBR) map.21 registration. Centerline coordinates of the ICAs generated from For each point in the waveform map, a short-time Fourier the 3D-DSA volumes were used for velocity estimation. An auto-

transform was applied to generate the local power spectrum PSfn, mated workflow for the PC VIPR analysis was developed in th where fn is the n characteristic frequency. The SBR at this point Matlab (MathWorks, Natick, Massachusetts). This software per- can be calculated as formed cross-sectional plane placement for every voxel along the 1872 Wu Oct 2018 www.ajnr.org FIG 1. A representative phantom study demonstrates the flow calculation from 4D-DSA. A, MIP image of the reconstructed phantom with centerline (red) overlaid. B, The contrast waveform map of the voxels along the centerline. The horizontal direction is the timeframe in the 4D-DSA scan, and the vertical direction is the position along the vessel centerline. Each point in the map M(t,z) represents the signal intensity of the voxel at distance z along the centerline and at time frame t. The highlighted area is the optimized waveform region where the pulsatility is strong and consistent. TCCs of 2 selected voxels along the centerline (marked as blue and red stars on A and blue and red curves on B) are shown in C. D, The least-squares differences of these 2 signals as a function of the time-shift, in which the minimal appears at ␶0 ϭ 5 shown as red arrow on D is considered as the time of bolus transport from the blue voxel to the red voxel. The time-shift as a function of the centerline position was fit to a linear relation (F), where the slope is the inverse of the velocity. E, The flow profile recorded from the flow probe (downsampled to 1/30 second to correspond to the 4D-DSA timeframes). Flow measurement is the average taken between the red lines, which correspond to the optimized window shown in B. AUI indicates arbitrary unit of intensity.

AJNR Am J Neuroradiol 39:1871–77 Oct 2018 www.ajnr.org 1873 3D centerline path, segmentation of the vessel boundaries in each versus 4D-DSA for the clinical study. The Pearson r was used for plane, and averaging of the normal velocities for all points inside correlation analysis, and a P value Ͻ .05 for the correlation coef- the segmented boundaries. The reported velocity value for a vessel ficient was considered statistically significant. was then the average result over all the centerline points that were present in both the PC VIPR and 3D-DSA volumes. RESULTS Phantom Study Statistical Analysis Figure 1 shows sample data from the phantom study. First, the The validation process was performed using linear regression fit centerline was generated from the static 3D volume (Fig 1A). Fig- between flow probe (Vfp) and Vdsa (calculated using 4D-DSA) for ure 1B is the contrast waveform map (TCCs along the centerline the phantom study. Correlation analysis and Bland-Altman plots path) of the selected centerline. Each point in the map M(t,z) were used to compare the velocity measurements using PC VIPR represents the signal intensity of the voxel at distance z along the

centerline and at timeframe t [Cz(t)]. The intensity variations along the verti- cal axis reflect the concentration of io- dine contrast medium versus the dis- tance from the injection site for a given time point. Variations along the hori- zontal axis reflect variations in concen- tration versus time for a given point along the centerline. The slope of the in- tensity ridges reflects the blood velocity. In the displayed format, higher blood velocities produce ridges that are closer to vertical, whereas lower blood veloci- ties produce ridges closer to horizontal. The highlighted region is the optimized waveform region for the flow calcula- tion determined using the SBR method. TCCs of 2 selected voxels along the cen- terline (marked as blue and red stars on FIG 2. Linear regression between the flow probe measurements and the 4D-DSA calculations from 15 phantom studies. Evaluations have been obtained with significance levels P ϭ 4.521E-11 for Fig 1A and blue and red dashed lines on the slope and P ϭ 0.18 for the intercept.

FIG 3. A representative ICA study. A, The centerline positions of the ICA. B, The contrast waveform map of the voxels along the centerline. The highlighted area is the optimized waveform region for the flow calculation. C, The time-shift as a function of the centerline position was fit to a linear relation in which the slope is the inverse of the velocity. 1874 Wu Oct 2018 www.ajnr.org FIG 4. Correlation analysis between velocity estimated from 4D-DSA and measured using PC VIPR. The significance level is P ϭ .0002.

4.521E-11 and P ϭ 0.18, respectively. The relative root mean square error (normalized by mean velocity) was 11.3%.

Clinical Study The average ICA flow rates were calcu- lated from 13 of the 14 ICAs that had 4D-DSA examinations. One subject was excluded due to the poor pulsatility in the TCCs. Figure 3 shows an example of the ICA studies. Average ICA flow rates were also measured with the PC VIPR studies. Correlation between 2 measure- ments using 4D-DSA and PC VIPR is shown in Fig 4. Average ICA flow rates calculated using 4D-DSA and measured FIG 5. Bland-Altman analysis of 4D-DSA compared with PC VIPR. using PC VIPR are well-correlated, with the Pearson r ϭ 0.835 and a significance Fig 1B) are shown in Fig 1C. Figure 1D shows the least-squares level P ϭ .0002. Figure 5 is the Bland-Altman analysis of 4D-DSA differences of these 2 signals as a function of the time-shift, where compared with PC VIPR, which shows all points within 2 SDs, the minimal appears at ␶0 ϭ 5 is considered as the time of bolus with a bias of 6.5 cm/s and 2-SD limits of agreement of Ϫ3.56 cm/s transport from the blue voxel to the red voxel. Figure 1F demon- and 16.5 cm/s. strates the linear fitting between the time-shift of each voxel along the centerline and its corresponding centerline position. Figure DISCUSSION 1E is the flow profile recorded from the flow probe (downsampled The shifted least-squares algorithm is a simple-but-robust to 1/30 second to correspond to the 4D-DSA timeframes). Flow method to calculate flow rates by finding the bolus transport time measurement was taken as an average over the same temporal between any 2 points along the centerline and performing linear window used in the 4D-DSA flow calculation. regression with the corresponding positions. The combination of The linear regression fit between the flow probe measurement the spatial and temporal data available in the 4D-DSA reconstruc- and the 4D-DSA calculation is shown in Fig 2. The linear regres- tion provides the opportunity to apply this technique to measure- ϭ ϩ sion equation is as follows: Vdsa 1.0069Vfp 2.8482, where Vdsa ment of blood flow in routine 4D-DSA acquisitions. The flow is the velocity calculated from 4D-DSA and Vfp is the measure- phantom study demonstrated that by using this method, we were ment using the flow probe. The determination coefficient R2 of able to calculate the flow rates with 11% relative root mean square the fit was 0.9677. The 95% confidence interval slope was 0.8966– error compared with ultrasonic flow probe measurements. Re- 1.1172 and the 95% CI intercept was Ϫ1.4917–7.1881. These sults from the 13 human ICAs showed that the calculated veloci- evaluations have been obtained with significance levels of P ϭ ties from 4D-DSA studies correlated well with the measurements AJNR Am J Neuroradiol 39:1871–77 Oct 2018 www.ajnr.org 1875 FIG 6. Fluctuation of the velocity estimation with a different average range. A, The flow profile recorded from the flow probe. Red lines define the starting average window. This window was gradually expanded to the purple line. The corresponding mean velocity was shown as the blue curve. A zoom-in of the green window is shown in B. Similarly, velocity calculations have been performed by gradually expanding the window coverage from yellow to purple as shown in the waveform map (C). Estimated velocity varies (D) as the window edge slides from position A toward B. AU indicates arbitrary units. using the 3D phase-contrast MR imaging method. We believe that The proposed algorithm worked best when there was good the reason for the higher velocities in the 4D-DSA studies com- and consistent pulsatility in the 4D-DSA TCCs. With each cardiac pared with the PC VIPR measurements (6.5 cm/s bias) is the in- cycle, as an intra-arterial contrast bolus is injected, there is an crement in flow caused by the intra-arterial power injection of oscillation in the ratio of contrast medium to nonopacified blood. contrast medium. This pulsatility can be identified and extracted from the contrast The proposed method calculates the mean flow rate within a waveform map by thresholding the SBR map. For vessels more selected window defined by the SBR map. This truncated average distal to the contrast injection site where the pulsatility can be low, of pulsatile flow varies depending on the window size and loca- injection protocols should be modified to enhance the pulsatility tion. This variation can be demonstrated in Fig 6A and B, where strength or contrast kinetics. The present work focused on the the average of the flow profile was performed with a gradually evaluation of blood velocity within the ICAs. Investigation of expanding average window. The truncated average oscillates de- more downstream segments, including the MCAs and basilar ar- pending on where the truncation window ends and eventually teries, should be performed in the future. converges to the mean signal as the window expands. Similar os- cillation has been observed in the flow calculation using 4D-DSA. Figure 6D shows the estimated average velocity changes as the CONCLUSIONS right side of the window gradually expands as shown in Fig 6C. 4D DSA is a novel technique that provides both the geometric and The current algorithm automatically selects a window purely on temporal information required to measure blood flow velocity the basis of the SBR map so that the selected window range may and flow. In this study, calculated velocity values from 4D-DSA fall anywhere between 4 and 6 pulsatile cycles. Further criteria TCCs correlated well with human ICA measurements derived could be included to trim the window to an integer number of from phase-contrast MR imaging and flow phantom measure- pulsatile cycles to minimize this error. ments using an ultrasonic flow probe. 1876 Wu Oct 2018 www.ajnr.org Disclosures: Gabe Shaughnessy—RELATED: Grant: National Institutes of Health.* 9. Pereira VM, Bonnefous O, Ouared R, et al. A DSA-based method Erick L. Oberstar—RELATED: Grant: University of Wisconsin, Madison, Comments: using contrast-motion estimation for the assessment of the intra- National Institutes of Health Grant R01HL116567*. Sebastian Schafer—UNRELATED: aneurysmal flow changes induced by flow-diverter stents. AJNR Employment: Siemens Healthineers, Comments: full-time employee. Michael A. Am J Neuroradiol 2013;34:808–15 CrossRef Medline Speidel—RELATED: Grant: Siemens Healthineers, Comments: Partial support was received from a research grant*; UNRELATED: Patents (Planned, Pending or Issued): 10. Rhode KS, Lambrou T, Hawkes DJ, et al. Novel approaches to the Wisconsin Alumni Research Foundation, Comments: I am a coinventor on a Wiscon- measurement of arterial blood flow from dynamic digital X-ray im- sin Alumni Research Foundation patent (pending) on resolving artifacts in 4D angio- ages. IEEE Trans Med Imaging 2005;24:500–13 CrossRef Medline graphic data. Money of the patents was paid to Wisconsin Alumni Research Foun- 11. Pereira VM, Ouared R, Brina O, et al. Quantification of internal dation. Charles A. Mistretta—RELATED: Grant: National Institute of Biomedical carotid artery flow with digital subtraction angiography: validation Imaging and Bioengineering, Comments: Strother/Mistretta R01 on 4DDSA*; of an optical flow approach with Doppler ultrasound. AJNR Am J UNRELATED: Patents (Planned, Pending or Issued): Wisconsin Alumni Research Foun- dation. Charles M. Strother—RELATED: Grant: University of Wisconsin School of Neuroradiol 2014;35:156–63 CrossRef Medline Medicine and Public Health*; UNRELATED: Patents (Planned, Pending or Issued): 12. Shpilfoygel SD, Close RA, Valentino DJ, et al. X-ray videodensito- Wisconsin Alumni Research Foundation, Comments: I have a pending IP for flow metric methods for blood flow and velocity measurement: a critical quantification; the IP is assigned to the Wisconsin Alumni Research Foundation. review of literature. Med Phys 2000;27:2008–23 CrossRef Medline *Money paid to the institution. 13. Mistretta CA, Oberstar EL, Davis JB, et al. 4D DSA and 4D fluoroscopy: preliminary implementation. Proc SPIE Int Soc Opt REFERENCES Eng 2010;7622 CrossRef 1. Murayama Y, Usami S, Abe T, et al. Transvenous Doppler guidewire 14. Strother CM, Bender F, Deuerling-Zheng Y, et. al. Parametric color sonographic monitoring during treatment of a complex vertebral coding of digital subtraction angiography. AJNR Am J Neuroradiol arteriovenous fistula associated with neurofibromatosis type 1. 2010;31:919–24 CrossRef Medline Neuroradiology 1999;41:328–33 CrossRef Medline 15. Benndorf G. Color-coded digital subtraction angiography: the end 2. Mast H, Mohr J, Thompson J, et al. Transcranial Doppler ultra- of a monochromatic era? AJNR Am J Neuroradiol 2010;31:925–27 sonography in cerebral arteriovenous malformations: diagnostic CrossRef Medline sensitivity and association of flow velocity with spontaneous hem- 16. Davis B, Royalty K, Kowarschik M, et al. 4D digital subtraction orrhage and focal neurological deficit. Stroke 1995;26:1024–27 angiography: implementation and demonstration of feasibility. CrossRef Medline AJNR Am J Neuroradiol 2013;34:1914–21 CrossRef Medline 3. Turski P, Levine R, Turnipseed W, et al. MR angiography flow 17. Gu T, Korosec FR, Block W, et al. PC VIPR: a high-speed 3D phase- analysis: neurovascular applications. Magn Reson Imaging Clin N contrast method for flow quantification and high-resolution an- Am 1995;3:541–55 Medline giography. AJNR Am J Neuroradiol 2005;26:743–49 Medline 4. Rosen L, Silverman NR, Videodensitometric measurements of 18. Chang W, Landgraf B, Johnson KM, et al. Velocity measurements in blood flow using crosscorrelation techniques. Radiology 1973;109: the middle cerebral arteries of healthy volunteers using 3D radial 305–10 CrossRef Medline phase-contrast HYPRFlow: comparison with transcranial Doppler 5. Silverman NR, Rosen L. Arterial blood flow measurement: assess- sonography and 2D phase-contrast MR imaging. AJNR Am J Neu- ment of velocity estimation methods. Invest Radiol 1977:12:319–24 roradiol 2011;32:54–59 CrossRef Medline CrossRef Medline 19. Schneiders JJ, Ferns SP, van Ooij P, et al. Comparison of phase- 6. Seifalian AM, Hawkes DJ, Colchester AC, et al. A new algorithm for deriving pulsatile blood flow waveforms tested using simulated dy- contrast MR imaging and endovascular sonography for intracra- namic angiographic data, Neuroradiology 1989;31:263–69 Medline nial blood flow velocity measurements. AJNR Am J Neuroradiol 7. Hoffmann KR, Doi K, Fencil LE. Determination of instantaneous 2012;33:1786–90 CrossRef Medline and average blood flow rates from digital angiograms of vessel 20. Lee TC, Kashyap RL, Chu CN. Building skeleton models via 3-D phantoms using distance-density curves. Invest Radiol 1991;26: medial surface axis thinning algorithms. Graphical Models and Im- 207–12 Medline age Processing 1994;56:462–78 CrossRef 8. Imbert B, Meunier J, Mongrain R, et al. Stenosis parameter assess- 21. Shaughnessy G, Hoffman C, Schafer S, et al. Quantitative flow and ment from contrast medium tracking in cineangiography with an velocity measurements of pulsatile blood flow with 4D DSA. Proc optical flow method. Proc SPIE Int Soc Opt Eng 1997;l3034:631–40 SPIE 10132, Medical Imaging 2017: Physics of Medical Imaging, CrossRef 101325R (19 April 2017); doi: 10.1117/12.2254143 CrossRef

AJNR Am J Neuroradiol 39:1871–77 Oct 2018 www.ajnr.org 1877 ORIGINAL RESEARCH HEAD & NECK

Correlation between Human Papillomavirus Status and Quantitative MR Imaging Parameters including Diffusion- Weighted Imaging and Texture Features in Oropharyngeal Carcinoma

X M. Ravanelli, X A. Grammatica, X E. Tononcelli, X R. Morello, X M. Leali, X S. Battocchio, X G.M. Agazzi, X M. Buglione di Monale e Bastia, X R. Maroldi, X P. Nicolai, and X D. Farina

ABSTRACT BACKGROUND AND PURPOSE: The incidence of Oropharyngeal Squampus Cell Carcinoma (OPSCC) cases is increasing especially in the Western countries due to the spreading of human papilloma virus (HPV) infection. Radiological investigations, MRI in particular, are used in the daily clinical practice to stage OPSCC. The aim of this study was to investigate the association of quantitative MR imaging features including diffusion-weighted imaging and human papillomavirus status in oropharyngeal squamous cell carcinoma.

MATERIALS AND METHODS: We retrospectively analyzed 59 patients with untreated histologically proved T2–T4 oropharyngeal squa- mous cell carcinoma. Human papillomavirus status was determined by viral DNA detection on tissue samples. MR imaging protocol included T2-weighted, contrast-enhanced T1-weighted (volumetric interpolated brain examination), and DWI sequences. Parametric maps of apparent diffusion coefficient were obtained from DWI sequences. Texture analysis was performed on T2 and volumetric-interpolated brain examination sequences and on ADC maps. Differences in quantitative MR imaging features between tumors positive and negative for human papillomavirus and among subgroups of patients stratified by smoking status were tested using the nonparametric Mann-Whitney U test; the false discovery rate was controlled using the Benjamini-Hochberg correction; and a predictive model for human papillomavirus status was built using multivariable logistic regression.

RESULTS: Twenty-eight patients had human papillomavirus-positive oropharyngeal squamous cell carcinoma, while 31 patients had human papillomavirus-negative oropharyngeal squamous cell carcinoma. Tumors positive for human papillomavirus had a significantly lower mean ADC compared with those negative for it (median, 850.87 versus median, 1033.68; P Ͻ .001). Texture features had a lower discriminatory power for human papillomavirus status. Skewness on volumetric interpolated brain examination sequences was signifi- cantly higher in the subgroup of patients positive for human papillomavirus and smokers (P ϭ .003). A predictive model based on smoking status and mean ADC yielded a sensitivity of 83.3% and specificity 92.6% in classifying human papillomavirus status.

CONCLUSIONS: ADC is significantly lower in oropharyngeal squamous cell carcinoma positive for human papillomavirus compared with oropharyngeal squamous cell carcinoma negative for it. ADC and smoking status allowed noninvasive prediction of human papillomavirus status with a good accuracy. These results should be validated and further investigated on larger prospective studies.

ABBREVIATIONS: HPV ϭ human papillomavirus; OPSCC ϭ oropharyngeal squamous cell carcinoma; SSF ϭ spatial scaling factor; VIBE ϭ volumetric interpolated brain examination

ropharyngeal squamous cell carcinoma (OPSCC) is probably specific demographic characteristics (younger and with fewer co- Othe most studied head and neck neoplasm in the past decade, morbidities) and better prognosis compared with those with HPV- mainly due to the discovery of the role of human papillomavirus negative OPSCC (usually older with more comorbidities, smokers, infection in carcinogenesis. This led to identification of a subset of and alcohol consumers).1-4 Other characteristics of HPV-positive patients with human papillomavirus (HPV)-positive OPSCC with OPSCC are the high ratio between nodal and primary tumor bur- den5 and the greater presence of intranodal cystic degeneration.6-8 MR imaging is routinely used for staging OPSCC in many centers. In Received December 22, 2017; accepted after revision May 27, 2018. addition to standard morphologic sequences, diffusion-weighted From the Departments of Radiology (M.R., E.T., M.L., G.M.A., R.M., D.F.), Otolaryn- gology–Head and Neck Surgery (A.G., R.M., P.N.), Pathology (S.B.), and Radiother- imaging offers an insight in tumor ultrastructure. Some recent stud- apy (M.B.d.M.e.B.), University of Brescia, Brescia, Italy. ies have suggested that HPV-positive tumors are associated with Please address correspondence to Alberto Grammatica, MD, Department of Otorhinolaryngology–Head and Neck Surgery, University of Brescia, Piazza Spedali lower apparent diffusion coefficient compared with HPV-negative Civili 1, 25123 Brescia, Italy; e-mail: [email protected] ones.9-11 However, these results are equivocal.12 http://dx.doi.org/10.3174/ajnr.A5792 Texture analysis allows quantitative parameters to be ex-

1878 Ravanelli Oct 2018 www.ajnr.org FIG 1. Example of freehand contouring of the tumor across its largest cross-sectional area on T2 (A), bϭ1000 (B), and ADC map (C) images. Tumor segmentation on ADC maps has been helped by side-by-side visualization of other sequences, including bϭ1000 DWI. tracted from CT, MR imaging, or PET images by applying various Hybrid Capture 2 test shows a sensitivity and specificity equiva- mathematic computations and algorithms.13 The technique has lent to that of the polymerase chain reaction.25 been recently introduced in medical imaging research and has provided promising results for cancer prognostication14-19 and a MR Imaging Protocol noninvasive signature of relevant genotypic and phenotypic tu- MR imaging was performed using a 1.5T scanner (Magnetom mor patterns.20-22 Bogowicz et al23 recently demonstrated that a Aera; Siemens, Erlangen, Germany). The MR imaging protocol radiomic CT signature can predict HPV status in head and neck included: axial TSE T2 (TR, 5570 ms; TE, 102 ms; slice thickness, cancer, even if its accuracy was not high enough to represent a 3 mm; matrix, 448 ϫ 224), axial echo-planar DWI (TR, 3900 ms; valid alternative to p16 immunohistochemistry (a surrogate bio- TE, 59 ms; slice thickness, 3 mm; matrix, 132 ϫ 132) with b-values marker) and direct viral DNA or messenger RNA studies, which of 0 and 1000 s/mm2, and contrast-enhanced 3D fat-saturated remain the criterion standard. To the best of our knowledge, no gradient-echo T1 (volumetric interpolated brain examination study has used MR imaging texture analysis to noninvasively de- [VIBE]) with an isotropic resolution of 0.7 mm. Apparent diffu- termine HPV status in oropharyngeal cancer. The first objective sion coefficient maps were automatically generated using an op- of this study was to test the correlation between ADC and HPV timized noise filter. status on a larger and more homogeneous OPSCC sample com- pared with previous studies. The second objective was to investi- Texture Analysis gate correlations between HPV status and a set of texture features Texture analysis was performed on primary tumors. MR images extracted from both morphologic and DWI sequences. were transferred to an off-line PC and analyzed using proprietary texture analysis software TexRAD (TexRAD; Cambridge, UK). MATERIALS AND METHODS Three head and neck radiologists in consensus (M.R., M.L., E.T.), Patients blinded to HPV status, drew an ROI on a single section of axial T2, This retrospective study included patients with untreated histo- ADC map, and postcontrast VIBE images (Fig 1), encompassing logically demonstrated T2–T4 OPSCC assessed by the Multidis- the primary tumor on its largest axial area. TexRAD software uses ciplinary Head and Neck Group of our institution between March the filtration-histogram method as described by Miles et al.26 The 2010 and April 2017. HPV status was determined by direct viral filtration-histogram method comprises an initial filtration step DNA study on tissue samples. Pretreatment MR imaging was that highlights image features of a specified size, followed by his- available for all patients. Patients with primary tumors too small togram analysis of the filtered image. The size of highlighted im- to be analyzed or with a low quality of MR images due to artifacts age features is denoted by the spatial scaling factor (SSF), ex- were excluded from image analysis. Tumor, Node, Metastasis pressed in millimeters. Histograms generated from unfiltered and classification was performed according to the eighth edition of the filtered images were quantified by the following parameters: mean TNM classification of head and neck cancer.24 (average value of the pixels within the ROI), SD (a measure of how much variation or dispersion exists from the average), mean of HPV Status Assessment positive pixels (average value of the pixels greater than zero within HPV status was evaluated with the digene HC2 High-Risk the ROI), entropy (a metric that reflects texture irregularity and HPV DNA Test (QIAGEN; https://www.qiagen.com/us/ positively correlates with tumor heterogeneity), skewness (a mea- shop/detection-solutions/human-pathogens/digene-hc2-high- sure of histogram asymmetry; positive values indicate histograms risk-hpv-dna-test/#orderinginformation), an in vitro nonradio- skewed to the right, while negative values indicate histograms active nucleic acid hybridization assay with signal amplification skewed to the left; skewness ϭ 0 indicates a perfectly symmetric using a chemiluminescent microtiter plate. This test is able to histogram), and kurtosis (a measure of the peakedness of the his- detect 18 HPV types, including high-risk (16, 18, 31, 33, 35, 39, 45, togram; positive kurtosis indicates a histogram that is more 51, 52, 56, 58, 59, 68) and low-risk types (6, 11, 42, 43, 44). The peaked than a Gaussian distribution, while negative kurtosis in- AJNR Am J Neuroradiol 39:1878–83 Oct 2018 www.ajnr.org 1879 Table 1: Baseline characteristics of patients and tumors Total (n = 59) HPV+ (n = 28) HPV− (n = 31) P Value Age (yr) 64.9 64.4 65.4 .904 Sex Male 43 19 24 .413 Female 16 9 7 Smoker 34/59 8/28 26/31 Ͻ.0001 Site 38, Palatine tonsil 20, Palatine tonsil 18, Palatine tonsil .611 17, Tongue base 7, Tongue base 10, Tongue base 3, Soft palate 1, Soft palate 2, Soft palate 1, Posterior wall 1, Posterior wall TNM classification TT2ϭ 25 T2 ϭ 16 T2 ϭ 9 .084 T3 ϭ 2T3ϭ 1T3ϭ 1 T4 ϭ 32 T4 ϭ 11 T4 ϭ 21 (16, T4a; 5, T4b) NN0ϭ 14 N0 ϭ 7N0ϭ 7 .002 N1 ϭ 15 N1 ϭ 13 N1 ϭ 2 N2 ϭ 18 N2 ϭ 6N2ϭ 12 (8, N2b; 4, N2c) N3 ϭ 12 N3 ϭ 2N3ϭ 10 (N3b ϭ 10) MM0ϭ 57 M0 ϭ 27 M0 ϭ 30 .144 M1 ϭ 2M1ϭ 1M1ϭ 1 Note:—HPVϩ indicates human papillomavirus-positive; HPVϪ, human papillomavirus-negative.

Table 2: Quantitative MR imaging features significantly different between patients positive and negative for HPV HPV− HPV+ P Value Variable No. Mean No. Mean Uncorrected Corrected Mean ADC 29 1045.2072 24 860.0675 .000112 .00168 MPP T2 SSF 4 31 149.4606 27 192.3400 .0064 .108 MPP T2 SSF 3 31 136.8981 27 173.5967 .0072 .108 Mean T2 SSF 2 31 17.6865 27 52.8400 .0182 .15525 Mean T2 SSF 3 31 34.4752 27 82.4037 .0207 .15525 Mean T2 SSF 1 31 5.8284 27 19.9081 .031 .1625 Mean T2 SSF 4 31 54.0761 27 107.7759 .0325 .1625 Kurtosis VIBE SSF 4 30 0.3773 26 Ϫ0.09423 .0358 .446 Note:—HPVϪ indicates human papillomavirus-negative; HPVϩ, human papillomavirus-positive; MPP, mean of positive pixels. dicates that the histogram is flatter than a Gaussian distribution). positive (47%), and 31 patients were HPV-negative (53%). Table In this study, SSFs of 1, 2, 3, and 4 mm were used. A total of 30 1 summarizes baseline patient characteristics. Twenty-six HPV- parameters [6 ϫ (unfiltered ϩ 4 SSF)] were computed for each negative lesions were found in smokers (class 1); 20 HPV-positive sequence. lesions, in nonsmokers (class 2); eight HPV-positive lesions, in smokers (class 3); and 5 HPV-negative lesions, in nonsmokers Statistical Analysis (class 4). Use of tobacco was observed in a significantly higher Texture parameters were compared between HPV-positive and ratio of patients with HPV-negative OPSCC (P Ͻ .0001, ␹2 test). HPV-negative groups using the Mann-Whitney U or Student t Age was not different between the 2 groups. test (when normally distributed). The Benjamini-Hochberg cor- MR imaging texture analysis was performed on all 3 sequences rection was applied to control the false discovery rate. The ␹2 test (T2, DWI, and VIBE) in 50 patients. DWI was not analyzed in 5 was used to compare categoric variables. Patients were divided in patients (because of artifacts in 3 and because the sequence was 4 subclasses, according to smoking and HPV status, which were not available in 2). The VIBE sequence was not analyzed in 3 considered as classification variables in a second step: smokers patients (2 because of artifacts, 1 because the sequence was not and HPV-negative (class 1), nonsmokers and HPV-positive (class available). In 1 patient, neither T2 nor DWI was analyzed because 2), smokers and HPV-positive (class 3), and nonsmokers and of low image quality. HPV-negative (class 4). Multivariable logistic regression was used to build a predictive model for HPV status. The statistical signif- Table 2 summarizes the texture features that were significantly icance level was set at P ϭ .05. Statistical tests were performed associated with HPV status. Mean ADC was the parameter with using MedCalc for Windows, Version 17.8.6 (MedCalc Software, the highest discriminatory power, while other parameters had sig- Mariakerke, Belgium). nificant but higher P values. After Benjamini-Hochberg correc- tion for the false discovery rate, only mean ADC maintained sta- RESULTS tistical significance. If one combined mean ADC and smoking Sixty-eight patients with OPSCC were enrolled; however, 4 were habit in multivariable logistic regression, the resulting model al- excluded because of the small size of the primary tumor and 5 for lowed correctly classifying 88.2% of cases, corresponding to an the low quality of MR images due to motion artifacts. Thus, 59 area under the receiver operating characteristic curve of 0.944 patients were analyzed (43 men and 16 women); 28 were HPV- (sensitivity, 83.3%; specificity, 92.6%) (Fig 2). When subclasses 1880 Ravanelli Oct 2018 www.ajnr.org determined by a combination of HPV status and history of smok- quently associated with higher Ki-67 levels,30 which have been ing were considered, some texture parameters were significantly found to be negatively correlated to ADC in several cancer different between class 3 (HPV-positive, smoking) and classes 1 types.31-35 The same aforementioned study by Swartz et al29 also and 2; class 4 (HPV-negative, nonsmokers) was excluded because demonstrated a negative correlation between ADC and Ki-67 of the very low patient numbers. In particular, patients in class 3 expression. (HPV-positive, smokers) had a significantly higher skewness: The increasing evidence for the relationship between the ADC SSF ϭ 2 mm on VIBE sequences compared with the other 2 classes value and HPV status is relevant given that the results from several (Table 3). This association maintained its statistical significance preliminary studies36-43 suggest an association between higher after Benjamini-Hochberg correction. tumor pretreatment ADC and a poor response to chemoradiation and prognosis in head and neck cancer, though none considered DISCUSSION HPV status in multivariable analysis. Therefore, our results em- Our study confirms the preliminary results obtained by previous phasize the need for future studies on DWI to include HPV status publications regarding the correlation between ADC derived as a possible important confounder for ADC in OPSCC. The pres- from DWI and HPV status in OPSCC; finding that HPV-positive ent study was performed on a larger patient cohort than previous tumors had a significantly lower ADC than HPV-negative tu- studies.10-12 The number of HPV-positive and HPV-negative tu- mors. Even if the reason for this finding remains unknown, sev- mors is better balanced (28 and 31, respectively) compared with eral possible explanations may be conceived. On the basis of their the group analyzed by Driessen et al,9 which included only 6 27 previous study, Driessen et al hypothesized that the low stromal HPV-positive tumors. Furthermore, the latter study included a volume observed in HPV-positive head and neck cancer could miscellanea of head and neck cancers, while our study was selec- explain the cancer lower ADC. Furthermore, cancer cell nests in tively directed to analyze oropharyngeal cancer. Different from HPV-driven cancer are often surrounded by zones of lymphoid articles published by Chan et al,10 Nakahira et al,11 and Schouten 28 cells, which could increase tissue cell density, thus leading to et al,12 HPV status in our study was defined by direct viral DNA lower ADC. This hypothesis is supported by a very recent pilot and messenger RNA studies rather than by p16 immunohisto- 29 study published by Swartz et al, who found a strongly significant chemistry. This is an important strength because p16 immuno- negative correlation between ADC and CD3-positive cell count in staining is a surrogate marker of HPV status and, despite its high 20 patients with OPSCC. sensitivity, has only moderate specificity because p16 may be con- In addition, HPV-positive OPSCC seems to be more fre- stitutively activated in HPV-negative OPSCC.44 Remarkably, the absolute ADC values that we found are compellingly lower than those obtained in the previously cited studies in both HPV-posi- tive and HPV-negative OPSCC. Even if the reason for these dis- crepancies cannot be fully explained, they may be conceivably imputed to differences in sequences and segmentation strategy. Kolff-Gart et al45 demonstrated that ADC values measured on different head and neck tissues differ significantly among differ- ent MR imaging systems and sequences. Juan et al46 demonstrated that non-echo-planar sequences (like those used by Schouten et al12) produce significantly higher ADC values for the major sali- vary glands compared with echo-planar sequences. Choices of different b-values and intrinsic signal-to-noise ratio might also

influence ADC. Finally, ROI segmentation on B0 images (as per- formed by Driessen et al9and Chan et al10) might include peritu- moral edema, leading to increased ADC mean values. This is the first study investigating the correlation between MR imaging texture features and HPV status in head and neck cancer. The relationship between texture parameters and HPV status was FIG 2. Receiver operating characteristic (ROC) curve of the predic- weaker than that observed between mean ADC and HPV status. tive model based on mean ADC and smoking status. The classification Even if some parameters showed raw P values Ͻ .05 (1 below .01), variable was HPV status, while the variable was the predicted proba- bility produced by multivariable logistic regression. AUC indicates they lost significance after correction for the false discovery rate, area under the curve. which is strongly recommendable when a large number of

Table 3: Quantitative MR imaging features significantly different between patients in class 3 (HPV؉ and smokers) and those in classes 1 (and 2 (smokers only and HPV؉ only, respectively Class 1 or 2 Class 3 P Value Variable No. Mean No. Mean Uncorrected Corrected Skewness VIBE SSF 2 42 0.165 8 0.769 .000798 .02394 Kurtosis VIBE SSF 2 42 0.4544 8 1.4763 .0076 .114 Skewness VIBE SSF 3 42 0.1713 8 0.5788 .0184 .184 Skewness VIBE SSF 1 42 0.1119 8 0.4038 .0364 .273

AJNR Am J Neuroradiol 39:1878–83 Oct 2018 www.ajnr.org 1881 variables without a priori validation are tested.47 Conversely, a illomavirus on the clinical presentation of oropharyngeal carci- strongly significant association was found between skewness noma in the United States. Laryngoscope 2017;127:2270–78 CrossRef SSF ϭ 2 mm (a descriptor of histogram asymmetry) on postcon- Medline 6. Huang YH, Yeh CH, Cheng NM, et al. Cystic nodal metastasis in trast VIBE and groups derived from the combination of HPV and patients with oropharyngeal squamous cell carcinoma receiving tobacco smoking status. In particular, the patients in the group of chemoradiotherapy: relationship with human papillomavirus sta- HPV-positive/smokers had a significantly higher skewness when tus and failure patterns. PLoS One 2017;12:e0180779 CrossRef comparing the HPV-positive/nonsmoker and HPV-negative/ Medline smoker groups. Even if this result has no obvious explanation, 7. Cantrell SC, Peck BW, Li G, et al. Differences in imaging character- istics of HPV-positive and HPV-negative oropharyngeal cancers: a it is interesting in line with previous observations that the com- blinded matched-pair analysis. AJNR Am J Neuroradiol 2013;34: bination of HPV infection and tobacco smoking leads to a 2005–09 CrossRef Medline 48 higher risk of treatment failure and an accumulation of ge- 8. Goldenberg D, Begum S, Westra WH, et al. Cystic lymph node me- netic mutations.49 tastasis in patients with head and neck cancer: an HPV-associated This study has some limitations. It is retrospective and based phenomenon. Head Neck 2008;30:898–903 CrossRef Medline on ROIs traced on a single slice rather than on volumetric tumor 9. Driessen JP, van Bemmel AJ, van Kempen PM, et al. Correlation of human papillomavirus status with apparent diffusion coefficient of analysis. Due to the relatively low number of patients, analysis of diffusion-weighted MRI in head and neck squamous cell carcino- texture data did not include complex classification algorithms, mas. Head Neck 2016;38(Suppl 1):E613–18 CrossRef Medline which would have required both training and validation sub- 10. Chan MW, Higgins K, Enepekides D, et al. Radiologic differences groups, each with sufficient numbers of patients. Thus, further between human papillomavirus-related and human papillomavi- information regarding relevant texture patterns could still be rus-unrelated oropharyngeal carcinoma on diffusion-weighted im- aging. ORL J Otorhinolaryngol Relat Spec 2016;78:344–52 CrossRef studied within our data. A more comprehensive approach will be Medline adopted when our sample size reaches an adequate size. Even if 11. Nakahira M, Saito N, Yamaguchi H, et al. Use of quantitative diffu- the identification of a noninvasive detection method for HPV sion-weighted magnetic resonance imaging to predict human pap- status was not the objective of the present study, the good perfor- illoma virus status in patients with oropharyngeal squamous cell mance of the simple regression model based on mean ADC and carcinoma. Eur Arch Otorhinolaryngol 2014;271:1219–25 CrossRef Medline smoking status (which led to correct classification in 88% of 12. Schouten CS, de Graaf P, Bloemena E, et al. Quantitative diffusion- cases) is an interesting result that deserves to be validated on ex- weighted MRI parameters and human papillomavirus status in ternal datasets. Furthermore, the accuracy of the model could be oropharyngeal squamous cell carcinoma. AJNR Am J Neuroradiol further improved by the addition of texture information. 2015;36:763–67 CrossRef Medline 13. Castellano G, Bonilha L, Li LM, et al. Texture analysis of medical CONCLUSIONS images. Clin Radiol 2004;59:1061–69 CrossRef Medline 14. De Cecco CN, Ganeshan B, Ciolina M, et al. Texture analysis as Our study confirms the correlation between ADC derived from imaging biomarker of tumoral response to neoadjuvant chemora- DWI and HPV status in OPSCC with a significantly lower ADC in diotherapy in rectal cancer patients studied with 3-T magnetic res- HPV-positive tumors compared to HPV-negative. onance. Invest Radiol 2015;50:239–45 CrossRef Medline Bases on these findings, we developed a simple predictive 15. Ganeshan B, Miles KA, Babikir S, et al. CT-based texture analysis poten- model based on ADC and smoking status that can be used as a tially provides prognostic information complementary to interim FDG-PET for patients with Hodgkin’s and aggressive non-Hodgkin’s non-invasive and cost-effective detection method for HPV status. lymphomas. Eur Radiol 2017;27:1012–20 CrossRef Medline This model deserves to be validated on external datasets and to be 16. Goh V, Ganeshan B, Nathan P, et al. Assessment of response to ty- perfected with other parameters to increase its sensitivity. rosine kinase inhibitors in metastatic renal cell cancer: CT texture A further noteworthy observation was that patients who are as a predictive biomarker. Radiology 2011;261:165–71 CrossRef both HPV-positive and smokers have significantly higher MRI Medline 17. Ravanelli M, Farina D, Morassi M, et al. Texture analysis of advanced skewness, which is in line with published literature. This result non-small cell lung cancer (NSCLC) on contrast-enhanced com- does not yet have an obvious explanation and therefore would puted tomography: prediction of the response to the first-line che- require confirmation by adding more patients and to be validated motherapy. Eur Radiol 2013;23:3450–55 CrossRef Medline on external datasets. 18. Smith AD, Gray MR, del Campo SM, et al. Predicting overall survival in patients with metastatic melanoma on antiangiogenic therapy and RECIST stable disease on initial posttherapy images using CT REFERENCES texture analysis. AJR Am J Roentgenol 2015;205:W283–93 CrossRef 1. Chaturvedi AK, Engels EA, Pfeiffer RM, et al. Human papillomavirus Medline and rising oropharyngeal cancer incidence in the United States. 19. Zhang H, Graham CM, Elci O, et al. Locally advanced squamous cell J Clin Oncol 2011;29:4294–301 CrossRef Medline carcinoma of the head and neck: CT texture and histogram analysis 2. Applebaum KM, Furniss CS, Zeka A, et al. Lack of association of allow independent prediction of overall survival in patients treated alcohol and tobacco with HPV16-associated head and neck cancer. with induction chemotherapy. Radiology 2013;269:801–09 CrossRef J Natl Cancer Inst 2007;99:1801–10 CrossRef Medline Medline 3. Ang KK, Harris J, Wheeler R, et al. Human papillomavirus and sur- 20. Aerts HJ, Velazquez ER, Leijenaar RT, et al. Decoding tumour phe- vival of patients with oropharyngeal cancer. N Engl J Med 2010;363: notype by noninvasive imaging using a quantitative radiomics ap- 24–35 CrossRef Medline proach. Nat Commun 2014;5:4006 CrossRef Medline 4. Ragin CC, Taioli E. Survival of squamous cell carcinoma of the head 21. Miles KA, Ganeshan B, Rodriguez-Justo M, et al. Multifunctional and neck in relation to human papillomavirus infection: review and imaging signature for V-KI-RAS2 Kirsten rat sarcoma viral onco- meta-analysis. Int J Cancer 2007;121:1813–20 CrossRef Medline gene homolog (KRAS) mutations in colorectal cancer. J Nucl Med 5. Stenmark MH, Shumway D, Guo C, et al. Influence of human pap- 2014;55:386–91 CrossRef Medline 1882 Ravanelli Oct 2018 www.ajnr.org 22. Weiss GJ, Ganeshan B, Miles KA, et al. Noninvasive image texture 37. Kim S, Loevner L, Quon H, et al. Diffusion-weighted magnetic res- analysis differentiates K-ras mutation from pan-wildtype NSCLC onance imaging for predicting and detecting early response to and is prognostic. PLoS One 2014;9:e100244 CrossRef Medline chemoradiation therapy of squamous cell carcinomas of the head 23. Bogowicz M, Riesterer O, Ikenberg K, et al. Computed tomography and neck. Clin Cancer Res 2009;15:986–94 CrossRef Medline radiomics predicts HPV status and local tumor control after defin- 38. Hatakenaka M, Shioyama Y, Nakamura K, et al. Apparent diffusion itive radiochemotherapy in head and neck squamous cell carci- coefficient calculated with relatively high b-values correlates with noma. Int J Radiat Oncol Biol Phys 2017;99:921–28 CrossRef Medline local failure of head and neck squamous cell carcinoma treated with 24. Huang SH, O’Sullivan B. Overview of the 8th Edition TNM Classi- radiotherapy. AJNR Am J Neuroradiol 2011;32:1904–10 CrossRef fication for Head and Neck Cancer. Curr Treat Options Oncol 2017; Medline 18:40 CrossRef Medline 39. Noij DP, Pouwels PJ, Ljumanovic R, et al. Predictive value of diffusion- 25. Clavel C, Masure M, Putaud I, et al. Hybrid capture II, a new sensi- weighted imaging without and with including contrast-enhanced tive test for human papillomavirus detection: comparison with hy- magnetic resonance imaging in image analysis of head and neck squa- brid capture I and PCR results in cervical lesions. J Clin Pathol 1998; mous cell carcinoma. Eur J Radiol 2015;84:108–16 CrossRef Medline 51:737–40 CrossRef Medline 40. Ohnishi K, Shioyama Y, Hatakenaka M, et al. Prediction of local 26. Miles KA, Ganeshan B, Hayball MP. CT texture analysis using the failures with a combination of pretreatment tumor volume and ap- filtration-histogram method: what do the measurements mean? parent diffusion coefficient in patients treated with definitive ra- Cancer Imaging 2013;13:400–06 CrossRef Medline diotherapy for hypopharyngeal or oropharyngeal squamous cell 27. Driessen JP, Caldas-Magalhaes J, Janssen LM, et al. Diffusion- carcinoma. J Radiat Res 2011;52:522–30 CrossRef Medline weighted MR imaging in laryngeal and hypopharyngeal carcinoma: 41. Preda L, Conte G, Bonello L, et al. Combining standardized uptake association between apparent diffusion coefficient and histologic value of FDG-PET and apparent diffusion coefficient of DW-MRI findings. Radiology 2014;272:456–63 CrossRef Medline improves risk stratification in head and neck squamous cell carci- 28. Westra WH. The morphologic profile of HPV-related head and noma. Eur Radiol 2016;26:4432–41 CrossRef Medline neck squamous carcinoma: implications for diagnosis, prognosis, 42. Lambrecht M, Van Calster B, Vandecaveye V, et al. Integrating pre- and clinical management. Head Neck Pathol 2012;6(Suppl 1):S48–54 treatment diffusion weighted MRI into a multivariable prognostic CrossRef Medline model for head and neck squamous cell carcinoma. Radiother Oncol 29. Swartz JE, Driessen JP, van Kempen PM, et al. Influence of tumor 2014;110:429–34 CrossRef Medline and microenvironment characteristics on diffusion-weighted im- 43. Ng SH, Lin CY, Chan SC, et al. Clinical utility of multimodality aging in oropharyngeal carcinoma: a pilot study. Oral Oncol 2018; imaging with dynamic contrast-enhanced MRI, diffusion- 77:9–15 CrossRef Medline weighted MRI, and 18F-FDG PET/CT for the prediction of neck 30. Liu J, Zhang M, Rose B, et al. Ki67 expression has prognostic signif- control in oropharyngeal or hypopharyngeal squamous cell car- icance in relation to human papillomavirus status in oropharyn- cinoma treated with chemoradiation. PLoS One 2014;9:e115933 geal squamous cell carcinoma. Ann Surg Oncol 2015;22:1893–900 CrossRef Medline CrossRef Medline 31. Heijmen L, Ter Voert EE, Nagtegaal ID, et al. Diffusion-weighted MR 44. Prigge ES, Arbyn M, von Knebel Doeberitz M, et al. Diagnostic accu- imaging in liver metastases of colorectal cancer: reproducibility racy of p16INK4a immunohistochemistry in oropharyngeal squa- and biological validation. Eur Radiol 2013;23:748–56 CrossRef mous cell carcinomas: a systematic review and meta-analysis. Int J Medline Cancer 2017;140:1186–98 CrossRef Medline 32. Schob S, Meyer J, Gawlitza M, et al. Diffusion-weighted MRI reflects 45. Kolff-Gart AS, Pouwels PJ, Noij DP, et al. Diffusion-weighted imag- proliferative activity in primary CNS lymphoma. PLoS One 2016;11: ing of the head and neck in healthy subjects: reproducibility of ADC e0161386 CrossRef Medline values in different MRI systems and repeat sessions. AJNR Am J 33. Shin JK, Kim JY. Dynamic contrast-enhanced and diffusion-weighted Neuroradiol 2015;36:384–90 CrossRef Medline MRIofestrogenreceptor-positiveinvasivebreastcancers:associations 46. Juan CJ, Chang HC, Hsueh CJ, et al. Salivary glands: echo-planar between quantitative MR parameters and Ki-67 proliferation status. J versus PROPELLER diffusion-weighted MR imaging for assess- Magn Reson Imaging 2017;45:94–102 CrossRef Medline ment of ADCs. Radiology 2009;253:144–52 CrossRef Medline 34. Surov A, Gottschling S, Mawrin C, et al. Diffusion-weighted imaging 47. Chalkidou A, O’Doherty MJ, Marsden PK. False discovery rates in in meningioma: prediction of tumor grade and association with PET and CT studies with texture features: a systematic review. PLoS histopathological parameters. Transl Oncol 2015;8:517–23 CrossRef One 2015;10:e0124165 CrossRef Medline Medline 48. Maxwell JH, Kumar B, Feng FY, et al. Tobacco use in human papil- 35. Surov A, Stumpp P, Meyer HJ, et al. Simultaneous (18)F-FDG-PET/ lomavirus-positive advanced oropharynx cancer patients related to MRI: associations between diffusion, glucose metabolism and his- increased risk of distant metastases and tumor recurrence. Clin topathological parameters in patients with head and neck squa- Cancer Res 2010;16:1226–35 CrossRef Medline mous cell carcinoma. Oral Oncol 2016;58:14–20 CrossRef Medline 49. Zevallos JP, Yim E, Brennan P, et al. Molecular profile of human 36. King AD, Thoeny HC. Functional MRI for the prediction of treat- papillomavirus–positive oropharyngeal squamous cell carcinoma ment response in head and neck squamous cell carcinoma: poten- stratified by smoking status. International Journal of Radiation On- tial and limitations. Cancer Imaging 2016;16:23 CrossRef Medline cology Biology Physics 2016;94:864 CrossRef

AJNR Am J Neuroradiol 39:1878–83 Oct 2018 www.ajnr.org 1883 ORIGINAL RESEARCH HEAD & NECK

Negative Predictive Value of NI-RADS Category 2 in the First Posttreatment FDG-PET/CT in Head and Neck Squamous Cell Carcinoma

X P. Wangaryattawanich, X B.F. Branstetter IV, X M. Hughes, X D.A. Clump II, X D.E. Heron, and X T.J. Rath

ABSTRACT BACKGROUND AND PURPOSE: FDG PET/CT has a high negative predictive value in patients with head and neck squamous cell carcinoma who responds completely to non-operative therapy. However, the treatment failure rate in patients with a partial but incomplete response is unclear. Our aim was to investigate the negative predictive value of the first posttreatment FDG-PET/CT in patients with head and neck squamous cell carcinoma with incomplete response interpreted as Neck Imaging Reporting and Data System (NI-RADS) category 2.

MATERIALS AND METHODS: We retrospectively identified patients with head and neck squamous cell carcinoma treated with chemo- radiation or radiation therapy with curative intent in our institution between 2008 and 2016. We included patients whose first posttreat- ment FDG-PET/CT was interpreted as showing marked improvement of disease but who had a mild residual mass or FDG avidity in either the primary tumor bed or lymph nodes (NI-RADS 2). The negative predictive value of FDG-PET/CT was calculated, including the 95% CI, using the Newcombe method. Two-year disease-free survival was the reference standard.

RESULTS: Seventeen of 110 patients (15%) experienced locoregional treatment failure within 2 years of completing treatment, yielding a negative predictive value of 85% (95% Cl, 77%–90%). The most common location of tumor recurrence was the cervical lymph nodes (59%). The median time interval between completion of therapy and treatment failure was 10 months (range, 5–24 months).

CONCLUSIONS: In patients with an incomplete response after treatment of head and neck squamous cell carcinoma, the negative predictive value of the first posttreatment FDG-PET/CT was 85%, which is lower than the 91% negative predictive value of FDG-PET/CT in patients with an initial complete response. Patients with an incomplete response (NI-RADS 2) should undergo more frequent clinical and imaging surveillance than patients with an initial complete response (NI-RADS 1).

ABBREVIATIONS: AJCC ϭ American Joint Committee on Cancer; HNSCC ϭ head and neck squamous cell carcinoma; IR ϭ incomplete response; NI-RADS ϭ Neck Imaging Reporting and Data System; NPV ϭ negative predictive value; TF ϭ treatment failure

ET/CT is critical to the management of patients with head and PET/CT in patients with a complete response has been estab- Pneck squamous cell carcinoma (HNSCC), given its high accu- lished.10 It was demonstrated in a prospective, randomized con- racy in pretreatment staging and the detection of persistent and trolled trial that concluded that planned neck dissection can be recurrent disease after therapy compared with CT or MR imaging, deferred in patients with HNSCC with a complete response on allowing salvage treatment to be initiated in a timely manner.1-12 initial posttreatment PET/CT following definitive chemoradia- The high negative predictive value (NPV) of posttreatment tion therapy.11,13 However, the treatment failure (TF) rate in pa- tients who have a partial, incomplete response (IR) on the initial posttreatment PET/CT remains controversial, and management Received April 4, 2018; accepted after revision June 28. 14,15 From the Departments of Radiology (P.W, B.F.B., M.H., T.J.R.), Otolaryngology of an initial IR is inconsistent. (B.F.B., M.H., T.J.R.), Biomedical Informatics (B.F.B.), and Radiation Oncology (D.A.C., The American College of Radiology has recently published the D.E.H.), University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania. Neck Imaging Reporting and Data System (NI-RADS), a standard- Paper previously presented at: Annual Meeting of the American Society of Neuro- radiology and the Symposium of the ASNR Foundation, June 2–7, 2018; Vancouver, ized radiologic reporting system for head and neck cancer imaging to British Columbia, Canada. facilitate communication between radiologists and referring physi- Please address correspondence to Tanya J. Rath, MD, UPMC Presbyterian Hospital, 200 Lothrop St, 2nd Floor, Suite 200 East Wing, Pittsburgh, PA 15213, cians and to determine the appropriate next management steps for e-mail: [email protected] an individual patient.16 This reporting system also assists in sharing Indicates article with supplemental on-line table. data among institutions, which may facilitate the advancement of http://dx.doi.org/10.3174/ajnr.A5767 head and neck cancer research. In NI-RADS, the results of posttreat-

1884 Wangaryattawanich Oct 2018 www.ajnr.org ment imaging surveillance are classified into 4 numeric categories vironment during the 50-minute radiotracer uptake phase. Axial based on imaging suspicion for residual or recurrent tumors (cate- PET and diagnostic contrast-enhanced CT images were obtained gory 1, no evidence of recurrence; category 2, low suspicion; category from the calvarial vertex through the upper thighs after urinary 3, high suspicion; category 4, definitive disease recurrence).16-18 voiding. Emission images were obtained at 50–60 minutes after The purpose of this study was to investigate the NPV of the radiopharmaceutical injection. Diagnostic CT images were ob- first posttreatment FDG-PET/CT in patients with HNSCC with tained 45 seconds after administration of 125 mL of intravenous an IR interpreted as NI-RADS category 2. contrast (iopidamol 370 mg I/mL, Isovue-370; Bracco, Princeton, New Jersey). CT parameters included the following: 120–30 kV- MATERIALS AND METHODS (peak); variable milliampere (AutomA; GE Healthcare); pitch, Study Design and Patient Selection 1.5–2; and collimation, 3.75-mm. The images of the head and We conducted a retrospective study that was approved by our neck were reconstructed in 2.5-mm slice thicknesses with a small institutional review board (PRO08120419) and was in compli- FOV, whereas the images of the thorax, abdomen, and pelvis were ance with the Health Insurance Portability and Accountability reconstructed in 3.75-mm slice thicknesses with a full-body FOV. Act. Medical records from our institutional Head and Neck On- cologic Data Repository were reviewed to include patients with Clinical Assessment, Treatment, and Surveillance histologically confirmed HNSCC treated with primary definitive Protocol chemoradiation therapy or radiation therapy at our institution All patients had staging contrast-enhanced imaging using PET/ between 2008 and 2016 with available contrast-enhanced staging CT, neck CT, or MR imaging. Patient treatment protocols, in- imaging and at least 24 months of clinical and radiographic fol- cluding radiation dose and chemotherapy regimen, were deter- low-up at our institution after the conclusion of treatment. Patients mined by the standard practice guidelines of the multidisciplinary who had non-squamous cell malignancies or a history of previously head and neck oncology team at our institution. After completion treated head and neck cancer were excluded. Pretreatment tumor of treatment, patients had clinical surveillance by means of phys- staging of all patients was performed using the American Joint ical examination and endoscopy every 6–8 weeks for the first year Committee on Cancer (AJCC) Cancer Staging Manual, 7th edi- per our institutional protocol. The first PET/CT scan was per- tion.19 We selected patients whose first posttreatment PET/CT formed approximately 8 weeks after the last course of chemora- diation therapy or radiation therapy, and subsequent PET/CT was performed 2–3 months after the completion of treatment and scans were then performed at 5, 8, and 14 months after therapy.20 was interpreted as showing marked improvement in size and/or If patients had clinical signs or symptoms suspicious for TF, FDG avidity of a locoregional tumor, but with persistent mild PET/CT and histologic confirmation were pursued outside the FDG avidity relative to background and/or small residual soft tissue usual imaging surveillance regimen. at either the treated primary tumor site or regional metastatic lymph nodes. These PET/CT findings were categorized as NI-RADS 2 (low Image Interpretation suspicion for residual viable tumor). These patients can be divided All PET/CT scans were interpreted in routine clinical workflow by into 3 major groups: 1) patients who had a partial IR at the pri- board-certified neuroradiologists with dedicated experience in mary tumor beds, with complete response at regional nodal head and neck radiology. Postprocessing fusion software (Mi- sites (P2N1); 2) patients who had a complete response at pri- rada; Mirada Medical, Denver, Colorado) was used to assist in mary tumor beds, but partial IR at regional nodal sites (P1N2); interpretation. The images were interpreted qualitatively, without and 3) patients who had a partial IR at both primary tumor specific standard uptake value thresholds. Patients were consid- beds and regional nodal sites (P2N2). ered appropriate for inclusion in this study if their first surveil- These patients were followed clinically and radiologically lance scan showed a marked decrease in the size and FDG avidity for 2 years after the conclusion of treatment to determine TF of the documented primary tumor and metastatic lymph nodes rates; TF was defined as the persistence of viable residual tumor with only mild residual soft-tissue abnormality and/or FDG up- or locoregional tumor recurrence after a disease-free interval. take at either the primary tumor or regional nodes. TF was confirmed by means of histopathology or unequivocal evidence of disease progression on subsequent imaging and Statistical Methods clinical evaluation. The primary outcome measure was 2-year disease-free survival. A 95% confidence interval for the negative predictive value for the PET/CT Parameters first surveillance PET/CT was calculated using the Newcombe 21 PET/CTs were performed on several PET/CT scanners (Discov- method. Disease-free survival was visualized with Kaplan-Meier ery, GE Healthcare, Milwaukee, Wisconsin; and Biograph mCT, survival curves. Statistical analyses were performed with SAS Ver- Siemens, Erlangen, Germany), ranging from 2 to 64 input chan- sion 9.4 (SAS Institute, Cary, North Carolina). nels. Except for water, patients fasted for at least 4–6 hours before the PET/CT scan and were instructed to avoid strenuous exercise RESULTS before the test. Serum glucose levels were obtained, and imaging We performed a first posttreatment PET/CT on 2077 patients in was deferred if glucose levels were Ͼ200 mg/dL. Each patient our institution between 2008 and 2016. Of these, 464 patients received 10–20 mCi of [18F] FDG dosed by body weight, after (22%) were classified as having a NI-RADS 2 response. One hun- which the patient remained seated or recumbent in a relaxed en- dred ten patients met the inclusion criteria, with a mean age of 59 AJNR Am J Neuroradiol 39:1884–88 Oct 2018 www.ajnr.org 1885 tients with residual FDG avidity had a (110 ؍ Patient characteristics (N Characteristics TF rate like that in the patients who had Sex no residual FDG avidity (16% [13/82] Male 85 (77%) versus 14% [4/28]). Female 25 (23%) There was no difference in predicting Age (range) (mean) (median) (yr) 30–87, 59, 58 Primary tumor location TF between NI-RADS 2 for the nodal Oropharynx 51 (46%) site versus NI-RADS 2 for the primary Oral cavity 19 (17%) tumor bed (87% versus 86%). Of the Larynx 19 (17%) 110 patients, 51 patients were scored as Hypopharynx 6 (6%) having NI-RADS 2 due to IR of meta- Nasopharynx 6 (6%) Unknown primary 5 (5%) static nodal disease (P1N2 and P2N2), Paranasal sinuses/nasal cavity 4 (3%) and 7 of these patients had TF, yielding HPV status (oropharynx) (n ϭ 51) an NPV of 86%. Seventy-five patients Positive 35 (69%) were scored as having NI-RADS 2 based Negative 8 (15.5%) on positive findings at the primary tu- Unknown 8 (15.5%) Tumor stage (AJCC Cancer Staging Manual, 7th ed) mor sites (P2N1 and P2N2), and 10 of T-stage these patients developed TF, yielding an T0 5 (5%) NPV of 87%. T1 22 (20%) Representative PET/CT images of a T2 30 (27%) T3 25 (23%) patient with TF following a posttreat- T4 28 (25%) ment examination scored as NI-RADS 2 N-stage are shown in Fig 1. N0 28 (25%) N1 17 (15%) N2 61 (57%) DISCUSSION N3 4 (3%) The NI-RADS classification system for M-stage surveillance imaging in HNSCC is used M0 107 (97%) to convey the degree of radiologic cer- M1 3 (3%) MX 0 (0%) tainty regarding the presence of recur- TNM stage rent or residual disease in treated pa- I 1 (1%) tients.16-18 Previous work has demon- II 9 (8%) strated that patients classified in the III 22 (20%) NI-RADS category 1 on their first sur- IV 78 (71%) Interval between the first posttreatment PET/CT and completion 3–46, 11, 9 veillance PET/CT have an NPV of of chemoradiation (range) (mean) (median) (wk) 91%.10 The results from the current Duration of follow-up (range) (mean) (median) (yr) 1.3–9.7, 4.9, 4.4 study indicate that the NPV in patients Total patients with residual or recurrent disease within 2 yr after 17/110 (15%) with NI-RADS category 2 is lower at completion of primary chemoradiation therapya 85%. Due to the higher incidence of TF, Note:—HPV indicates human papillomavirus; TNM, Tumor, Node, Metastasis. patients with an IR should undergo a NPV ϭ 85% (95% CI, 77%–90%). more frequent clinical and imaging sur- years. Most patients were men (n ϭ 85, 77%) with an oropharyn- veillance than patients with a complete response. For example, at geal primary tumor site (n ϭ 51, 46%) and stage IV disease (n ϭ our institution, patients who are categorized as NI-RADS 1 un- 78, 71%). Thirty-five of 51 patients with an oropharyngeal tumor dergo their next imaging surveillance after 6 months, whereas were positive for human papillomavirus (69%). Patient demo- patients categorized as NI-RADS 2 undergo imaging at 3 months. graphics; Tumor, Node, Metastasis staging; treatment; and fol- Figure 316 demonstrates a scheme for PET/CT surveillance at our low-up information are summarized in the Table. institution. This recommendation is applied to patients pri- Seventeen of the 110 patients (15%) experienced locoregional marily treated nonoperatively with definitive chemoradiation or TF within 2 years after treatment completion, yielding an NPV of radiation therapy. 85% (95% confidence interval, 77%–90%). Histopathologic con- The literature regarding the TF rates in patients with HNSCC firmation was obtained in 13 patients; 4 patients had no tissue with an initial IR is limited. There are no established treatment or confirmation but had unequivocal clinical and imaging evidence of TF. The characteristics of patients with TF are summarized in surveillance protocols for these patients. The most recent Na- 22 the On-line Table. The most common location of locoregional tional Comprehensive Cancer Network guidelines of 2018 rec- recurrence was the cervical lymph nodes (65%; 11/17). Fifty-three ommend either observation, fine needle aspiration, or planned percent (9/17) of patients with TF had oropharyngeal primaries, neck dissection if the initial 12-week posttreatment PET/CT most of which were positive (56%; 5/9) for human papillomavi- shows an equivocal response of nodal disease (ie, size of lymph rus. The median time interval between completion of therapy and node Ͻ1 cm with abnormal FDG uptake which is suspicious for locoregional TF was 10 months (range, 5–24 months). The pa- disease or size of the lymph node of Ͼ1 cm with no FDG uptake). 1886 Wangaryattawanich Oct 2018 www.ajnr.org PET uptake is more relevant than a re- sidual nodal mass in assessing therapy response. However, there was no differ- ence in the TF rates between the patients with and without residual FDG avidity in our study (16% [13/82] versus 14% [4/28]). In our series, 9 of 51 patients (18%) with oropharyngeal cancer who had an IR developed locoregional failure within 2 years. More than 90% of pa- tients in our series had TF within 5–16 months after the conclusion of treat- FIG 1. Patient 8. A 69-year-old woman with advanced oropharyngeal squamous cell carcinoma. A, ment. Our results suggest that closer im- Pretreatment PET/CT shows a large FDG-avid right faucial tonsil and lateral oropharyngeal wall aging and clinical surveillance in pa- tumor with an FDG-avid right level II nodal metastasis. B, At 8 months after completion of tients with an IR on initial PET/CT is therapy, there is complete response of the primary tumor but mild residual FDG uptake in the warranted and may need to extend to 16 treated right level II node. C, Surveillance PET/CT scan obtained at 16 months after treatment shows increased FDG avidity and size of the right level II node, consistent with regional treatment months to detect TF early with the goal failure, confirmed by salvage neck dissection. There is no disease recurrence at the primary tumor of optimizing patient outcomes. site. The TF rate of patients with NI- RADS 2 in our study is like that in the initial published performance of NI- RADS.17 Krieger et al17 analyzed a local recurrence rate of 58 of 618 targets that were scored NI-RADS 2. The overall rate of recurrence was 17.2%, with similar rates for the primary tumor bed and nodes. However, in our study, TF most frequently occurred in regional lymph nodes (65% [11/17]) versus the primary tumor site (41% [7/17]), with 1 patient having both local and regional recur- rence. In addition, the results of our study reflect the capability of NI-RADS in predicting TF rates in patients with NI-RADS category 2. We support the current effort to standardize radiology reporting of PET/CT in patients with HNSCC with the goals of improving com- munication among physicians and guid- ing the next step in management.16-18 FIG 2. Two-year disease-free survival curve of patients with HNSCC with an incomplete response This study has several limitations. It on the first posttreatment PET/CT scan. Time zero is defined as completion of treatment. is inherently limited by the retrospective study design resulting in variation in the Current evidence supports using PET/CT surveillance in patients timing of the first posttreatment PET/CT, which can impact the with HNSCC with an initial complete response following defini- false-positive rate, mainly due to radiation-induced inflamma- 9,11,14,23,24 11 tive treatment. Mehanna et al recently demonstrated, tion causing FDG avidity.1,8 In our series, most patients had the in a multicenter prospective randomized controlled trial, that first posttreatment PET/CT at 8–9 weeks, which has been shown there was no statistical difference in 2-year overall survival be- to have an accuracy similar to that of PET/CT performed 11–14 tween patients who underwent planned neck dissection versus weeks after treatment.20 Regarding tumor staging, we used the those who underwent PET/CT-guided surveillance following de- AJCC Cancer Staging Manual, 7th edition, because of the hetero- finitive nonoperative therapy. However, imaging surveillance re- sulted in fewer operations and substantial cost savings for the geneity of the clinical data and because it was used in the manage- nonoperative group. ment of the patients. Approximately 15% of the patients with Porceddu et al14 prospectively investigated the performance of oropharyngeal cancer were also not tested for human papilloma- PET-directed management of the neck in 112 patients with node- virus status, limiting the application of the AJCC Cancer Staging positive HNSCC whose primary tumors showed complete re- Manual, 8th edition,25 in some patients. Moreover, a standard sponse to therapy. The results of this study suggest that residual uptake value analysis was not used in PET/CT interpretation; AJNR Am J Neuroradiol 39:1884–88 Oct 2018 www.ajnr.org 1887 dicting the outcome in patients with head and neck cancer. Laryn- goscope 2014;124:2732–38 CrossRef Medline 7. Koshkareva Y, Branstetter BF 4th, Gaughan JP, et al. Predictive accuracy of first post-treatment PET/CT in HPV-related oropharyngeal squa- mous cell carcinoma. Laryngoscope 2014;124:1843–47 CrossRef Medline 8. Ho AS, Tsao GJ, Chen FW, et al. Impact of positron emission tomography/computed tomography surveillance at 12 and 24 months for detecting head and neck cancer recurrence. Cancer 2013;119:1349–56 CrossRef Medline 9. Meerwein CM, Queiroz M, Kollias S, et al. Post-treatment surveil- lance of head and neck cancer: pitfalls in the interpretation of FDG PET-CT/MRI. Swiss Med Wkly 2015;145:w14116 CrossRef Medline 10. McDermott M, Hughes M, Rath T, et al. Negative predictive value of surveillance PET/CT in head and neck squamous cell cancer. AJNR Am J Neuroradiol 2013;34:1632–36 CrossRef Medline 11. Mehanna H, Wong WL, McConkey CC, et al; PET-NECK Trial Man- agement Group. PET-CT surveillance versus neck dissection in ad- vanced head and neck cancer. N Engl J Med 2016;374:1444–54 FIG 3. University of Pittsburgh PET/CT Surveillance Flowchart for CrossRef Medline Head and Neck Squamous Cell Carcinoma. NI-RADS 1 indicates no 12. Chen YJ, Rath T, Mohan S. PET-computed tomography in head and evidence of recurrence; NI-RADS 2, low suspicion; NI-RADS 3, high neck cancer: current evidence and future directions. Magn Reson suspicion; NI-RADS 4, definitive disease recurrence. Imaging Clin N Am 2018;26:37–49 CrossRef Medline 13. Nayak JV, Walvekar RR, Andrade RS, et al. Deferring planned however, the repeatability of quantitative standard uptake value neck dissection following chemoradiation for stage IV head and measurements, particularly in lesions with low FDG uptake, has neck cancer: the utility of PET-CT. Laryngoscope 2007;117: been proved to be poor,26 and there is no established standard 2129–34 CrossRef Medline uptake value cutoff reliably distinguishing benign from malignant 14. Porceddu SV, Pryor DI, Burmeister E, et al. Results of a prospective tissue. Furthermore, the true benefits of this PET/CT surveillance study of positron emission tomography-directed management of residual nodal abnormalities in node-positive head and neck can- guideline for early detection of recurrent or residual disease and cer after definitive radiotherapy with or without systemic therapy. the potential impact on patients’ overall survival require further Head Neck 2011;33:1675–82 CrossRef Medline investigation, preferably with an additional long-term prospec- 15. Brizel DM, Prosnitz RG, Hunter S, et al. Necessity for adjuvant neck tive study. Last, our patients were derived from a single institution dissection in setting of concurrent chemoradiation for advanced with extensive experience in PET/CT imaging of head and neck head-and-neck cancer. Int J Radiat Oncol Biol Phys 2004;58:1418–23 CrossRef Medline cancer; therefore, the results may not be universally applicable. 16. Aiken AH, Farley A, Baugnon KL, et al. Implementation of a novel surveillance template for head and neck cancer: neck imaging re- CONCLUSIONS porting and data system (NI-RADS). J Am Coll Radiol 2016;13: In patients with incomplete response (NI-RADS 2) after treat- 743–46 CrossRef Medline ment of HNSCC, the NPV of the first posttreatment FDG- 17. Krieger DA, Hudgins PA, Nayak GK, et al. Initial performance of NI- RADS to predict residual or recurrent head and neck squamous cell PET/CT was 85%, which was lower than the 91% NPV of FDG- carcinoma. AJNR Am J Neuroradiol 2017;38:1193–99 CrossRef Medline PET/CT in patients with a complete response (NI-RADS 1). Patients 18. Aiken AH, Hudgins PA. Neck imaging reporting and data system. with an incomplete response should undergo more frequent clinical Magn Reson Imaging Clin N Am 2018;26:51–62 CrossRef Medline and imaging surveillance than patients with a complete response. 19. Edge SB, Byrd DR, Compton CC, et al. AJCC cancer staging manual. 7th ed. New York: Springer; 2010:21–100 20. Leung AS, Rath TJ, Hughes MA, et al. Optimal timing of first post- REFERENCES treatment FDG PET/CT in head and neck squamous cell carcinoma. 1. King KG, Kositwattanarerk A, Genden E, et al. Cancers of the oral cavity Head Neck 2016;38(Suppl 1):E853–58 CrossRef Medline and oropharynx: FDG PET with contrast-enhanced CT in the post- 21. Newcombe RG. Statistical applications in orthodontics, Part II: treatment setting. Radiographics 2011;31:355–73 CrossRef Medline confidence intervals for proportions and their differences. J Orthod 2. Tantiwongkosi B, Yu F, Kanard A, et al. Role of (18)F-FDG PET/CT 2000;27:339–40 CrossRef Medline in pre and post treatment evaluation in head and neck carcinoma. 22. Adelstein D, Gillison ML, Pfister DG, et al. NCCN Guidelines World J Radiol 2014;6:177–91 CrossRef Medline Insights: Head and Neck Cancers, Version 2.2017. J Natl Compr 3. Cheung PK, Chin RY, Eslick GD. Detecting residual/recurrent head Canc Netw 2017;15:761–70 CrossRef Medline neck squamous cell carcinomas using PET or PET/CT: systematic 23. Sjo¨vall J, Chua B, Pryor D, et al. Long-term results of positron emis- review and meta-analysis. Otolaryngol Head Neck Surg 2016;154: sion tomography-directed management of the neck in node-posi- 421–32 CrossRef Medline tive head and neck cancer after organ preservation therapy. Oral 4. Sagardoy T, Fernandez P, Ghafouri A, et al. Accuracy of (18) FDG Oncol 2015;51:260–66 CrossRef Medline PET-CT for treatment evaluation 3 months after completion of che- 24. Bird T, Barrington S, Thavaraj S, et al. (18)F-FDG PET/CT to assess moradiotherapy for head and neck squamous cell carcinoma: response and guide risk-stratified follow-up after chemoradiother- 2-year minimum follow-up. Head Neck 2016;38(Suppl 1):E1271–76 apy for oropharyngeal squamous cell carcinoma. Eur J Nucl Med CrossRef Medline Mol Imaging 2016;43:1239–47 CrossRef Medline 5. Sheikhbahaei S, Taghipour M, Ahmad R, et al. Diagnostic accuracy 25. Amin AB, Edge SB, Greene FL, et al. AJCC cancer staging manual. 8th of follow-up FDG PET or PET/CT in patients with head and neck ed. New York:Springer;2017:55–184 cancer after definitive treatment: a systematic review and meta- 26. de Langen AJ, Vincent A, Velasquez LM, et al. Repeatability of 18F- analysis. AJR Am J Roentgenol 2015;205:629–39 CrossRef Medline FDG uptake measurements in tumors: a metaanalysis. J Nucl Med 6. Chen SW, Hsieh TC, Yen KY, et al. Interim FDG PET/CT for pre- 2012;53:701–08 CrossRef Medline

1888 Wangaryattawanich Oct 2018 www.ajnr.org ORIGINAL RESEARCH HEAD & NECK

The Diagnostic Value of Diffusion-Weighted Imaging in Differentiating Metastatic Lymph Nodes of Head and Neck Squamous Cell Carcinoma: A Systematic Review and Meta-Analysis

X C.H. Suh, X Y.J. Choi, X J.H. Baek, and X J.H. Lee

ABSTRACT BACKGROUND: Accurate lymph node staging is crucial for proper treatment planning for metastasis in patients with head and neck squamous cell carcinoma.

PURPOSE: Our aim was to evaluate the diagnostic performance of DWI for differentiating metastatic cervical lymph nodes from benign cervical lymph nodes in patients with head and neck squamous cell carcinoma and to identify optimal cutoff values for ADC.

DATA SOURCES: A computerized literature search was performed to identify relevant original articles in Ovid MEDLINE and EMBASE.

STUDY SELECTION: Studies evaluating the diagnostic performance of DWI for differentiating metastatic cervical lymph nodes from benign cervical lymph nodes were selected.

DATA ANALYSIS: Diagnostic meta-analysis was conducted with a bivariate random-effects model, and a hierarchical summary receiver operating characteristic curve was obtained. Meta-regression was also performed.

DATA SYNTHESIS: Nine studies with 337 patients were included. In all studies, ADC values derived from metastatic lymph nodes were significantly lower than ADC values derived from benign lymph nodes. The median ADC cutoff value was 0.965 ϫ 10Ϫ3 mm2/s. The pooled sensitivity and specificity for the diagnostic performance of DWI in differentiating metastatic lymph nodes from benign lymph nodes were 90% (95% CI, 84%–94%) and 88% (95% CI, 80%–93%), respectively. In the meta-regression, sensitivity was significantly higher in the studies using a 3-mm slice thickness (93% [95% CI, 88%–98%]) than in studies using a slice thickness of Ͼ3 mm (86% [95% CI, 77%–95%], P Ͻ .01).

LIMITATIONS: A small number of studies were included in our meta-analysis.

CONCLUSIONS: DWI demonstrated high diagnostic performance for differentiating metastatic lymph nodes from benign lymph nodes in patients with head and neck squamous cell carcinoma, and the median ADC cutoff value was 0.965 ϫ 10Ϫ3 mm2/s. A 3-mm DWI slice thickness can provide a slight improvement in sensitivity.

ABBREVIATIONS: HNSCC ϭ head and neck squamous cell carcinoma; LN ϭ lymph node; NCCN ϭ National Comprehensive Cancer Network; QUADAS-2 ϭ Quality Assessment of Diagnostic Accuracy Studies-2

ymph node (LN) metastasis is an adverse prognostic factor in work-up in patients with HNSCC.1 CT and MR imaging are use- Lpatients with head and neck squamous cell carcinoma (HNSCC), ful for determining morphologic criteria, including shape, size, and accurate LN staging is crucial for proper treatment planning. The internal architecture, extracapsular extension, and vascular fea- National Comprehensive Cancer Network (NCCN) guidelines tures associated with LN metastasis; however, diagnostic perfor- recommend CT and/or MR imaging with contrast for the initial mance is limited, especially in normal-sized non-necrotic LNs.2,3 During the past decade, diffusion-weighted imaging has been used for differentiating metastatic cervical LNs from benign cer- Received February 28, 2018; accepted after revision July 7. vical LNs in patients with HNSCC. Some studies have reported a From the Department of Radiology and Research Institute of Radiology, University 4-12 of Ulsan College of Medicine, Asan Medical Center, Seoul, Republic of Korea. high diagnostic performance for DWI, whereas other studies 13,14 Please address correspondence to Young Jun Choi, MD, PhD, Department of Radi- have demonstrated disappointing results. In addition, vari- ology and Research Institute of Radiology, University of Ulsan College of Medi- ous cutoff values have been proposed for the apparent diffusion cine, Asan Medical Center, 86 Asanbyeongwon-Gil, Songpa-Gu, Seoul 138-736, Re- public of Korea; e-mail: [email protected] coefficient. Indicates article with supplemental on-line tables and Appendix. We considered it timely to review the DWI protocols, param- http://dx.doi.org/10.3174/ajnr.A5813 eters, and reported diagnostic performances because there are no

AJNR Am J Neuroradiol 39:1889–95 Oct 2018 www.ajnr.org 1889 published systematic reviews or meta-analyses on the topic, to our The risk of bias was assessed for each selected study, according knowledge. Therefore, we aimed to evaluate the diagnostic per- to the Quality Assessment of Diagnostic Accuracy Studies-2 formance of DWI for differentiating metastatic cervical LNs from (QUADAS-2) criteria.16 Two reviewers (C.H.S. and Y.J.C.) inde- benign cervical LNs in patients with HNSCC and to identify the pendently performed study selection, data extraction, and quality optimal ADC cutoff value. assessment.

MATERIALS AND METHODS Statistical Analysis This study adhered to the Preferred Reporting Items for System- A diagnostic meta-analysis of the DWI was conducted with a bi- 17-19 atic Reviews and Meta-Analyses guidelines.15 variate random-effects model. Individual study sensitivity/ specificity and pooled sensitivity/specificity were plotted using a Search Strategy coupled forest plot. The pooled positive likelihood ratio, negative A preliminary literature search demonstrated that various termi- likelihood ratio, and diagnostic odds ratio were calculated. “Pos- nology was used to indicate cervical LN metastasis, including itive likelihood ratio” was defined as the likelihood that a DWI “cervical nodal metastasis,” “malignant cervical lymph nodes,” result positive for differentiating metastatic LNs from benign LNs “metastatic cervical lymph node,” and “cervical lymphadenopa- would occur in patients with metastatic LNs. “Negative likelihood thy.” These synonyms for cervical LN metastasis were used in ratio” was defined as the likelihood that a DWI result negative for the search terms for Ovid MEDLINE and EMBASE: ((cervical differentiating metastatic LNs from benign LNs would occur in lymph node metastasis) OR (cervical nodal metastasis) OR patients without metastatic LNs. The “diagnostic odds ratio” was (malignant cervical lymph nodes) OR (metastatic cervical defined as the odds of having a positive DWI result in patients lymph node) OR (cervical lymphadenopathy)) AND ((diffu- with metastatic LNs compared with the odds of having a positive sion-weighted) OR (DWI) OR (apparent diffusion coefficient) DWI result in patients without metastatic LNs. A hierarchical OR (ADC)). The literature search was not limited to a publi- summary receiver operating characteristic curve with 95% confi- cation date or study setting but was limited to English-lan- dence and prediction regions was obtained, and the area under guage publications. Any additional relevant studies identified the hierarchical summary receiver operating characteristic curve were investigated, and the literature search was updated until was calculated. January 3, 2018. Heterogeneity across the studies was explored using the incon- sistency index (I2) and Cochran Q-statistics.20 I2 values of Ͼ50% Study Selection indicated the presence of heterogeneity across the studies.21 We used the following eligibility criteria: 1) patients with biopsy- Visual assessment of a coupled forest plot (inverse correlation proved HNSCC who underwent preoperative MR imaging in- indicating the presence of a threshold effect) and a Spearman cluding DWI, 2) histopathology as a reference standard, 3) provi- correlation (a coefficient of Ͼ0.6 indicating the presence of a sion of the diagnostic performance and corresponding ADC threshold effect) were performed to evaluate any threshold cutoff value for differentiating metastatic cervical LNs from be- effect (positive correlation between sensitivity and the false- nign cervical LNs, and 4) published original articles. Case reports/ positive rate).22 Visual assessment of the difference between series (including Ͻ10 patients), reviews, conference abstracts, the 95% confidence and prediction regions in the hierarchical and studies including other types of tumor (including nasopha- summary receiver operating characteristic curve (a large dif- ryngeal carcinoma or lymphoma), or a study population over- ference indicating heterogeneity) was also performed. The lapping other studies were excluded. Authors of the studies were presence of publication bias was assessed by a Deeks funnel contacted for further information when 2 ϫ 2 tables could not be plot asymmetry test,23 and a slope coefficient with P Ͻ .1 was acquired. considered significant small-study bias. Data Extraction and Quality Assessment Meta-regression was performed using the following covari- The following information was extracted from the selected stud- ates: 1) analysis method (per LN versus per patient); 2) the per- Ͻ ies using a standardized form: 1) study characteristics: authors, centage of metastatic LNs ( 36.7% [median value of the included publication years, institution, study period, study design, and studies] versus Ն36.7%); 3) underlying disease (HNSCC versus data analysis (per LN versus per patient); 2) demographic charac- oral squamous cell carcinoma); 4) study design (prospective ver- teristics: sample size, mean age, age range, sex, and proportion of sus retrospective); 5) consecutive enrollment; 6) number of read- metastatic LNs; 3) MR imaging characteristics: magnetic field ers (2 versus 1); 7) magnetic field strength (3T versus 1.5T); 8) strength, MR imaging vendor, MR imaging scanner, coil, DWI maximum b-value (Ͻ1000 versus 1000 s/mm2); 9) slice thickness sequence, b-values (seconds/square millimeter), TR, TE, slice (3 versus Ͼ3 mm); and 10) ADC measurement (whole node ver- thickness, interslice gap, matrix, FOV, number of signal acqui- sus single section). sitions, scan time, number of readers, experience of readers; All statistical analyses were conducted by one of the reviewers and 4) outcomes: a 2 ϫ 2 contingency table (number of true- (C.H.S., with 5 years of experience in conducting systematic re- positive, false-positive, false-negative, and true-negative re- views and meta-analyses) using commercially available software sults) demonstrating the presence of metastatic LNs according (STATA 15.0, StataCorp, College Station, Texas; and R statistical to the ADC values and optimal ADC cutoff values for differen- and computing software, Version 3.4.1; http://www.r-project. tiating metastatic from benign LNs. org/). P Ͻ .05 indicated statistical significance. 1890 Suh Oct 2018 www.ajnr.org RESULTS included in this meta-analysis and were considered for further Study Search analyses.4-12 Figure 1 provides an overview of the search strategy and study- selection procedure. After 15 non-English studies were excluded, Study Characteristics and Quality Assessment our search yielded 214 records, of which 35 articles remained after The relevant study characteristics are summarized in On-line Ta- screening of the titles and abstracts. The full text of these studies ble 1. Eight of 9 studies analyzed the diagnostic performance of was reviewed, and 26 studies were excluded as follows (On-line DWI per LN,4,5,7-12 whereas 1 study performed analysis on a per- Appendix): studies evaluating patients with enlarged cervical LNs patient basis.6 The number of included patients ranged from 16 to (not all patients had HNSCC [n ϭ 9]), studies including patients 80, and the number of LNs ranged from 34 to 651. There were 7 with non-HNSCC malignancy (nasopharyngeal carcinoma or prospective studies4-8,10,11 and 1 retrospective one,9 with the lymphoma [n ϭ 5]), a study population partially overlapping study design not being explicit in a further study.12 Informed other studies (n ϭ 4), studies that did not allow a 2 ϫ 2 contin- consent was obtained in 8 studies,4,6-12 as was approval by an gency table to be obtained (n ϭ 4), and studies not in the field of ethics committee or institutional review board.4-8,10-12 interest (n ϭ 4). There were no studies reporting the diagnostic The results of the methodologic quality assessment accord- performance of DWI without mentioning the corresponding ing to QUADAS-2 are presented in Fig 2. Most studies were ADC cutoff value. Ultimately, 9 studies with 337 patients were considered to have a low risk of bias and minimal concerns regarding applicability. Common weaknesses involved uncertainties in blinding to the reference standard when analyzing the MR imaging re- sults and a poorly documented time interval between MR imaging and the reference standard. In the patient-se- lection domain, 2 studies had a high risk of bias due to a case-control de- sign9 or inappropriate exclusion crite- ria.8 In the index test domain, 3 studies had an unclear risk of bias because no information was provided on blinding to the reference standard.6,7,12 In the reference standard and flow/timing domain, 1 study had a high risk of bias and a high concern regarding applica- bility because both histopathology and follow-up imaging results were used as a reference standard.9 No studies were excluded from the meta-analysis on the basis of the quality assessment. FIG 1. Flowchart depicting the literature search and study selection.

FIG 2. Grouped bar charts indicating methodologic quality according to the QUADAS-2 criteria and expressed as the percentage of studies meeting each criterion. For each quality domain, the proportions of studies suggesting a low, high, or unclear risk of bias and/or concerns regarding applicability are illustrated in green, red, and blue, respectively. AJNR Am J Neuroradiol 39:1889–95 Oct 2018 www.ajnr.org 1891 FIG 3. Coupled forest plots showing the diagnostic performance of each study. Vertical lines in the coupled forest plots show the pooled sensitivity and specificity.

MR Imaging Characteristics individual sensitivities ranged from 80% to 97%, and the individ- The detailed MR imaging parameters are summarized in On-line ual specificities ranged from 65% to 96%. Table 2. Four studies used 3T,4,7,8,11 and 5 studies used The pooled sensitivity and specificity for the diagnostic per- 5,6,9,10,12 1.5T MR imaging. Six studies used echo-planar imaging formance of DWI in differentiating metastatic LNs from be- 4,5,7,8,10,11 as a DWI sequence, with the other studies not being nign cervical LNs were 90% (95% CI, 84%–94%) and 88% 6,9,12 4,5,7,12 explicit. Two b-values were used for DWI in 4 studies ; (95% CI, 80%–93%), respectively (Fig 3). The pooled positive 6,8,9,11 10 3 b-values, in 4 studies ; and 6 b-values, in 1 study. Four likelihood ratio, negative likelihood ratio, and diagnostic odds 4,5,7,11 studies used a 3-mm slice thickness, and 4 studies used a 4- ratio were 7.3 (95% CI, 4.4–12.1), 0.11 (95% CI, 0.07–0.19), 6,8-10 or 5-mm slice thickness. and 64 (95% CI, 27–156), respectively. The area under the ROIs were drawn manually around each LN in all studies, with hierarchical summary receiver operating characteristic curve the ROIs being placed on solid portions and avoiding cystic or was 0.95 (95% CI, 0.93–0.97), which suggests high diagnostic necrotic portions. An ADC value covering the whole node was performance (Fig 4). obtained in 5 studies,4-6,8,9 while an ADC value for a single section Heterogeneity was present, with I2 values exceeding 50% for was obtained in 4 studies.7,10-12 The mean ADC was calculated in both sensitivity and specificity. Visual assessment of the coupled 8 studies,4-8,10-12 and the minimum ADC (the lowest value) was forest plots revealed no threshold effect, and the Spearman corre- calculated in 1 study.9 The smallest LN sizes for ADC calculation lation coefficient was Ϫ0.471 (95% CI, Ϫ0.865–0.281), also indi- were set at a minimal axial diameter of 2 mm,5,7 3 mm,11 4 mm,4 or 10 mm.8 cating no threshold effect. The slope coefficient for the Deeks funnel plot for differentiating metastatic LNs from benign LNs is ϭ Data Analysis presented in Fig 5 and suggests slight asymmetry in the data (P In all studies, ADC values derived from metastatic LNs were sig- .03) and possible publication bias. nificantly lower than ADC values derived from benign LNs. The The Table shows the results of the meta-regression to explore optimal ADC cutoff values varied slightly among individual stud- the influence of 10 covariates on pooled sensitivity and specificity. ies, ranging from 0.851 ϫ 10Ϫ3 mm2/s to 1.038 ϫ 10Ϫ3 mm2/s. Slice thickness was revealed to be a significant factor affecting The median ADC cutoff value was 0.965 ϫ 10Ϫ3 mm2/s. The study heterogeneity. Sensitivity was significantly higher in studies 1892 Suh Oct 2018 www.ajnr.org showed slightly higher sensitivity; however, the differences did not reach statistical significance.

DISCUSSION The present systematic review and meta-analysis demonstrated that in patients with HNSCC, the ADC derived from metastatic LNs was significantly lower than the ADC derived from benign LNs. The median ADC cutoff value was 0.965 ϫ 10Ϫ3 mm2/s. In addition, our study demonstrated a high pooled sensitivity and specificity for the diagnostic performance of DWI for differenti- ating metastatic from benign LNs in patients with HNSCC. The meta-regression revealed that studies using a 3-mm slice thick- ness had higher sensitivity than studies using a slice thickness of Ͼ3 mm. Therefore, DWI using a 3-mm slice thickness should be optimally considered for differentiating metastatic from benign cervical LNs. A dedicated sequence optimization is essential to obtain optimized DWI. In the meta-regression, sensitivity was signif- icantly higher in studies using a 3-mm slice thickness (93% [95% CI, 88%–98%]) than in studies using a slice thickness of Ͼ3 mm (86% [95% CI, 77%–95%], P Ͻ .01). A 3-mm slice thickness may help to detect smaller sized LNs. In addition, 3T MR imaging (92% [95% CI, 86%–98%]), the use of a maxi- mum b-value of 1000 s/mm2 (91% [95% CI, 86%–97%]), and ADC measurement of the whole node (93% [95% CI, 89–97]) FIG 4. Hierarchical summary receiver operating characteristic curve showed slightly higher sensitivity, though the differences did with 95% confidence and prediction regions of DWI for differentiat- not reach statistical significance. A previous report also men- ing metastatic LNs from benign LNs in patients with HNSCC. Each tioned that a high gradient strength substantially increases the circle indicates 1 included study. signal-to-noise ratio and that applying a larger number of b- values not only reduces the influence of noise propagation in ADC calculations but also decreases the risk of motion-related artifacts.10 When one uses DWI to differentiate metastatic from benign LNs in patients with HNSCC, use of a 3-mm slice thickness, a 3T scanner, a maximum b-value of 1000 s/mm2, and ADC measurement of the whole node should all be con- sidered to obtain a high diagnostic performance. Considerable effort is required to achieve standardization, and further stud- ies are needed. Among the included studies, the ADCs derived from met- astatic LNs were consistently lower than the ADCs derived from benign LNs. The lower ADC values are probably due to the tumor microstructure in metastatic LNs, which typically show a larger number of cells, cellular polymorphism, and in- creased mitosis in comparison with benign LNs; these charac- teristics may reduce the extracellular extravascular space and FIG 5. The Deeks funnel plot showing the presence of publication decrease the ADC value.24 In our study, the optimal ADC cut- bias. Numbers in circles refer to study number. ESS indicates ef- Ϫ Ϫ fective sample size. off values ranged from 0.851 ϫ 10 3 mm2/s to 1.038 ϫ 10 3 mm2/s, with 7 of 9 studies reporting optimal cutoff values be- Ϫ Ϫ using a 3-mm slice thickness (93% [95% CI, 88%–98%]) than tween 0.94 ϫ 10 3 mm2/s and 1.038 ϫ 10 3 mm2/s, which is a in studies using a slice thickness of Ͼ3 mm (86% [95% CI, relatively small variation. In addition, the median ADC cutoff 77%–95%], P Ͻ .01). Otherwise, the analysis method, percent- value was 0.965 ϫ 10Ϫ3 mm2/s. age of metastatic LNs, underlying disease, study design, con- We recognize that our study has several limitations. First, a secutive enrollment, number of readers, magnetic field small number of studies were included in our meta-analysis; strength, maximum b-value, and ROI used for ADC measure- therefore, we cannot evaluate all potential causes of heteroge- ment were not significant factors affecting heterogeneity. MR neity. Although we found that slice thickness was a significant imaging using a 3T scanner, a maximum b-value of 1000 factor affecting study heterogeneity, other technical aspects, s/mm2, and an ADC measurement of the whole node all including different TRs/TEs and different sets of b-values, may AJNR Am J Neuroradiol 39:1889–95 Oct 2018 www.ajnr.org 1893 Metaregression results Pooled Pooled Covariates Subgroups Sensitivity (95% CI) P Value Specificity (95% CI) P Value Analysis method Per LN 89% (84%–94%) .42 87% (80–93%) .43 Per patient 95% (85%–100%) 97% (89–100%) Percentage of metastatic LN Ͻ36.7% 86% (78%–94%) .20 86% (77–95%) .25 Ն36.7% 92% (88%–97%) 89% (81%–97%) Underlying disease HNSCC 91% (86%–96%) .29 88% (81%–95%) .37 Oral SCC 87% (76%–97%) 86% (74%–99%) Study design Prospective 91% (86%–96%) .74 89% (84%–94%) .15 Retrospective 85% (64%–100%) 68% (24%–100%) Consecutive enrollment Yes 92% (87%–96%) .30 91% (89%–94%) .92 No 86% (76%–96%) 68% (56%–80%) No. of readers 2 90% (84%–95%) .42 87% (80%–94%) .53 1 89% (76%–100%) 85% (70%–100%) Magnetic field strength 3T 92% (86%–98%) .12 87% (79%–96%) .11 1.5T 89% (82%–95%) 88% (80%–96%) Maximum b-value Ͻ1000 88% (79%–96%) .16 86% (76%–97%) .27 1000 91% (86%–97%) 89% (82%–96%) Slice thickness 3 mm 93% (88%–98%) Ͻ.01 86% (78%–94%) .49 Ͼ3 mm 86% (77%–95%) 93% (88%–99%) ADC measurement Whole node 93% (89%–97%) .19 89% (82%–97%) .28 Single section 86% (79%–94%) 86% (76%–95%) Note:—ADC indicates apparent diffusion coefficient; HNSCC, head and neck squamous cell carcinoma; LN, lymph node; SCC, squamous cell carcinoma. account for some portion of the heterogeneity. In addition, the owned by the National Comprehensive Cancer Network, Inc, see low number of included studies may limit the power to achieve www.nccn.org statistical significance. Second, publication bias was reported. 2. Curtin HD, Ishwaran H, Mancuso AA, et al. Comparison of CT and MR imaging in staging of neck metastases. Radiology 1998;207: One possible reason is that 2 studies showing negative results 123–30 CrossRef Medline ϫ were excluded because of the nonavailability of 2 2 contin- 3. van den Brekel MW, Castelijns JA, Snow GB. Detection of lymph gency tables.13,14 Therefore, our results should be interpreted node metastases in the neck: radiologic criteria. Radiology 1994;192: cautiously, and the high diagnostic performance of DWI may 617–18 CrossRef Medline have been overestimated. To overcome these limitations, 4. Barchetti F, Pranno N, Giraldi G, et al. The role of 3 Tesla diffusion- weighted imaging in the differential diagnosis of benign versus ma- we included a relatively homogeneous study population (ie, lignant cervical lymph nodes in patients with head and neck squa- biopsy-proved HNSCC) and performed an extensive meta-re- mous cell carcinoma. Biomed Res Int 2014;2014:532095 CrossRef gression using 10 covariates. Moreover, we applied recent ro- Medline bust methodology (hierarchical logistic regression model- 5. de Bondt RB, Hoeberigs MC, Nelemans PJ, et al. Diagnostic accuracy ing17-19) and reported our results according to prestigious and additional value of diffusion-weighted imaging for discrimina- guidelines (the Preferred Reporting Items for Systematic Re- tion of malignant cervical lymph nodes in head and neck squamous 15 cell carcinoma. Neuroradiology 2009;51:183–92 CrossRef Medline view and Meta-Analysis and the Handbook for Diagnostic Test 6. Goel V, Parihar PS, Parihar A, et al. Accuracy of MRI in prediction of 25 Accuracy Reviews published by the Cochrane Collaboration ). tumour thickness and nodal stage in oral tongue and gingivobuccal Nevertheless, caution should be used when applying our re- cancer with clinical correlation and staging. J Clin Diagn Res 2016; sults to daily clinical practice. 10:Tc01–05 CrossRef Medline 7. Lee MC, Tsai HY, Chuang KS, et al. Prediction of nodal metastasis in head and neck cancer using a 3T MRI ADC map. AJNR Am J Neuro- CONCLUSIONS radiol 2013;34:864–69 CrossRef Medline DWI demonstrated a high diagnostic performance for differentiating 8. Si J, Huang S, Shi H, et al. Usefulness of 3T diffusion-weighted MRI metastatic from benign cervical LNs in patients with HNSCC, and for discrimination of reactive and metastatic cervical lymph nodes the median ADC cutoff value was 0.965 ϫ10Ϫ3 mm2/s. A 3-mm slice in patients with oral squamous cell carcinoma: a pilot study. Den- thickness for DWI can slightly improve sensitivity. Further large pro- tomaxillofac Radiol 2014;43:20130202 CrossRef Medline spective multicenter studies are required to confirm these findings. 9. Stecco A, Buemi F, Cassara`A,etal.Comparison of retrospective PET and MRI-DWI (PET/MRI-DWI) image fusion with PET/CT and MRI-DWI in detection of cervical and endometrial cancer lymph Disclosures: Jung Hwan Baek—UNRELATED: Consultancy: Consultant of StarMed node metastases. Radiol Med 2016;121:537–45 CrossRef Medline (2017–2018) and RF Medical (2017), Comments: radiofrequency ablation of the thyroid 10. Vandecaveye V, De Keyzer F, Vander Poorten V, et al. Head and neck gland. squamous cell carcinoma: value of diffusion-weighted MR imaging for nodal staging. Radiology 2009;251:134–46 CrossRef Medline REFERENCES 11. Wendl CM, Mu¨ller S, Eiglsperger J, et al. Diffusion-weighted imag- 1. NCCN Clinical Practice Guidelines in Oncology (NCCN Guide- ing in oral squamous cell carcinoma using 3 Tesla MRI: is there a lines®) for Head and Neck Cancers V.2.2017. https://www.nccn.org/ chance for preoperative discrimination between benign and malig- professionals/physician_gls/pdf/head-and-neck.pdf. Accessed Janu- nant lymph nodes in daily clinical routine? Acta Radiol 2016;57: ary 23, 2018. To view the most recent and complete version of the 939–46CrossRef Medline guidelines, National Comprehensive Cancer Network®, NCCN®, 12. Zhong J, Lu Z, Xu L, et al. The diagnostic value of cervical lymph NCCN Guidelines®, and all other NCCN content that are trademarks node metastasis in head and neck squamous carcinoma by using 1894 Suh Oct 2018 www.ajnr.org diffusion-weighted magnetic resonance imaging and computed to- clinical researchers, Part I: general guidance and tips. Korean J Ra- mography perfusion. Biomed Res Int 2014;2014:260859 CrossRef diol 2015;16:1175–87 CrossRef Medline Medline 19. Lee J, Kim KW, Choi SH, et al. Systematic review and meta-analysis 13. Liang L, Luo X, Lian Z, et al. Lymph node metastasis in head and of studies evaluating diagnostic test accuracy: a practical review for neck squamous carcinoma: efficacy of intravoxel incoherent mo- clinical researchers, Part II: statistical methods of meta-analysis. tion magnetic resonance imaging for the differential diagnosis. Eur Korean J Radiol 2015;16:1188–96 CrossRef Medline J Radiol 2017;90:159–65 CrossRef Medline 20. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta- 14. Lim HK, Lee JH, Baek HJ, et al. Is diffusion-weighted MRI useful for analysis. Stat Med 2002;21:1539–58 CrossRef Medline differentiation of small non-necrotic cervical lymph nodes in pa- 21. Higgins JP, Thompson SG, Deeks JJ, et al. Measuring inconsistency tients with head and neck malignancies? Korean J Radiol 2014;15: in meta-analyses. BMJ 2003;327:557–60 CrossRef Medline 810–16 CrossRef Medline 22. Deville´ WL, Buntinx F, Bouter LM, et al. Conducting systematic 15. Liberati A, Altman DG, Tetzlaff J, et al. The PRISMA statement for reviews of diagnostic studies: didactic guidelines. BMC Med Res reporting systematic reviews and meta-analyses of studies that Methodol 2002;2:9 CrossRef Medline evaluate health care interventions: explanation and elaboration. 23. Deeks JJ, Macaskill P, Irwig L. The performance of tests of publica- Ann Intern Med 2009;151:W65–94 Medline tion bias and other sample size effects in systematic reviews of di- 16. Whiting PF, Rutjes AW, Westwood ME, et al; QUADAS-2 Group. agnostic test accuracy was assessed. J Clin Epidemiol 2005;58:882–93 QUADAS-2: a revised tool for the quality assessment of diagnos- CrossRef Medline tic accuracy studies. Ann Intern Med 2011;155:529–36 CrossRef 24. Koh DM, Collins DJ. Diffusion-weighted MRI in the body: applica- Medline tions and challenges in oncology. AJR Am J Roentgenol 2007;188: 17. Suh CH, Park SH. Successful publication of systematic review and 1622–35 CrossRef Medline meta-analysis of studies evaluating diagnostic test accuracy. Korean 25. Deeks JJ BP, Gatsonis C, eds. 2013 Cochrane Handbook for Systematic J Radiol 2016;17:5–6 CrossRef Medline Reviews of Diagnostic Test Accuracy, Version 1.0.0. The Cochrane Col- 18. Kim KW, Lee J, Choi SH, et al. Systematic review and meta-analysis laboration. http://srdta.cochrane.org/handbook-dta-reviews. Accessed of studies evaluating diagnostic test accuracy: a practical review for January 8, 2017

AJNR Am J Neuroradiol 39:1889–95 Oct 2018 www.ajnr.org 1895 ORIGINAL RESEARCH HEAD & NECK

Radiologic-Pathologic Correlation of Tumor Thickness and Its Prognostic Importance in Squamous Cell Carcinoma of the Oral Cavity: Implications for the Eighth Edition Tumor, Node, Metastasis Classification

X E.A.M. Weimar, X S.H. Huang, X L. Lu, X B. O’Sullivan, X B. Perez-Ordonez, X I. Weinreb, X A. Hope, X L. Tong, X D. Goldstein, X J. Irish, X J.R. de Almeida, X S. Bratman, X W. Xu, and X E. Yu

ABSTRACT BACKGROUND AND PURPOSE: Addressing the performance of an imaging-based parameter compared to a “gold standard” pathologic measurement is essential to achieve accurate clinical T-classification. Our aim was to determine the radiologic-pathologic tumor thickness correlation and its prognostic value in oral squamous cell carcinoma.

MATERIALS AND METHODS: All pathologic T1–T3 (seventh edition of the Cancer Staging Manual of the American Joint Committee on Cancer) oral squamous cell carcinomas diagnosed between 2010 and 2015 were reviewed. Radiologic tumor thickness was measured on preoperative CT or MR imaging blinded to pathology. The radiologic-pathologic tumor thickness correlation was calculated. The impact of the imaging-to-surgery time interval and imaging technique on the correlation was explored. Intra-/interrater reliability on radiologic tumor thickness was calculated. The correlation of radiologic-versus-pathologic tumor thickness and its performance as the seventh edition T-category modifier was evaluated. Multivariable analysis assessed the prognostic value of the radiologic tumor thickness for overall survival adjusted for age, seventh edition T-category, and performance status.

RESULTS: For 354 consecutive patients, the radiologic-pathologic tumor thickness correlation was similar for the image-to-surgery interval of Յ4.0 weeks (␳ ϭ 0.76) versus 4–8 weeks (␳ ϭ 0.80) but lower in those with more than an 8-week interval (␳ ϭ 0.62). CT and MR imaging had similar correlations (0.76 and 0.80). Intrarater and interrater reliability was excellent (0.88 and 0.84). Excluding 19 cases with an imaging-to-surgery interval of Ͼ8 weeks, 335 patients were eligible for further analysis. The radiologic-pathologic tumor thickness correlation was 0.78. The accuracy for upstaging the T-classification based on radiologic tumor thickness was 83% for pathologic T1 and 74% for pathologic T2 tumors. Multivariable analysis confirmed the prognostic value of radiologic tumor thickness (hazard ratio ϭ 1.5, P ϭ .02) for overall survival.

CONCLUSIONS: This study demonstrates a good radiologic-pathologic tumor thickness correlation. Intrarater and interrater reliability for radiologic tumor thickness was excellent. Radiologically thicker tumor was predictive of inferior survival.

ABBREVIATIONS: DOI ϭ depth of invasion; HR ϭ hazard ratio; OS ϭ overall survival; OSCC ϭ oral cavity squamous cell carcinoma; pDOI ϭ pathologic depth of invasion; pTT ϭ pathologic tumor thickness; rDOI ϭ radiologic depth of invasion; rTT ϭ radiologic tumor thickness; TNM ϭ tumor, node, metastasis; TT ϭ tumor thickness

umor thickness (TT) and depth of invasion (DOI) are inde- siveness (measured from the adjacent normal mucosal basement Tpendently prognostic in oral cavity squamous cell carcinoma membrane to the deepest point of tumor invasion), while TT (OSCC).1-7 Although often used interchangeably, the precise def- represents the distance from the tumor surface to the deepest initions of TT and DOI differ. DOI assesses primary tumor inva- point of invasion.8 Recognizing the prognostic significance and clinical relevance, the eighth edition of the Cancer Staging Manual of the American Joint Committee on Cancer tumor, node, metas- Received March 17, 2018; accepted July 9. From the Departments of Neuroradiology and Head and Neck Imaging (E.A.M.W., E.Y.), Radiation Oncology (S.H.H., B.O., A.H., L.T., S.B.), Biostatistics (L.L., W.X.), Pa- Please address correspondence to Shao Hui Huang, MD, MRT(T), MSc, Depart- thology (B.P.-O., I.W.), and Otolaryngology–Head and Neck Surgery/Surgical On- ment of Radiation Oncology, Princess Margaret Cancer Centre/University of cology (D.G., J.I., J.R.d.A.), Princess Margaret Cancer Centre/University of Toronto, Toronto, 610 University Ave, Toronto, ON, Canada, M5G 2M9; e-mail: shaohui. Toronto, Ontario, Canada. [email protected]; Emilie A.M. Weimar, MD, Department Neuroradiology and Head and Neck Imaging, Princess Margaret Cancer Centre/University of S.H.H., E.Y., and W.X. contributed equally to this project. Toronto, 610 University Ave, Toronto, ON, M5G 2M9, Canada; e-mail: The Bartley-Smith/Wharton Fund, the Gordon Tozer Fund, the Wharton Head and [email protected]/[email protected] Neck Translational Fund, the Dr. Mariano Elia Fund, the Joe’s Team Fund, and the Indicates article with supplemental on-line photo. Petersen Turofsky Fund at the Princess Margaret Foundation supported the aca- demic activities of B.O., S.H.H., L.T., and W.X. http://dx.doi.org/10.3174/ajnr.A5782

1896 Weimar Oct 2018 www.ajnr.org tasis (TNM) classification has differentiated both terms and in- against the saturated background. Nonenhanced T1 images best de- troduced DOI into the OSCC T-classification.9,10 lineated tumor margins with lower intermediate signal intensity con- Inclusion of DOI is applicable to both clinical and pathologic trasted against the brighter (fatty) signal of the adjacent tissues. To T-classifications, though its prognostic value is primarily derived appreciate the difference between rTT and rDOI, we also recorded from surgical specimens.5 Although mainly managed via an op- rDOI, measured from an “interpreted mucosal plane” across the eration, some patients with OSCC do not undergo an operation closest intact surface of the normal mucosa (Fig 1). If both CT and due to the high risk, functional considerations, and personal MR imaging were available, rTT and rDOI were measured on both choice. In these cases, radiologic measurement combined with imaging modalities. If the same imaging technique was available at clinical assessment is the only way to assess TT and DOI to stage a multiple time points, rTT and rDOI were measured on the examina- tumor when an operation is not undertaken. Therefore, address- tion most closely approximating the date of the operation. ing the performance of an imaging-based parameter compared MR imaging or CT or both were reviewed, and rTT and rDOI with a criterion standard pathologic measurement is essential were measured by the first author (E.A.M.W.) blinded to the his- to achieve an accurate T-classification. Robust data confirming topathologic findings. To ensure consistency of radiology-pathol- the reliability of measuring radiologic depth of invasion ogy rTT measurements, we consulted our pathologists (B.P.-O. (rDOI) versus pathologic depth of invasion (pDOI) do not and I.W.), who confirmed that the maximum pTT value recorded exist, likely due to the unavailability of the latter because insti- on synoptic pathology reports represented the value measured on tutions traditionally only reported pathologic TT (pTT) not the slice with the thickest tumor chosen after evaluating the entire pDOI. Several studies have reported that TT measured on MR gross tumor and all slices of a specimen. In the case of rTT mea- imaging11-14 or CT15 correlates well with pTT. However, the surements, we followed the same process (ie, going through the interrater and intrarater reliability and prognostic value of ra- entire series of scans to find the best orientation, axial, coronal, or diologic tumor thickness (rTT) remain elusive. sagittal, and the image slice that visually represented the “thick- Confirming the reliability of radiologic-versus-pathologic est” portion of tumor to measure). measurement of a parameter and its prognostic value is para- mount to ensuring feasible implementation of the eighth edition Statistical Analysis clinical T-classification for OSCC. Ideally, this requires a compar- To appreciate the practicality, we evaluated the difference in rTT ison of rDOI versus pDOI. However, because only pTT was avail- and rDOI in “exophytic,” “ulcerated,” and “flat” tumor. To avoid able in our institution during the study period, we confined the potential confounding from tumor growth during the “wait time radiologic-pathologic correlation to TT, though both rTT and period” to an operation, we calculated the rTT-pTT Spearman rDOI were recorded. We hypothesized that the rTT-pTT correla- correlation coefficient (␳) among Յ4.0-, 4.0- to 8.0-, and Ͼ8.0- tion could be indicative of the rDOI-pDOI correlation. Because week subgroups to determine the acceptable time interval. To pTT has a similar implication for the T-classification compared justify whether CT and MR imaging rTT measurements can be with pDOI,16 we further evaluated the prognostic value of both combined to increase study power, we compared the performance rTT and rDOI for overall survival (OS). of CT-versus-MR imaging on the rTT assessment. Finally, to de- termine the reliability and reproducibility of rTT measurements, ␬ MATERIALS AND METHODS interrater and intrarater reliability was assessed using the Cohen Study Population coefficient in a subset of patients. Blinded rTT re-assessment by Following ethics board approval, we reviewed all newly diagnosed the initial interpreter (E.A.M.W.) and a second experienced neu- pathologic T1-T3 (seventh edition) OSCCs treated with definitive roradiologist (E.Y.) was undertaken after a 3-month interval. surgery from 2010 to 2015. We included all OSCC subsites except After excluding cases with unacceptably protracted imaging- the lip (typically different etiology [ie, sun exposure] rather than to-surgery time intervals, a valid study cohort for rTT-pTT cor- smoking/drinking).17 Exclusion criteria included unavailable im- relation analysis and prognostic assessment was assembled. We aging or pTT, a Ͼ12-week imaging-to-surgery time interval, or calculated the Spearman correlation coefficient of rTT versus pTT unassessable rTT due to imaging artifacts. Clinical and pathologic and the shrinkage factor (dividing the mean of pTT by the mean information was obtained from our institutional data base,18 in of rTT) for the entire cohort, oral tongue subgroup, and other which pTT and outcomes were prospectively recorded. OSCC subgroup. To evaluate the performance of rTT as a poten- tial T-classification modifier to upstage the seventh edition T-cat- Image Analysis egory to the eighth edition, we calculated the diagnostic accuracy Occasionally (Ͻ1% of cases), preoperative CT was provided from of rTT (with or without adjusting for the shrinkage factor) versus referring institutions using 5-mm collimation. Standard MR im- pTT within the seventh edition T1 (TT Յ5 versus Ͼ5 mm) and T2 aging protocol in our center includes nonenhanced T1, T2, and (TT Յ10 versus Ͼ10 mm) tumors. Shrinkage factor was used to T2 fat-saturated sequences in axial, coronal, and sagittal planes account for potential tumor shrinkage during specimen process- with 3-mm thicknesses. We used outside MR imaging studies that ing and fixation. Finally, to assess the prognostic value of rTT and included contrast-enhanced T1 sequences. However, rTT was usu- rDOI and its implication for staging refinement, we calculate OS ally assessed via nonenhanced T1 and T2 sequences in the most ap- using Kaplan-Meier methods and compared within the seventh propriate plane perpendicular to the mucosal surface. T2 fat-satu- edition T1 (cutoff: Յ5 versus Ͼ5 mm) and T2 tumors (cutoff: rated images helped to identify tumor, especially if the lesion was Յ10 versus Ͼ10 mm) using the log-rank test. Multivariable anal- small because such lesions are accentuated by their brighter T2 signal ysis calculated the hazard ratio (HR) of the risk of death for rTT AJNR Am J Neuroradiol 39:1896–1902 Oct 2018 www.ajnr.org 1897 FIG 1. A schematic representation of 3 main types of tumors. A, Flat tumors have the interpreted mucosal plane at the same level as the tumor; thus, depth of invasion and tumor thickness should be measured identically (TT ϭ DOI). B, Exophytic/bulging tumors extend outward from the interpreted mucosal plane, leading to a situation in which tumor thickness can be larger than the depth of invasion (TT Ͼ DOI). C, Endophytic/ ulcerated tumors have a gap between the interpreted mucosal plane and the tumor surface, and depth of invasion can be larger than tumor thickness (DOI Ͼ TT). and rDOI separately, adjusting for age, seventh edition Tumor/ was similar between Ͻ4.0 weeks (␳ ϭ 0.76) and 4.1–8.0 weeks Node category, and Eastern Cooperative Oncology Group perfor- (␳ ϭ 0.80) (P ϭ .83), but it was nonsignificantly lower for Ͼ8 mance status. We also calculated the HR of the eighth edition weeks (␳ ϭ 0.62, P ϭ .69). T-category using rTT and rDOI as the seventh edition T-category modifier, separately adjusted for the aforementioned covariates. Comparison of the rTT-pTT Correlation on CT versus MR All statistical analyses were 2-sided, and a P value of Յ.05 was Imaging considered statistically significant. A total of 206 patients had preoperative CT, and 187 had MR imaging (49 had both CT and MR). While MR imaging showed RESULTS slightly better correlation, the difference was minor (␳ ϭ 0.80 Of 463 consecutive OSCCs during the study period, 109 were ex- versus 0.78 for all cases; 0.75 versus 0.67 for MR imaging/CT both cluded (lip tumors: n ϭ 6; imaging-to-surgery time interval Ͼ12 available cases) after adjusting for the imaging-to-surgery interval weeks: n ϭ 12; unavailable pathologic reports: n ϭ 26; nonassessable (P ϭ .83). Thus, we combined rTT on CT and MR imaging as a tumor due to imaging artifacts: n ϭ 65). The remaining 354 were composite rTT for subsequent analyses. eligible for exploratory analyses (On-line Figure). Interrater and Intrarater Reliability for rTT Primary Tumor Type and Difference in rTT-versus-rDOI Measurements Measurements On the basis of the power calculation, 85 cases (provided at least Both rTT and rDOI were measured on all scans. Most (311/354, 85% power to detect significant difference) were randomly se- 87.9%) were flat tumors, where rTT and rDOI yielded the same lected for the intrarater and interrater reliability assessment. The measurement. Only 36 (10%) were exophytic (rTT Ͼ rDOI) and intrarater and intrarater concordance of rTT was 0.88 (95% CI, 7 (2%) were ulcerated tumors (rTT Ͻ rDOI). The median differ- 0.83–0.92) and 0.84 (95% CI, 0.77–0.90), respectively. ences between rTT and rDOI were 4.4 mm (range, 0.2–17.6 mm) and 1.7 mm (range, 0.1–14.4 mm) for exophytic and ulcerated rTT-pTT Correlation and rTT as the T-Category Modifier tumors, respectively. After we excluded 19 cases with more than an 8-week imaging-to- surgery interval due to their suboptimal rTT-pTT concordance, Influence of the Imaging-to-Surgery Time Interval on the the remaining 335 cases (189 oral tongue and 146 other oral cavity rTT-pTT Correlation subsites) were eligible for further analyses. The clinical character- To explore the potential impact of the imaging-to-surgery time istic of these 335 cases are listed in Table 1. interval on the rTT-pTT correlation, we stratified the 354 cases The distribution of rTT versus pTT showed a clear linear cor- into 3 subgroups: 0–4 weeks (n ϭ 205, 58%), 4.1–8.0 weeks (n ϭ relation for the entire cohort as well as tongue and other subsites 130, 37%), and Ͼ8 weeks (n ϭ 19, 5%). The rTT-pTT correlation (Fig 2). The rTT-pTT correlation adjusted for the imaging-to- 1898 Weimar Oct 2018 www.ajnr.org Table 1: Clinical characteristics of 335 patients Covariate Entire Cohort Oral Tongue Other Subsitea P Valueb Total case No. 335 189 (56%) 146 (44%) Age (median) (range) (yr) 62 (22–96) 60 (22–96) 64 (28–96.4) .004c Sex .260 Female 129 (39%) 78 (41%) 51 (35%) Male 206 (61%) 111 (59%) 95 (65%) Smoking PY (median) (range) 10 (0–100) 5 (0–100) 20 (0–86) Ͻ.001c pT Category (7th edition) Ͻ.001c T1 119 (36%) 47 (25%) 72 (49%) T2 161 (48%) 101 (53%) 60 (41%) T3 55 (16%) 41 (22%) 14 (10%) pN Category (7th edition) Ͻ.001c N0 181 (54%) 83 (44%) 98 (67%) pNϩ 154 (46%) 106 (56%) 48 (33%) pTT (median) (range) (cm) 0.9 (0.1–4.0) 1.1 (0.1–4.0) 0.6 (0.1–3.9) Ͻ.001c rTT (median) (range) (cm) 1.0 (0.1–3.4) 1.2 (0.1–3.0) 0.6 (0.1–3.4) Ͻ.001c rTT on CT (median) (range) (cm) 1.0 (0.1–3.4) 1.3 (0.1–2.7) 0.7 (0.1–3.4) Ͻ.001c rTT on MR (median) (range) (cm) 1.2 (0.1–3.0) 1.3 (0.1–3.0) 0.3 (0.1–2.8) Ͻ.001c rDOI on CT (median) (range) (cm) 0.9 (0.0–3.1) 1.3 (0.1–2.7) 0.6 (0.0–3.1) Ͻ.001c rDOI on MR (median) (range) (cm) 1.1 (0.1–3.0) 1.3 (0.1–3.0) 0.3 (0.1–2.8) Ͻ.001c Note:—PY indicates pack-year; pNϩ, pathologic-positive lymph nodes. a Other oral cavity subsite included the following: n ϭ 75, floor of mouth; n ϭ 37, buccal mucosa; n ϭ 16, lower alveolar and gingiva; n ϭ 8, retromolar trigone; n ϭ 6, upper alveolar and gingiva; n ϭ 4, hard palate. b P value was for comparison between oral tongue and other oral cavity subsites. c Significant.

FIG 2. Distribution of radiologic and pathologic tumor thickness.

Table 2: Radiologic-pathologic tumor thickness concordance accuracy in the entire cohort was high for both T1 and T2 tumors adjusted for imaging-to-surgery time interval and diagnostic accuracy of rTT as the seventh edition T-category modifier (83% and 82% and 74% and 70% with and without correction for Spearman Correlation Shrinkage the shrinkage factor, respectively). rTT vs pTT Coefficient (␳) (95% CI) Factor Entire cohort (n ϭ 335) 0.78 (0.73–0.82) 0.81 The Prognostic Value of rTT and rDOI for Overall Survival Oral tongue (n ϭ 189) 0.74 (0.66–0.80) 0.90 The median follow-up was 3.6 years. A trend toward lower OS was Other subsites (n ϭ 146) 0.74 (0.65–0.80) 0.71 observed in thicker tumors within each seventh edition T-cate- gory: Three-year OS for T1_rTT Ͼ5mm(n ϭ 55) versus Յ5mm surgery time interval was 0.78 for the entire cohort and 0.74 for (n ϭ 64) was 78% versus 92% (P ϭ .13); T2_rTT Ͼ10 mm (n ϭ both oral tongue and other subsite subgroups with shrinkage fac- 95) versus Յ10 mm (n ϭ 66) was 67% versus 82% (P ϭ .19). Only tors of 0.81, 0.90, and 0.71, respectively (Table 2). Because 0.80 2 T3 tumors were Յ10 mm, and no deaths occurred. Three-year was the most commonly used shrinkage factor18 and almost iden- OS for T3_rTT Ͼ10 mm (n ϭ 53) was 49% (P ϭ .23). If one tical to 0.81 derived herein, we used 0.80 as a shrinkage factor to replaced rTT with rDOI, the results were almost identical: account for potential shrinkage of tumor during specimen pro- T1_rDOI Ͼ5mm(n ϭ 49) versus Յ5mm(n ϭ 70) was 78% cessing and fixation. versus 91% (P ϭ .16); T2_rDOI Ͼ10 mm (n ϭ 92) versus Յ10 Sensitivity, specificity, positive predictive value, negative pre- mm (n ϭ 69) was 66% versus 83% (P ϭ .14). Only 3 T3 tumors dictive value, and accuracy for the ability of rTT to upstage the had rDOI Յ10 mm, and no deaths occurred. Three-year OS for original size-based T-category from T1 to T2 (TT Ͼ0.5 cm) and T3_rDOI Ͼ10 mm (n ϭ 52) was 50% (P ϭ .23). Multivariable from T2 to T3 (TT Ͼ1.0 cm), with and without correction for the analysis confirmed, similar to pTT (HR 1.38, P ϭ .01), that both shrinkage factor of 0.80, are summarized in Table 3. The overall rTT and rDOI were prognostic for OS with HRs of 1.50 (1.06– AJNR Am J Neuroradiol 39:1896–1902 Oct 2018 www.ajnr.org 1899 Table 3: Diagnostic accuracy of rTT as the seventh edition ter formaldehyde fixation, like findings in other studies. Most T-category modifier interesting, the shrinkage factor was smaller for oral tongue com- No With pared with other subsite tumors (0.91 versus 0.70). This was also Shrinkage Shrinkage 13 Variable Factor Factor 0.8 observed by Lwin et al, who reported shrinkage factors of 0.87, rTT identifying 7th edition T1 0.65, and 0.59 for oral tongue, floor of mouth, and others, respec- tumor with pTT Ͼ 0.5 cm tively. We hypothesized that the tongue, an organ with more free Total case No. 119 119 margins, has less propensity to shrink than tumors that are more True positive 38 34 deeply embedded in surrounding tissues. True negative 59 65 Because the eighth edition TNM includes DOI for the clinical False positive 17 11 False negative 5 9 T-classification, confirming its reliability and prognostication Sensitivity 88% (75–96) 79% (64–90) clinically and radiologically is important because not all patients Specificity 78% (67–86) 86% (76–93) undergo an operation. Clinicians need to use both clinical assess- Positive predictive value 69% (55–81) 76% (60–87) ment and imaging to best determine the clinical T-classification Negative predictive value 92% (83–97) 88% (78–94) for this population. A practical challenge in assessing rDOI is the Accuracy 82% 83% rTT identifying 7th edition starting point of the “plumb line.” Pathologic assessment used the T2 tumor with pTT Ͼ 1.0 cm adjacent mucosal basement membrane, which is invisible on im- Total case No. 161 161 aging because the thickness of the oral mucosal epithelium is Ͻ0.5 True positive 64 54 mm,11 representing a negligible difference between the potential True negative 49 65 False positive 31 15 originating points of measurement (mucosal surface versus base- False negative 17 27 ment membrane). Correspondingly, for practical reasons, we Sensitivity 79% (69–87) 67% (55–77) proposed that imaging could use an interpreted mucosal plane Specificity 61% (50–72) 81% (71–89) across the “surface” of the adjacent normal mucosa for rDOI Positive predictive value 67% (57–77) 78% (67–87) measurement. Negative predictive value 74% (62–84) 71% (60–80) Accuracy 70% 74% Our study confirmed that both rDOI and rTT are indepen- dently associated with inferior OS in addition to seventh edition 2.12) (P ϭ .021) and 1.77 (1.22–2.56) (P ϭ .003), respectively T-category. When one applies rTT and rDOI to modify seventh (Fig 3). Finally, when the T-category was reclassified to the eighth edition T1 and T2, the separation in OS is evident (though non- edition T-category using either rTT or rDOI as T-modifiers (ad- significant due to an insufficient sample size). Nonetheless, the justed for age, N-category, and ECOG performance status), an trend supports consideration of either for modification of a pre- incremental HR with higher eighth edition T-category was appar- viously size-based T-classification. Furthermore, the multivari- ent: HRs for eighth edition T2, T3, and T4 compared with T1 ϭ able analysis confirmed that both the rTT- and rDOI-based eighth 1.88 (0.76–4.68), 2.86 (1.20–6.84), and 4.65 (1.86–11.6) based edition T-categories demonstrate a clear distinction in HRs be- on rTT (P Ͻ .001), and 1.67 (0.72–3.88), 2.88 (1.26–6.58), and tween each T-category, an essential requirement for staging. 4.2 (1.77–9.95) based on rDOI, respectively (P Ͻ .001). Thus, rTT can be a surrogate if rDOI is unavailable. The similar prognostic performance of rTT and rDOI echoes pathology- 16 DISCUSSION based findings. Dirven et al compared pTT and pDOI in 927 This large cohort study shows a high rTT-pTT correlation. By patients with OSCC and found that 79% of cases had a Ͻ1-mm means of pTT as a reference, the diagnostic accuracy of rTT for difference between both parameters and prognostic performance, upstaging seventh edition T1 and T2 tumors is good (Ͼ70%). An like those of T-category modifiers, and suggested that TT can be excellent intrarater and interrater reliability of measuring rTT used as a surrogate in retrospective studies for eighth edition confirms the reliability of recording this parameter in clinical TNM classification. practice. In addition, most OSCCs are flat tumors, in which the Study limitations include its retrospective nature and unavail- rTT and rDOI are identical. Both rTT and rDOI confer indepen- ability of pDOI. pTT was obtained prospectively from synoptic dent prognostic significance in addition to a size-based T-cate- reports and by convention, measured from the tumor surface to gory, supporting inclusion of either parameter in the eighth edi- the deepest point of invasion. pDOI was unavailable as the refer- tion TNM classification. Suboptimal rTT-pTT correlation when ence for rDOI. However, it was difficult to recognize whether a imaging is performed Ͼ8 weeks before an operation suggests that tumor had an exophytic or ulcerative component on imaging, and repeat staging imaging might be required to accurately depict tu- in most, rTT and rDOI were similar. Because spatial resolution mor extent when a protracted interval to treatment occurs. While remains a disadvantage on imaging compared with pathology, MR imaging–based rTT seems to have a slightly higher correla- very thin tumors that were not reliably measurable on imaging tion with pTT compared with CT-based rTT, the difference was were coded as rTT Ͻ1 mm for this study. Because the smallest nonsignificant, permitting the combining of CT and MR imaging cutoff for rDOI in the eighth edition T-category is 5 mm, this measurements to ensure a sufficiently large sample size. arbitrary coding is not expected to affect the reclassification of the The high rTT-pTT correlation (0.78) in this study is consistent T-category. with others.11-15,19-21 The correlation was similar for both oral Although most CT scans were obtained at our institution us- tongue and other OSCC subsites. We found that pTT is generally ing 2-mm slice thickness, rarely (Ͻ1% cases) did preoperative CT thinner than rTT, potentially attributable to tumor shrinkage af- from referring institutions use a 5-mm slice thickness. On careful 1900 Weimar Oct 2018 www.ajnr.org FIG 3. Prognostic value of rTT and rDOI for overall survival. AHR indicates adjusted hazard ratio, adjusted for age, seventh edition T-category, N-category, and Eastern Cooperative Oncology Group performance status. review of the data, our statisticians (W.X., L.L.) determined that REFERENCES this difference would not significantly influence our results. 1. Asakage T, Yokose T, Mukai K, et al. Tumor thickness predicts cer- vical metastasis in patients with stage I/II carcinoma of the tongue. CONCLUSIONS Cancer 1998;82:1443–48 CrossRef Medline 2. Fukano H, Matsuura H, Hasegawa Y, et al. Depth of invasion as a rTT measurement assessed by either CT or MR imaging is an predictive factor for cervical lymph node metastasis in tongue acceptable representation for pTT in OSCC. rTT can upstage the carcinoma. Head Neck 1997;19:205–10 CrossRef Medline seventh edition size-based T-category to the eighth-edition T-cat- 3. Mohit-Tabatabai MA, Sobel HJ, Rush BF, et al. Relation of thickness egory with good accuracy. Both rTT and rDOI are independent of floor of mouth stage I and II cancers to regional metastasis. Am J survival predictors and can stratify risk of death in addition to Surg 1986;152:351–53 CrossRef Medline 4. Spiro RH, Huvos AG, Wong GY, et al. Predictive value of tumor traditional tumor size. Similar to pTT for pDOI, rTT can be a thickness in squamous carcinoma confined to the tongue and floor surrogate for rDOI. Finally, we propose using the interpreted mu- of the mouth. Am J Surg 1986;152:345–50 CrossRef Medline cosal plane (ie, a plane crossing an adjacent normal mucosal sur- 5. Ebrahimi A, Gil Z, Amit M, et al. International Consortium for face) to measure rDOI. Outcome Research (ICOR) in Head and Neck Cancer. Primary tumor staging for oral cancer and a proposed modification in- ACKNOWLEDGMENTS corporating depth of invasion: an international multicenter ret- rospective study. JAMA Otolaryngol Head Neck Surg 2014;140: The author (B.O.) would like to acknowledge the O. Harold War- 1138–48 CrossRef Medline wick Prize of the Canadian Cancer Society for supporting his ac- 6. Huang SH, Hwang D, Lockwood G, et al. Predictive value of tumor ademic activities. thickness for cervical lymph-node involvement in squamous cell AJNR Am J Neuroradiol 39:1896–1902 Oct 2018 www.ajnr.org 1901 carcinoma of the oral cavity: a meta-analysis of reported studies. 15. Madana J, Laliberte´ F, Morand GB, et al. Computerized tomography Cancer 2009;115:1489–97 CrossRef Medline based tumor-thickness measurement is useful to predict postoper- 7. Kane SV, Gupta M, Kakade AC, et al. Depth of invasion is the most ative pathological tumor thickness in oral tongue squamous cell significant histological predictor of subclinical cervical lymph carcinoma. J Otolaryngol Head Neck Surg 2015;44:49 CrossRef node metastasis in early squamous carcinomas of the oral cavity. Medline Eur J Surg Oncol 2006;32:795–803 Medline 16. Dirven R, Ebrahimi A, Moeckelmann N, et al. Tumor thickness ver- 8. Lydiatt WM, Patel SG, O’Sullivan B, et al. Head and neck cancers: sus depth of invasion: analysis of the 8th edition American Joint major changes in the American Joint Committee on Cancer eighth Committee on Cancer Staging for oral cancer. Oral Oncol 2017;74: edition Cancer Staging Manual. CA Cancer J Clin 2017;67:122–37 30–33 CrossRef Medline CrossRef Medline 17. Park JO, Jung SL, Joo YH, et al. Diagnostic accuracy of magnetic 9. Amin M, Edge S, Greene F, et al. AJCC Cancer Staging Manual. 8th ed. resonance imaging (MRI) in the assessment of tumor invasion New York: Springer-Verlag; 2017 depth in oral/oropharyngeal cancer. Oral Oncol 2011;47:381–86 10. Brierley J, Gospodarowicz M, Wittekind C. UICC TNM Classification CrossRef Medline of Malignant Tumours. 8th ed. Southern Gate, Chichester, West Sus- 18. Wong K, Huang SH, O’Sullivan B, et al. Point-of-care outcome as- sex, UK: John Wiley & Sons, Ltd; 2017 sessment in the cancer clinic: audit of data quality. Radiother Oncol 11. Iwai H, Kyomoto R, Ha-Kawa SK, et al. Magnetic resonance de- 2010;95:339–43 CrossRef Medline termination of tumor thickness as predictive factor of cervical 19. Alsaffar HA, Goldstein DP, King EV, et al. Correlation between clin- metastasis in oral tongue carcinoma. Laryngoscope 2002;112: ical and MRI assessment of depth of invasion in oral tongue squa- 457–61 CrossRef Medline mous cell carcinoma. J Otolaryngol Head Neck Surg 2016;45:61 12. Lam P, Au-Yeung KM, Cheng PW, et al. Correlating MRI and histo- CrossRef Medline logic tumor thickness in the assessment of oral tongue cancer. AJR 20. Jung J, Cho NH, Kim J, et al. Significant invasion depth of early oral Am J Roentgenol 2004;182:803–08 CrossRef Medline tongue cancer originated from the lateral border to predict regional 13. Lwin CT, Hanlon R, Lowe D, et al. Accuracy of MRI in prediction of metastases and prognosis. Int J Oral Maxillofac Surg 2009;38:653–60 tumour thickness and nodal stage in oral squamous cell carcinoma. CrossRef Medline Oral Oncol 2012;48:149–54 CrossRef Medline 21. Goel V, Parihar PS, Parihar A, et al. Accuracy of MRI in prediction of 14. Preda L, Chiesa F, Calabrese L, et al. Relationship between histologic tumour thickness and nodal stage in oral tongue and gingivobuccal thickness of tongue carcinoma and thickness estimated from pre- cancer with clinical correlation and staging. J Clin Diagn Res 2016; operative MRI. Eur Radiol 2006;16:2242–48 CrossRef Medline 10:TC01–05 CrossRef Medline

1902 Weimar Oct 2018 www.ajnr.org BRIEF/TECHNICAL REPORT HEAD & NECK

MR Imaging of the Facial Nerve through the Temporal Bone at 3T with a Noncontrast Ultrashort Echo Time Sequence

X J.P. Guenette, X R.T. Seethamraju, X J. Jayender, X C.E. Corrales, and X T.C. Lee

ABSTRACT SUMMARY: The pointwise encoding time reduction with radial acquisition (PETRA) ultrashort echo time MR imaging sequence at 3T enables visualization of the facial nerve from the brain stem, through the temporal bone, to the stylomastoid foramen without intravenous contrast. Use of the PETRA sequence, or other ultrashort echo time sequences, should be considered in the MR imaging evaluation of certain skull base tumors and perhaps other facial nerve and temporal bone pathologies.

ABBREVIATIONS: CNR ϭ contrast-to-noise ratio; IR-FSPGR ϭ inversion recovery-prepared fast spoiled gradient recalled-echo; PETRA ϭ pointwise encoding time reduction with radial acquisition; UTE ϭ ultrashort echo time

he cisternal, canalicular, labyrinthine, geniculate, tympanic, sistent image quality over a wider range of conditions than similar Tand mastoid segments of the facial nerve are either not detect- sequences.7 This report describes the evaluation of the facial nerve able or only faintly visible on noncontrast T1-weighted MR im- segments through the temporal bone using both the default PETRA ages.1 Moreover, although contrast enhancement of the sur- protocol and a shortened 4-minute PETRA protocol. rounding venous plexus can aid in facial nerve visualization, the 2 extent of enhancement is variable and unreliable. We hypothe- MATERIALS AND METHODS sized that visibility of the facial nerve could be improved with As part of a prospective study designed to evaluate the extracranial ultrashort echo time (UTE) imaging. UTE sequences capture sig- facial nerve, an MR imaging examination of the brain and face 3 nal from rapidly decaying short-T2 tissue, such as cortical bone that included a PETRA sequence was performed in 8 healthy sub- 4 and middle ear ossicles and, therefore, should capture signal jects (6 men, 2 women; 31 Ϯ 8 years of age) between January and from the petrous and mastoid portions of the temporal bone. March 2018. All subjects provided informed consent. The study Moreover, peripheral nerves have detectable signal in the ultra- was approved by our institutional review board and performed in 5 short T2 spectrum, so UTE imaging should provide visualization compliance with the Health Insurance Portability and Account- of the facial nerve. Finally, UTE imaging minimizes air-related ability Act. susceptibility artifacts6 and would therefore minimize artifacts from the middle ear cavity and mastoid air cells that may contrib- Imaging Protocols ute to nonvisualization of the facial nerve on spin-echo and gra- All imaging was performed on a Magnetom Prisma 3T MR imag- dient-echo sequences. The pointwise encoding time reduction ing system with a 64-channel head/neck coil (Siemens, Erlangen, with radial acquisition (PETRA) UTE sequence provides more con- Germany). The default PETRA protocol as installed by Siemens (Table 1) was performed on 7 subjects and 14 facial nerves. Due to Received April 25, 2018; accepted after revision June 3. the 6-minute duration of the protocol, parameter modifica- From the Division of Neuroradiology (J.P.G., J.J., T.C.L.) and Division of Otolaryn- tions were tested to reduce imaging time while retaining the gology–Head and Neck Surgery (C.E.C.), Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts; and Siemens Medical Solutions USA, Inc. signal-to-noise ratio and contrast-to-noise ratio (CNR) of the (R.T.S.), Boston, Massachusetts. facial nerve and surrounding structures. A modified 4-minute This study was supported by the American Society of Head and Neck Radiology protocol (Table 1) was performed on 5 subjects and 10 facial through the 2017 William N. Hanafee Research Grant. Jagadeesan Jayender was supported by the National Institute of Biomedical Imaging and Bioengineering, nerves. Imaging with both PETRA protocols was performed on National Institutes of Health, through grant number P41EB015898. 4 subjects. Please address correspondence to Jeffrey P. Guenette, MD, Division of Neuroradi- ology, Brigham and Women’s Hospital, 75 Francis St, Boston, MA 02135; e-mail: [email protected]; @gunetty Imaging Analysis Indicates open access to non-subscribers at www.ajnr.org Two independent reviewers scored the visibility of the cisternal, http://dx.doi.org/10.3174/ajnr.A5754 canalicular, labyrinthine, geniculate, tympanic, and mastoid seg-

AJNR Am J Neuroradiol 39:1903–06 Oct 2018 www.ajnr.org 1903 ments of the facial nerve and the greater superficial petrosal nerve ebellum. The scored data from both reviewers were summed and on each side of each subject on all series of default PETRA and averaged for each facial nerve segment for both the default PETRA 4-minute PETRA images. One reviewer was a neuroradiologist and 4-minute PETRA images. Using sample size, means, and SDs, with 10 years’ attending-level experience, and 1 reviewer was a we calculated independent-sample t tests to compare the signal postgraduate year-5 radiology resident completing a year-long intensity of the facial nerve on the 4-minute PETRA protocol neuroradiology mini-fellowship. To allow indirect comparison images, as scored for this study, with the signal intensity of the with a prior study that evaluated these facial nerve segments using facial nerve on the IR-FSPGR images, as scored in the study by a 3D inversion recovery fast spoiled gradient recalled (IR-FSPGR) Dehkharghani et al.1 sequence,1 we scored the signal intensity as follows: 0, no signal; 1, To determine the SNR and CNR, in all 4 subjects in whom faint visualization; 2, signal equivalent to that of the normal cer- both protocols were performed, we drew ROIs on a single slice over the following structures of the default and 4-minute Table 1: Parameters for default and 4-minute PETRA sequences PETRA images: facial nerve tympanic segment, ossicle, middle Default 4-Minute ear cavity, mastoid, brain stem, and extracorporeal back- TR 1 3.32 ms 3.32 ms ground region lateral to the temporal bone. The SNR and CNR TR 2 2250 ms 2250 ms were calculated as follows: SNR ϭ Mean Signal Intensity of the TE 0.07 ms 0.07 ms ϭ Ϫ TI 1300 ms 2000 ms Region/SD of Background Noise; CNR SNR Region 1 SNR TI 2 900 ms 900 ms Region 2. Averages 1 1 Slice thickness 0.9 mm 0.9 mm FOV 300 ϫ 300 300 ϫ 300 RESULTS Matrix 320 ϫ 320 320 ϫ 320 All evaluated segments of the facial nerve were visible by both ϫ ϫ ϫ ϫ Voxel size 0.9 0.9 0.9 mm 0.9 0.9 0.9 mm reviewers (Figure) with a score of either 1 or 2 with 90.5% overall Bandwidth 400 kHz 400 kHz Flip angle 6° 6° reviewer agreement. The signal intensity of all facial nerve seg- Radial views 60,000 40,000 ments was significantly greater for the 4-minute PETRA protocol Acquisition time 5:57 minutes 4:05 minutes images compared the IR-FSPGR images (Table 2). SNR and CNR values were consis- tently similar or higher with the 4-min- ute PETRA protocol compared with the default PETRA protocol (Tables 3 and 4).

DISCUSSION This study demonstrates that noncon- trast UTE imaging with the PETRA se- quence can provide visualization of the facial nerve from the brain stem to the stylomastoid foramen. Such visualiza- tion is not possible with more routine sequences currently used in skull base or internal auditory canal imaging.1,2 The PETRA sequence is a proved alternative to an IR-FSPGR sequence in pediatric brain imaging8 and could perhaps simi- larly substitute for IR-FSPGR sequences in skull base imaging. It is known that ultrashort T2 (Ͻ1 ms) MR signal can be detected in pe- ripheral nerves and imaged. This signal originates from protons in myelin phos- pholipids5 and likely accounts, at least in part, for the consistent visibility of the facial nerve with the PETRA sequence. In addition, it is likely that artifacts re- FIGURE. Normal right facial nerve in a 27-year-old healthy female volunteer. Axial oblique (A and lated to the air/tissue interfaces in the B) and sagittal oblique (C) noncontrast MR images obtained with the 4-minute PETRA protocol temporal bone interfere with detection show the cisternal, canalicular, labyrinthine, geniculate, tympanic, and mastoid segments of the of signal from the facial nerve on more right facial nerve (white arrows) and a portion of an ossicle (black arrow). D, Sagittal oblique image obtained with the default PETRA protocol shows a slightly sharper but similar appearance routine sequences. Minimization of air- to C. related susceptibility artifacts with the 1904 Guenette Oct 2018 www.ajnr.org Table 2: Signal intensity comparison of PETRA and IR-FSPGR sequences as previously reporteda IR-FSPGR (from P Value (4-Minute Default PETRA 4-Minute PETRA Dehkharghani et al1) PETRA vs IR-FSPGR) Cisternal 1.89 (0.31) 2.00 (0.00) 0.86 (0.42) Ͻ.001 Canalicular 1.86 (0.36) 2.00 (0.00) 0.83 (0.44) Ͻ.001 Labyrinthine 1.79 (0.42) 1.85 (0.37) 0.88 (0.39) Ͻ.001 Geniculate 2.00 (0.00) 2.00 (0.00) 1.03 (0.22) Ͻ.001 Tympanic 1.96 (0.19) 2.00 (0.00) 0.95 (0.32) Ͻ.001 Superficial Petrosal 2.00 (0.00) 2.00 (0.00) 0.96 (0.30) Ͻ.001 Mastoid 2.00 (0.00) 2.00 (0.00) 1.01 (0.30) Ͻ.001 a To allow direct comparison with the data published by Dehkharghani et al,1 except for the P values, we report data as mean (SD) of assigned signal intensity, in which each facial nerve segment was assigned a value of 0–2 (0, no detectable signal; 1, faint visualization; 2, signal equivalent to normal brain).

Table 3: Signal-to-noise ratio for default and 4-minute PETRA sequences Subject C Subject D Subject F Subject G Default 4-Minute Default 4-Minute Default 4-Minute Default 4-Minute PETRA PETRA PETRA PETRA PETRA PETRA PETRA PETRA Facial nerve 125 165 113 159 88 140 99 173 Ossicle 58 116 65 114 39 94 57 118 Middle ear 5 51 31 75 36 83 14 59 Mastoid 36 55 62 106 28 54 42 63 Brain stem 168 281 155 290 151 277 176 300

Table 4: Contrast-to-noise ratio for default and 4-minute PETRA sequences Subject C Subject D Subject F Subject G Default 4-Minute Default 4-Minute Default 4-Minute Default 4-Minute PETRA PETRA PETRA PETRA PETRA PETRA PETRA PETRA Facial nerve/mastoid 89 111 51 43 75 69 57 88 Facial nerve/middle ear 120 114 82 67 65 46 85 91 Ossicle/mastoid 22 62 3 7 13 32 15 44 Ossicle/middle ear 53 65 34 31 4 9 43 47

UTE sequences6 may reduce such interference and may also con- MR imaging evaluation of certain skull base tumors and perhaps tribute to the consistent visibility of the facial nerve with the other facial nerve and temporal bone pathologies. PETRA sequence. Comparison of temporal bone imaging with multiple different UTE sequences could help elucidate the bio- Disclosures: Jeffrey P. Guenette—RELATED: Grant: American Society of Head and physical properties that allow imaging of the facial nerve through Neck Radiology.* Ravi Teja Seethamraju—UNRELATED: Employment: Siemens Med- ical Solutions USA, Inc.; Stock/Stock Options: Siemens Medical Solutions USA, Inc. the temporal bone and help optimize the technique. Jayender Jagadeesan—RELATED: Grant: National Institutes of Health, Comments: The default PETRA protocol runs nearly 6 minutes. We through the National Institute of Biomedical Imaging and Bioengineering grant num- shortened the protocol to 4 minutes by reducing the radial ber P41EB015898*; UNRELATED: Board Membership: Navigation Sciences; Consul- tancy: Navigation Sciences; Grants/Grants Pending: Siemens Research Grant*; Pat- acquisitions from 60,000 to 40,000. To recover SNR and CNR, ents (Planned, Pending or Issued): system for localizing deformable tumors*; Stock/ we increased the first TI, which is used in the pointwise acqui- Stock Options: Navigation Sciences. Thomas C. Lee—RELATED: Grant: American sition of central k-space data.6,8 Our increase of the TI from Society of Head and Neck Radiology, Comments: This study was supported by the American Society of Head and Neck Radiology through the 2017 William N. Hanafee 1300 to 2000 ms would be expected to increase signal in the Research Grant*. *Money paid to the institution. brain and nerve tissues at some expense to gray and white matter contrast. These parameter changes resulted in mild sig- nal artifacts in the middle ear cavity, demonstrated by higher REFERENCES middle ear cavity SNR. 1. Dehkharghani S, Lubarsky M, Aiken AH, et al. Redefining normal The major limitation of this study is that imaging was per- facial nerve enhancement: healthy subject comparison of typical en- formed with a single MR imaging system on a small sample of hancement patterns—unenhanced and contrast-enhanced spin- healthy subjects and does not prove reproducibility on other MR echo versus 3D inversion recovery-prepared fast spoiled gradient- imaging systems or in a clinical setting. In addition, the compar- echo imaging. AJR Am J Roentgenol 2014;202:1108–13 CrossRef Medline ison of signal intensities between PETRA and IR-FSPGR se- 2. Hong HS, Yi BH, Cha JG, et al. Enhancement pattern of the normal quences was made across studies performed by different research- facial nerve at 3.0 T temporal MRI. Br J Radiol 2010;83:118–21 ers with different subjects and equipment. CrossRef Medline 3. Robson MD, Gatehouse PD, Bydder M, et al. Magnetic resonance: an CONCLUSIONS introduction to ultrashort TE (UTE) imaging. J Comput Assist To- A 3T MR imaging 4-minute noncontrast PETRA protocol enables mogr 2003;27:825–46 CrossRef Medline 4. Naganawa S, Nakane T, Kawai H, et al. Visualization of middle ear visualization of the facial nerve from the brain stem, through the ossicles in elder subjects with ultra-short echo time MR imaging. temporal bone, to the stylomastoid foramen. Use of the PETRA Magn Reson Med Sci 2017;16:93–97 CrossRef Medline sequence, or other UTE sequences, should be considered in the 5. Horch RA1, Gore JC, Does MD. Origins of the ultrashort-T2 1H NMR AJNR Am J Neuroradiol 39:1903–06 Oct 2018 www.ajnr.org 1905 signals in myelinated nerve: a direct measure of myelin content? in zero echo time MRI: principles compared. Magn Reson Med 2018; Magn Reson Med 2011;66:24–31 CrossRef Medline 79:2036–45 CrossRef Medline 6. Grodzki DM, Jakob PM, Heismann B. Ultrashort echo time imaging 8. Aida N, Niwa T, Fujii Y, et al. Quiet T1-weighted pointwise encod- using pointwise encoding time reduction with radial acquisition ing time reduction with radial acquisition for assessing myelina- (PETRA). Magn Reson Med 2012;67:510–18 CrossRef Medline tion in the pediatric brain. AJNR Am J Neuroradiol 2016;37: 7. Froidevaux R, Weiger M, Brunner DO, et al. Filling the dead-time gap 1528–34 CrossRef Medline

1906 Guenette Oct 2018 www.ajnr.org ORIGINAL RESEARCH HEAD & NECK

CT versus MR Imaging in Estimating Cochlear Radiation Dose during Gamma Knife Surgery for Vestibular Schwannomas

X A.M. Faramand, X H. Kano, S. Johnson, X A. Niranjan, X J.C. Flickinger, and X L.D. Lunsford

ABSTRACT BACKGROUND AND PURPOSE: Leksell stereotactic radiosurgery is an effective option for patients with vestibular schwannomas. Some centers use a combination of stereotactic CT fused with stereotactic MR imaging to achieve an optimal target definition as well as minimize the radiation dose delivered to adjacent structures that correlate with hearing outcomes. The present prospective study was designed to determine whether there is cochlear dose variability between MR imaging and CT.

MATERIALS AND METHODS: Fifty consecutive patients underwent stereotactic radiosurgery for vestibular schwannomas. Dose-planning was performed using high-definition fused stereotactic MR imaging and stereotactic CT images. The 3D cochlear volume was determined by delineating the cochlea on both CT and T2-weighted MR imaging. The mean radiation dose, maximum dose, and 3- and 4.20-Gy cochlear volumes were identified using standard Leksell Gamma Knife software.

RESULTS: The median mean radiation dose delivered to the cochlea was 3.50 Gy (range, 1.20–6.80 Gy) on CT and 3.40 Gy (range, 1–6.70 Gy) on MR imaging (concordance correlation coefficient ϭ 0.86, r2 ϭ 0.9, P Յ .001). The median maximum dose delivered to the cochlea was 6.7 Gy on CT and 6.6 Gy on MR imaging (concordance correlation coefficient ϭ 0.89, r2 ϭ 0.90, P Յ .001). Dose-volume histograms generated from CT and MR imaging demonstrated a strong level of correlation in estimating the 3- and 4.20-Gy volumes (concordance correlation coefficient ϭ 0.81, r2 ϭ 0.82, P Յ .001 and concordance correlation coefficient ϭ 0.87, r2 ϭ 0.89, P Յ .001).

CONCLUSIONS: Both MR imaging and CT provide similar cochlear dose parameters. Despite the reported superiority of CT in identifying bony structures, high-definition MR imaging alone is sufficient to identify the radiation doses delivered to the cochlea.

ABBREVIATIONS: CCC ϭ concordance correlation coefficient; SRS ϭ stereotactic radiosurgery; VS ϭ vestibular schwannomas

estibular schwannomas (VS), also known as acoustic neuro- radiation fall-off in the structures surrounding the tumor target. The Vmas, are benign tumors that most commonly arise from the aim of SRS is tumor control with minimal collateral damage to adja- vestibular portion of cranial nerve VIII, the vestibulocochlear cent cranial nerve and brain stem structures. Multiple reports dem- nerve. The most common presenting symptoms of VS are hearing onstrate successful long-term SRS outcomes.5,6 1 loss, tinnitus, and imbalance. Depending on the presentation, Factors such as patient age at the time of SRS, hearing status the options currently available for the management of this tumor before the procedure, tumor size, the interval between diagnosis include observation with serial imaging, surgical resection, and and treatment, and cochlear radiation dose have been found to radiosurgery. Leksell gamma knife stereotactic radiosurgery (SRS) influence hearing preservation rates.6-9 Cochlear dose remains (Elekta Instruments, Stockholm, Sweden) is a widely accepted treat- the only variable that can be modified during treatment planning 2-4 ment technique for VS. SRS involves delivering highly focused to improve hearing-preservation rates. radiation to the 3D tumor volume in a single session, with rapid The use of MR imaging for dose-planning in SRS for VS has been shown to be safe and efficient.10 However, the concern that

Received December 7, 2017; accepted after revision July 1, 2018. MR imaging accuracy may be affected by magnetic susceptibility From the Department of Neurological Surgery and Radiation Oncology, University issues has led many centers to use both stereotactic CT and ste- of Pittsburgh Medical Center, Pittsburgh, Pennsylvania. reotactic MR imaging to achieve a more accurate target as well as Please address correspondence to Hideyuki Kano, MD, PhD, Department of Neu- cochlear definition.11-13 Whether bone window CT provides su- rological Surgery, University of Pittsburgh, Suite B-400, UPMC Presbyterian, 200 Lothrop St, Pittsburgh, PA, 15213; e-mail: [email protected] perior resolution of the cochlea compared with high-definition http://dx.doi.org/10.3174/ajnr.A5808 T2 MR imaging is not known.14,15

AJNR Am J Neuroradiol 39:1907–11 Oct 2018 www.ajnr.org 1907 In this study, we compared the definition of the cochlear vol- cochlea were generated from the available CT and T2 images, and ume determined by MR imaging with the cochlear volume de- the mean and maximum doses were obtained for each patient. fined by CT to detect whether any variance in the radiation dose Additionally, the cochlear volume receiving 3- and 4.2-Gy radia- delivered to the cochlea was detectable. tion doses was obtained using the dose-volume histogram gen- erated by the software. After the treating neurosurgeon, radi- MATERIALS AND METHODS ation oncologist, and the medical physicist approved the Patient Population treatment plan, the treatment began. The median gross tumor Between May 2016 and October 2017, fifty consecutive patients volume as determined by postgadolinium T1 MR imaging was with VS underwent MR imaging and CT-guided SRS at the Uni- 0.97 cm3 (range, 0.02–16.5 cm3). The median margin dose was versity of Pittsburgh Medical Center. There were 26 men and 24 12 Gy (range, 11–12.50 Gy), and the median maximum dose was 24 women with a median age of 60 years (range, 28–77 years) at the Gy (range, 15.5–25 Gy). The median prescription isodose was 50% time of SRS. The most common presentation was unilateral hear- (range, 50%–70%). After the procedure, all patients were given 20- to ing loss (94%). The median duration of symptoms before SRS was 40-mg of IV methylprednisolone. All patients were discharged on the 18 months. The median speech discrimination score and pure- same day. Only 4-mm isocenters were used in the internal auditory tone average at the time of SRS were 64% and 37 dB, respectively. canal. Hearing level was classified as Gardner Robertson grades I and II in 32 patients (64%). SRS was the primary management in 46 Statistical Analysis patients (92%). Three patients had SRS for residual tumor after Continuous features were summarized using median and range. surgery, and 1 patient, for recurrent tumor after prior SRS. The Pearson r2 and the Lin concordance correlation coefficient (CCC) were used to assess the correlation of volumes and doses Radiosurgery Technique between the 2 imaging modalities. The CCC ranged between Ϫ1 Radiosurgery was performed using the Perfexion or ICON and 1, with values indicating perfect negative or positive concor- gamma knife models (Elekta). Our radiosurgical technique has dance, respectively. Statistical analysis was performed with SPSS 16 been previously described. The procedure begins by applying Statistics 24 (IBM, Armonk, New York). A P value Ͻ .05 was set the Leksell stereotactic skull frame (Elekta) with the patient under for statistical significance. conscious intravenous sedation and local anesthesia. After attach- ment of the fiducial system, patients undergo a high-resolution RESULTS 3D spoiled gradient-recalled-acquisition axial MR imaging (slice Cochlear Dose Parameters thickness ϭ 1 mm, TE/TR ϭ 4/16.7 ms, flip angle ϭ 20°, FOV ϭ The median cochlear volume identified by CT was 36.8 mm3 250 mm/100%, matrix ϭ 256 ϫ 256) after administration of IV (range, 4.90–77 mm3). The median cochlear volume as identified contrast. Both 1-mm axial fast spin-echo T2-weighted MR im- on T2-weighted MR imaging was 41 mm3 (range, 4.90–70 mm3). ages through the internal auditory canal and 3-mm T2 axial On the basis of CCC analysis, we found a poor correlation in whole-head images are acquired to evaluate tumor extent and cochlear volume between CT and MR imaging (CCC ϭ 0.16, r2 ϭ inner ear structures. Additionally, to better define the bony 0.17, P ϭ .245) (Fig 1). canal of the internal acoustic meatus, stereotactic non-con- The median of the average radiation dose delivered to the co- trast-enhanced spiral CT is performed (120 kV, 185 mA, full chlea was 3.50 Gy (range, 1.20–6.80 Gy) on CT and 3.40 Gy helical current, slice thickness ϭ 1.25 mm, display FOV ϭ 300 ϭ 2 ϭ Յ mm, matrix ϭ 512 ϫ 512). All imaging is obtained in stereo- (range, 1–6.70 Gy) on MR imaging (CCC 0.86, r 0.90, P tactic conditions using the appropriate fiducial box for either .001) (Fig 2). The median maximum dose delivered to the cochlea ϭ MR imaging or CT. was 6.70 Gy on CT and 6.60 Gy on MR imaging (CCC 0.89, 2 ϭ Յ Dose-planning was performed using the high-definition ste- r 0.90, P .001) (Fig 3). Based on CCC analysis, an almost reotactic MR imaging axial plane T2 volume 1-mm images and perfect correlation was observed between CT and MR imaging in was refined using the stereotactic 1.25-mm axial plane bone win- estimating the mean and maximum doses of radiation delivered dow CT images. Cochlear volume was obtained by windowing to the cochlea. and then tracing the volume of the cochlea on each axial plane The median cochlear volume receiving a 3-Gy radiation 3 3 image where it could be identified. Planning was by means of the dose as identified on CT was 25 mm (range, 0–47 mm ), and 3 3 standard Leksell Gamma Knife software. Volumetric stereotactic on MR imaging, it was 25 mm (range, 0–55.40 mm ). Dose- conformal planning of the tumor volume was completed by the volume histograms generated from CT and MR imaging dem- responsible neurosurgeon. A single radiation oncologist per- onstrated an almost perfect level of correlation in estimating formed tumor, cochlea, trigeminal nerve, and brain stem delinea- the cochlear volume receiving a 3-Gy radiation dose (CCC ϭ tion independently on the CT and MR images. The cochlear vol- 0.81, r2 ϭ 0.82, P Յ .001). umes were independently drawn as well by the responsible The median cochlear volume receiving a 4.2-Gy radiation dose neurosurgeon. Ultimately the CT and MR images were fused us- as identified on CT was 12.7 mm3 (range, 0–63 mm3), and on MR ing the Leksell GammaPlan software (Electa) to optimize the final imaging, it was 7.60 mm3 (range, 0–46 mm3). Based on CCC treatment plan. Selective sector beam-blocking (generally 1 sector analysis, an almost perfect correlation was observed between CT of 28 beams) was used to limit the radiation dose to the brain and MR imaging in estimating the cochlear volume receiving a stem, the trigeminal nerve, and the cochlea. 3D volumes of the 4.2-Gy radiation dose (CCC ϭ 0.87, r2 ϭ 0.89, P Յ .001) (Fig 4). 1908 Faramand Oct 2018 www.ajnr.org who received an average cochlear dose of Յ4.20 Gy to the center of the cochlea had better hearing preservation rates. Hearing preservation is an important consideration in patients undergoing management. Yang et al22 performed a literature review looking at the hearing- preservation rates in patients who un- derwent SRS for the management of VS. In their review, among 4234 patients with VS treated with SRS, there was a 51% chance of hearing preservation at pre–gamma knife levels at a mean fol- low-up of 3 years. Yomo et al23 retro- spectively reviewed the outcomes of 154 patients who underwent SRS for the management of vestibular schwanno- FIG 1. CCC plot comparing the cochlear volume as identified on CT and MR imaging. A CCC score of 0.16 indicates a poor correlation between the 2 imaging modalities in identifying cochlear mas. After a mean audiologic follow-up volume. of 52 months post-SRS, a maximum co- chlear dose of Ͻ4 Gy was found to be the sole prognostic factor for hearing preser- vation. Hasegawa et al24 reported hearing outcomes in 92 patients who underwent SRS for the management of vestibular schwannomas at a median follow-up of 83 months. They found that the mean co- chlear radiation dose and pre-SRS pure- tone average were the main predictors of hearing preservation. Both MR imaging and CT are com- monly used in treatment planning. However, it is widely accepted that MR imaging is superior when it comes to identifying soft-tissue structures, whereas CT provides superior resolution of bony structures.14,25 Using CCC analysis, we FIG 2. CCC plot comparing the mean radiation dose delivered to the cochlea as identified on CT found a poor correlation in cochlear vol- and MR imaging. A CCC score of 0.86 indicates an almost perfect relationship in identifying the ume identification between CT and MR mean cochlear dose between CT and MR imaging. imaging. This variance may also relate to the windowing level used to define the DISCUSSION cochlea using both CT and MR imaging. Kulkarni et al25 per- Prevention of further tumor growth and the preservation of cra- formed a comparison of gross target volumes as delineated inde- nial nerve function are the main goals of SRS management of VS. pendently on contrast-enhanced CT and T1- and T2-weighted During the past several years, multiple factors have been found to influence hearing-preservation rates. These include the age of the MR imaging in vestibular schwannomas. In their analysis, they patient, hearing status using scales such as the Gardner Robertson found that cochlear volume as identified on T2 images was signif- 3 classification, length of the time between diagnosis and treat- icantly larger (23.9 mm ) than the cochlear volume identified on 3 14 ment, difference in the pure-tone average between the tumor CT (15 mm ). Jacob et al compared CT- and MR imaging– and nontumor ear, and estimated length of the vestibuloco- based modiolus point dose measurements and found a moderate chlear nerve.6,9,17,18 The correlation between cochlear dose and level of correlation between CT and MR imaging in identifying hearing preservation has been studied extensively. Multiple re- the dose point of the cochlear modiolus. Treatment planning is ports have demonstrated that a higher radiation dose to the co- performed after the fusion of CT and MR images. This may ex- chlea is associated with a higher chance of hearing decline during plain how, despite the significant difference in cochlear volumes long-term follow-up.19-21 In a review of the outcomes of 53 pa- between the 2 modalities, the dose delivered to the cochlea was tients who underwent SRS for the management of vestibular found to be the same. schwannomas, Brown et al8 found that patient age and the per- The importance of accurately identifying the mean and max- centage of the cochlea that receives Ͼ5.30 Gy are the main pre- imum doses received by the cochlea is related to the fact that dictors of hearing preservation. Kano et al18 found that patients multiple reports observed that the radiation dose of 3–5.3 Gy was AJNR Am J Neuroradiol 39:1907–11 Oct 2018 www.ajnr.org 1909 and MR imaging in the same patient are not different indicates that we can elim- inate the need for additional radiation delivered by spiral CT in patients with VS undergoing Leksell gamma knife SRS.

Weaknesses of the Present Study The overall findings of this report may be limited by the small number of pa- tients included in the analysis. In this study, both the responsible surgeon and the radiation oncologist indepen- dently found that the cochlear volume drawn by each had no impact on the dose delivered to the cochlea. Future studies may allow a more robust anal- FIG 3. CCC plot comparing the maximum radiation dose delivered to the cochlea as identified on ysis of variance. Windowing of the co- CT and MR imaging. A CCC score of 0.89 indicates an almost perfect relationship in identifying the mean cochlear dose between CT and MR imaging. chlear volume by CT may influence the tracing of the cochlear volume in each axial slice. Variation in the slice thickness of CT (1.25 mm) and T2 vol- ume MR imaging (1 mm) may also af- fect these results.

CONCLUSIONS In this study, both CT and MR imag- ing provided similar cochlear dose pa- rameters during Leksell radiosurgery for VS. Despite the differences in co- chlear volume identified by either CT or MR imaging, the dose delivered to the cochlea in this study was not dif- ferent. In patients undergoing Leksell gamma knife SRS for VS, high-defini- tion MR imaging alone provides supe- FIG 4. CCC plot comparing the volume of the cochlea receiving 4.2-Gy volume as identified on rior 3D tumor volume definition and CT and MR imaging. A CCC score of 0.87 indicates an almost perfect correlation between CT and a satisfactory depiction of cochlear MR imaging in identifying the cochlear volume receiving 4.2 Gy. volume.

7,8,14,20,26 associated with better preservation of hearing status. We Disclosures: Hideyuki Kano—UNRELATED: Comments: supported by Elekta AB. L. found that both imaging modalities provide almost perfect values Dade Lunsford—UNRELATED: Consultancy: Insightec, Data and Safety Monitor- when determining the mean dose, maximum dose, and 3- and ing Board; Stock/Stock Options: Elekta AB. 4.20-Gy volumes. The rationale for using multiple imaging modalities during radiosurgery is to enable the accurate delivery of a high radiation REFERENCES dose to the identified target, while, at the same time, sparing doses 1. Delbrouck C, Hassid S, Massager N, et al. Preservation of hear- delivered to adjacent critical structures. Despite the previously ing in vestibular schwannomas treated by radiosurgery using Leksell gamma knife: preliminary report of a prospective Bel- reported advantage of CT in more accurately identifying the co- gian clinical study. Acta Otorhinolaryngol Belg 2003;57:197–204 chlea, we found that MR imaging provides similar dose-parame- Medline ter estimates compared with CT. 2. Lunsford LD, Niranjan A, Flickinger JC, et al. Radiosurgery of ves- During the 31 years that we have performed SRS using each tibular schwannomas: summary of experience in 829 cases. J Neu- of 6 models of the Leksell gamma knife, we have studied out- rosurg 2013;119(Suppl):195–99 Medline comes in 1956 patients. While age, hearing status, the interval 3. Williams BJ, Xu Z, Salvetti DJ, et al. Gamma knife surgery for large vestibular schwannomas: a single-center retrospective case-matched between diagnosis and treatment, and tumor volume cannot be comparison assessing the effect of lesion size. J Neurosurg 2013;119: controlled by the surgeon, tumor margin and maximum doses 463–71 CrossRef Medline and the dose delivered to the cochlea can be controlled. The find- 4. Bailo M, Boari N, Franzin A, et al. Gamma knife radiosurgery as ing that doses delivered to the cochlea as estimated by both CT primary treatment for large vestibular schwannomas: clinical re- 1910 Faramand Oct 2018 www.ajnr.org sults at long-term follow-up in a series of 59 patients. World Neuro- treatment planning: implications for acoustic neuroma outcomes. surg 2016;95:487–501 CrossRef Medline Otol Neurotol 2005;26:1220–28 CrossRef Medline 5. Re´gis J, Delsanti C, Roche PH, et al. Functional outcomes of radio- 16. Flickinger JC, Kondziolka D, Niranjan A, et al. Acoustic neuroma surgical treatment of vestibular schwannomas: 1000 successive radiosurgery with marginal tumor doses of 12 to 13 Gy. Int J Radiat cases and review of the literature [in French]. Neurochirurgie 2004; Oncol Biol Phys 2004;60:225–30 CrossRef Medline 50(2–3 Pt 2):301–11 Medline 17. Linskey ME, Flickinger JC, Lunsford LD. Cranial nerve length pre- 6. Mousavi SH, Kano H, Faraji AH, et al. Hearing preservation up to 3 dicts the risk of delayed facial and trigeminal neuropathies after years after gamma knife radiosurgery for Gardner-Robertson class acoustic tumor stereotactic radiosurgery. Int J Radiat Oncol Biol I patients with vestibular schwannomas. Neurosurgery 2015;76: Phys 1993;25:227–33 CrossRef Medline 584–90; discussion 590–91 CrossRef Medline 18. Kano H, Kondziolka D, Khan A, et al. Predictors of hearing preser- 7. Kano H, Kondziolka D, Khan A, et al. Predictors of hearing preser- vation after stereotactic radiosurgery for acoustic neuroma: clinical vation after stereotactic radiosurgery for acoustic neuroma. J Neu- article. J Neurosurg 2013;119(Suppl):863–73 Medline rosurg 2009;111:863–73 CrossRef Medline 19. Link MJ, Pollock BE. Chasing the Holy Grail of vestibular schwan- 8. Brown M, Ruckenstein M, Bigelow D, et al. Predictors of hearing loss noma management. World Neurosurg 2013;80:276–78 CrossRef after gamma knife radiosurgery for vestibular schwannomas: age, Medline cochlear dose, and tumor coverage. Neurosurgery 2011;69:605–13; 20. Massager N, Nissim O, Delbrouck C, et al. Irradiation of cochlear discussion 613–14 CrossRef Medline structures during vestibular schwannoma radiosurgery and associ- 9. Akpinar B, Mousavi SH, McDowell MM, et al. Early radiosurgery ated hearing outcome. J Neurosurg 2013;119(Suppl):733–39 CrossRef improves hearing preservation in vestibular schwannoma patients Medline with normal hearing at the time of diagnosis. Int J Radiat Oncol Biol 21. Timmer FC, Hanssens PE, van Haren AE, et al. Gamma knife radio- Phys 2016;95:729–34 CrossRef Medline surgery for vestibular schwannomas: results of hearing preserva- 10. Lobato-Polo J, Kondziolka D, Zorro O, et al. Gamma knife radiosur- tion in relation to the cochlear radiation dose. Laryngoscope 2009; gery in younger patients with vestibular schwannomas. Neurosur- 119:1076–81 CrossRef Medline gery 2009;65:294–300; discussion 300–01 CrossRef Medline 22. Yang I, Sughrue ME, Han SJ, et al. A comprehensive analysis of 11. Cohen DS, Lustgarten JH, Miller E, et al. Effects of coregistration of hearing preservation after radiosurgery for vestibular schwan- MR to CT images on MR stereotactic accuracy. J Neurosurg 1995;82: noma. J Neurosurg 2010;112:851–59 CrossRef Medline 772–79 CrossRef Medline 23. Yomo S, Carron R, Thomassin JM, et al. Longitudinal analysis of 12. Yomo S, Tamura M, Carron R, et al. A quantitative comparison of hearing before and after radiosurgery for vestibular schwannoma. radiosurgical treatment parameters in vestibular schwannomas: J Neurosurg 2012;117:877–85 CrossRef Medline the Leksell Gamma Knife Perfexion versus Model 4C. Acta Neuro- 24. Hasegawa T, Kato T, Yamamoto T, et al. Long-term hearing out- chir (Wien) 2010;152:47–55 CrossRef Medline comes after gamma knife surgery in patients with vestibular 13. Bowden G, Cavaleri J, Monaco E III, et al. Cystic vestibular schwan- schwannoma with hearing preservation: evaluation in 92 patients nomas respond best to radiosurgery. Neurosurgery 2017;81:490–97 with serial audiograms. J Neurooncol 2018;138:283–90 CrossRef CrossRef Medline Medline 14. Jacob JT, Carlson ML, Schiefer TK, et al. Significance of cochlear 25. Kulkarni BS, Bajwa H, Chandrashekhar M, et al. CT- and MRI-based dose in the radiosurgical treatment of vestibular schwannoma: gross target volume comparison in vestibular schwannomas. Rep controversies and unanswered questions. Neurosurgery 2014;474; Pract Oncol Radiother 2017;22:201–08 CrossRef Medline 74:466–74; discussion 474 CrossRef Medline 26. Baschnagel AM, Chen PY, Bojrab D, et al. Hearing preservation in 15. Poetker DM, Jursinic PA, Runge-Samuelson CL, et al. Distortion of patients with vestibular schwannoma treated with gamma knife magnetic resonance images used in gamma knife radiosurgery surgery. J Neurosurg 2013;118:571–78 CrossRef Medline

AJNR Am J Neuroradiol 39:1907–11 Oct 2018 www.ajnr.org 1911 ORIGINAL RESEARCH PEDIATRICS

Multidelay Arterial Spin-Labeling MRI in Neonates and Infants: Cerebral Perfusion Changes during Brain Maturation

X H.G. Kim, X J.H. Lee, X J.W. Choi, X M. Han, X S.-M. Gho, and X Y. Moon

ABSTRACT BACKGROUND AND PURPOSE: Arterial spin-labeling with multiple postlabeling delays can correct transit times. We tried to evaluate CBF in neonates and infants using multidelay arterial spin-labeling.

MATERIALS AND METHODS: Multidelay arterial spin-labeling was applied to 13 preterm neonates (mean postmenstrual age, 34.9 weeks), 13 term-equivalent-age neonates (mean postmenstrual age, 39.2 weeks), and 6 infants (mean postmenstrual age, 57.8 weeks). Transit time–corrected CBF in the caudate, thalamus, frontal GM, occipital GM, frontal WM, and occipital WM was measured, and relative CBF compared with the whole-brain CBF was calculated. Inter- and intragroup comparisons were performed among the 3 age groups. A correlation and nonlinear regression analysis were performed between postmenstrual age and CBF.

RESULTS: Intergroup comparisons showed significantly higher whole-brain CBF in infants (38.3 mL/100 g/min) compared with preterm (15.5 mL/100 g/min) and term-equivalent-age (18.3 mL/100 g/min) neonates (P Ͻ .001). In the intragroup comparison, all 3 groups showed significantly higher relative CBF values in the occipital WM (63.6%–90.3%) compared with the frontal WM (46.3%–73.9%). In term-equiva- lent-age neonates, the occipital GM (120.8%) had significantly higher relative CBF values than the frontal GM (103.5%). There was a significant negative correlation between postmenstrual age and the relative CBF of the thalamus (r ϭϪ0.449, P ϭ .010). There were significant positive relationships between postmenstrual age and the relative CBF of the frontal WM (R2 ϭ 0.298, P ϭ .001) and occipital WM (R2 ϭ 0.452, P Ͻ .001).

CONCLUSIONS: Multidelay arterial spin-labeling with transit time–corrected CBF showed developmental changes and regional differ- ences of CBF in neonates and infants.

ABBREVIATIONS: ASL ϭ arterial spin-labeling; F-GM ϭ frontal GM; F-WM ϭ frontal WM; O-GM ϭ occipital GM; O-WM ϭ occipital WM; PLD ϭ postlabeling delay; PMA ϭ postmenstrual age; rCBF ϭ relative CBF; T1b ϭ longitudinal relaxation time of blood; TEA ϭ term-equivalent-age; wbCBF ϭ whole-brain CBF

rain maturation is one of the most vital processes occurring increasing number of neonates undergo routine brain MR imag- Bduring neonatal life, and imaging studies can potentially pro- ing, advanced tools have been applied to observe maturation vide insight into normal brain development. MR imaging can processes in the neonatal brain. These include DWI,1,2 DTI,3 provide high-resolution structural and functional images. As an magnetization transfer imaging,4 and arterial spin-labeling (ASL).5

Received February 19, 2018; accepted after revision July 8. Among these methods, ASL is a noninvasive method that From the Departments of Radiology (H.G.K., J.W.C., M.H.) and Pediatrics (J.H.L.) and measures CBF. ASL uses endogenous blood water as a diffus- Office of Biostatistics (Y.M.), Institute of Medical Sciences, Ajou University School ible tracer5 and is increasingly applied in neurologic studies. By of Medicine, Ajou University Medical Center, Suwon, Korea; and MR Clinical Research and Development (S.-M.G.), GE Healthcare Korea, Seoul, Korea. using a radiofrequency inversion pulse, arterial blood protons 5 Hyun Gi Kim and Jang Hoon Lee contributed equally to the study and share first are magnetically labeled in the carotid artery level. ASL pro- authorship. vides a quantitative measure of regional brain function and can This work was funded by National Research Foundation of Korea 6 (NRF-2017R1D1A1B03034768). show changes in baseline function associated with aging. In Paper previously presented at: Annual Meeting of the Radiological Society of neonates, ASL has been applied to see changes in perfusion North America, November 26 to December 1, 2017; Chicago, Illinois. during brain maturation7 and to see the predictive value of Please address correspondence to Hyun Gi Kim, MD, PhD, Department of Radiology, these changes in patients who undergo hypothermia treatment Ajou University School of Medicine, Ajou University Medical Center, 164 World cup-ro, 8 Yeongtong-gu, Suwon, Korea, 16499; e-mail: [email protected] after hypoxic-ischemic injury. Indicates open access to non-subscribers at www.ajnr.org However, there are some limitations to applying ASL in neo- http://dx.doi.org/10.3174/ajnr.A5774 nates. First, neonates have different blood and brain tissue T1

1912 Kim Oct 2018 www.ajnr.org values than older children and adults.7,9 Adjusting these values to infants, 2 underwent MR imaging for suspicion of delayed devel- neonates is essential because inadequate application of ASL will opment; 2, for follow-up of meningitis; 1, for early closure of the lead to inaccurate calculation of CBF.10 Second, neonates have anterior fontanelle; and 1, for transient apnea. None of the in- decreased cerebral blood perfusion and low flow velocity.11 In cluded subjects showed abnormalities on conventional MR imag- ASL, labeled blood signals are acquired at the cerebral level after ing. All patients were sedated for the MR imaging examination enough time has passed to allow the labeled spins to reach the with oral chloral hydrate (0.5 mL/kg). An additional single dose of imaging section, and this time lapse is called the postlabeling delay IV midazolam (0.1 mg/kg) was administered to patients who (PLD). If the PLD is not long enough to reflect the brain perfusion awoke during scanning. of neonates with low flow velocity, the acquired perfusion values will be inaccurately low. Inaccurately low perfusion values will Image Acquisition lead to inadequate interpretation of CBF in neonates combined Multidelay ASL was included in our routine imaging protocol for with the inherent problem of a low signal-to-noise ratio in ASL. infants who underwent brain MR imaging with a 3T scanner ASL with multiple PLD acquisitions (multidelay ASL) has (750w; GE Healthcare). The routine brain MR imaging protocol been applied in adults. CBF values calculated from multidelay for neonates and infants included 3D T1-weighted spoiled gradi- ASL showed good correlation with the CBF results of PET,12 dy- ent-echo imaging, axial T2WI, DWI, gradient-echo imaging, and namic susceptibility contrast perfusion MR imaging,13 and CT multidelay ASL. perfusion.14 Compared with single-PLD ASL, multidelay ASL We used a multidelay ASL sequence based on Hadamard en- showed improved CBF quantification in patients with Moyamoya coding to obtain 7 perfusion-weighted images with different disease, who have a delay between labeling in the feeding arteries PLDs and effective labeling durations. For neonates and infants, and the arrival of the labeled blood in tissue. PLDs were the following: 2.00, 2.22, 2.48, 2.8, 3.15, 3.63, 4.32 The major trade-off of multidelay ASL used to be longer scan- seconds and 1.50, 1.72, 1.98, 2.28, 2.65, 3.13, 3.82 seconds, respec- 16 ning time. However, recent application of Hadamard encoding tively. Effective labeling durations were the following: 0.22, has made it possible to acquire multiple PLDs in a much shorter 0.26, 0.30, 0.37, 0.48, 0.68, 1.18 seconds, regardless of age. By time,15 and a more upgraded version of the multidelay ASL tech- means of Hadamard encoding, CBF images from 7 different PLDs nique (enhanced ASL technique; GE Healthcare, Milwaukee, could be generated from 8 acquisitions, and each CBF image was 15 Wisconsin) has been introduced. To the best of our knowledge, an average of 4 images. there has been no study that applies multidelay ASL to neonates. The arterial transit time was estimated using the weighted- 17 Therefore, the purpose of this study was to show cerebral perfu- delay method of Dai et al. On the arterial transit time map, the sion in neonates and infants using multidelay ASL with optimized median arterial transit times of preterm neonates, TEA neonates, T1 values. and infants were 2.27 seconds (interquartile range, 0.11 seconds), 2.26 seconds (interquartile range, 0.12 seconds), and 1.24 seconds (interquartile range, 0.53 seconds), respectively. Transit time- MATERIALS AND METHODS corrected CBF maps were computed from the cumulative perfu- Patients sion-weighted image.17-20 Other parameters for multidelay ASL This retrospective study was approved by the institutional review imaging were as follows: TR, 6791 ms; TE, 11 ms; FOV, 24 cm; 640 board of Ajou University Hospital. Brain MRIs obtained between sampling points on 5 spirals (matrix size, 640 ϫ 5); section thick- July 2016 and March 2017 were reviewed. In our institution, brain ness, 5 mm; number of sections, 20. MR imaging is performed in preterm neonates born before the 32nd week of gestation or in neonates with a birth weight of Image Analysis Ͻ1500 g. These neonates stay in the neonatal intensive care unit Acquired raw images of multidelay ASL were processed using during hospitalization and undergo brain MR imaging before dis- dedicated software to adjust for the longitudinal relaxation time charge. In addition to the above indications, infants undergo of blood (T1b) and tissue. This procedure was used because the brain MR imaging for clinical suspicion of brain injury. We ex- quantified perfusion of the acquired multidelay ASL was assumed cluded infants who had abnormalities on brain MR imaging, in- to have the T1b and longitudinal relaxation time of tissue values cluding hydrocephalus, hemorrhage, or hypoxic-ischemic injury. of adults, which are different from those of neonates. The longi- Thirty-two patients were included, and we divided the subjects tudinal relaxation time of tissue was assumed to be 1.5 seconds.21 into 3 groups according to age at the time of MR imaging: preterm Because neonates show a large individual variation for T1b, we neonates, term-equivalent-age (TEA) neonates, and infants. All individualized the T1b values for each neonate. The T1b was cal- those in the infant group were full-term with a postnatal age of 1 culated using the following equation9: month or older at the time of MR imaging. Therefore, there were Ϯ 13 preterm neonates (34.9 1.1 weeks’ postmenstrual age 1 / T1b ϭ a ϫ Hematocrit ϩ b, [PMA]), 13 TEA neonates (39.2 Ϯ 2.1 weeks’ PMA), and 6 infants (57.8 Ϯ 5.4 weeks’ PMA; 111.0 Ϯ 62.8 days’ postnatal age). where a ϭ 0.50 secondsϪ1, the specific relaxivity, and b ϭ 0.37 Several patients underwent brain MR imaging for indications secondsϪ1, the native relaxivity. Hematocrit values were acquired that were not routine. One preterm neonate underwent MR im- from blood tests performed within a week of MR imaging. The aging to rule out cranial complications of asphyxia. Among TEA mean values of hematocrit in preterm neonates, TEA neonates, neonates, 3 underwent MR imaging for apnea; 2, for macroceph- and infants were 31.1% Ϯ 8.0%, 39.0% Ϯ 8.3%, and 33.1% Ϯ aly; 2, for postresuscitation evaluation; and 1, for seizure. Among 5.1%, respectively. The mean values of T1b in preterm neonates, AJNR Am J Neuroradiol 39:1912–18 Oct 2018 www.ajnr.org 1913 TEA neonates, and infants were 1.91 Ϯ 0.14, 1.78 Ϯ 0.13, and cated statistical significance. SPSS, Version 25.0 (IBM, Armonk, 1.87 Ϯ 0.09, respectively. New York) and the R software package (Version 3.4.3; www. ROIs were manually drawn by 1 radiologist with 8 years of r-project.org) were used for analysis. experience in brain imaging on the PACS at the following brain regions for CBF in mL/100 g/min: the caudate, thalamus, fron- RESULTS tal GM (F-GM), occipital GM (O-GM), frontal WM (F-WM), Regional CBF and rCBF for each age group are summarized in and occipital WM (O-WM) (Fig 1). The size range of the ROIs Table 1. WM (7.3–35.6 mL/100 g/min) showed lower CBF than was 16.6–24.9 mm2. The whole-brain CBF (wbCBF) was ob- cortical GM (16.8–49.8 mL/100 g/min) in all age groups. In pre- tained, and we calculated relative CBF (rCBF) for each brain term and TEA neonates, deep GM structures of the caudate and region by comparing the specific CBF of the region with the thalamus showed the highest CBF (21.2–32.2 mL/100 g/min) with wbCBF. a rCBF of 136.4%–177.2% for all brain regions. In infants, the Statistical Analysis highest CBF was observed in the O-GM (49.8 mL/100 g/min) Continuous variables were first tested for normality using the followed by the caudate (42.9 mL/100 g/min) and F-GM (41.8 Kolmogorov-Smirnov test. For the intergroup comparison of mL/100 g/min). CBF for the 3 age groups, normally distributed independent vari- ables were compared using one-way analysis of variance followed Intergroup Comparison by the Tukey post hoc test. For the intragroup comparison of For absolute CBF values, there was a significant difference in CBF rCBF in the frontal and occipital lobes, the paired t test was used. in the 3 groups for all brain regions except the thalamus. Differ- PMA had a non-normal distribution. Correlations between PMA ences were significant between preterm neonates and infants and and CBF were analyzed using the Spearman correlation. For the between TEA neonates and infants, with a significantly higher CBF of WM, the best-fit nonlinear regression model (ie, the low- CBF being observed in infants for both comparisons. There was est Bayesian information criterion) was selected among qua- no significant difference between preterm neonates and TEA ne- dratic, logarithmic, and linear regression models. A P Ͻ .05 indi- onates. The wbCBF was significantly higher in infants (38.3 mL/ 100 g/min) compared with preterm (15.5 mL/100 g/min) and TEA (18.3 mL/100 g/min) neonates (P Ͻ .001). There was no significant difference between preterm and TEA neonates (P ϭ .388). For rCBF values, the thalamus, F-WM, and O-WM showed significant differences among the 3 groups. The rCBF of the thal- amus in preterm (174.1%) and TEA neonates (177.2%) was sig- nificantly higher than in infants (96.5%). The rCBF of the F-WM was significantly higher in infants (73.9%) than in preterm (46.7%) and TEA neonates (46.3%). The rCBF of the O-WM was significantly higher in infants (90.3%) than in preterm (63.6%) and TEA neonates (65.6%). Between preterm and TEA neonates, there was no significant difference in the rCBF of the thalamus, FIG 1. Transit time-corrected CBF is measured within the ROIs: frontal F-WM, and O-WM. The caudate, F-GM, and O-GM showed no GM (a); caudate (b); occipital GM (c); frontal WM (d); thalamus (e); and significant difference among the 3 groups. occipital WM (f).

Table 1: Regional CBF in the age groups of preterm neonates, term-equivalent-age neonates, and infantsa Preterm TEA Infants P Preterm Preterm TEA Brain Region (n = 13) (n = 13) (n =6) Value vs TEA vs Infants vs Infants CBF (mL/100 g/min) Caudate 21.2 (6.9) 25.5 (6.6) 42.9 (7.1) Ͻ.001 .260 Ͻ.001 Ͻ.001 Thalamus 27.7 (11.4) 32.2 (7.1) 36.2 (7.5) .168 .437 .166 .658 F-GM 16.8 (4.1) 18.8 (3.8) 41.8 (12.0) Ͻ.001 .688 Ͻ.001 Ͻ.001 O-GM 19.1 (6.2) 21.8 (3.6) 49.8 (19.8) Ͻ.001 .755 Ͻ.001 Ͻ.001 F-WM 7.3 (3.1) 8.4 (1.9) 28.3 (7.5) Ͻ.001 .744 Ͻ.001 Ͻ.001 O-WM 9.9 (3.5) 11.9 (3.3) 35.6 (14.1) Ͻ.001 .718 Ͻ.001 Ͻ.001 Whole-brain 15.5 (4.2) 18.3 (3.3) 38.3 (9.5) Ͻ.001 .388 Ͻ.001 Ͻ.001 rCBF (%) Caudate 136.4 (25.6) 140.8 (28.9) 116.0 (25.3) .184 .911 .291 .168 Thalamus 174.1 (33.1) 177.2 (28.1) 96.5 (17.6) Ͻ.001 .960 Ͻ.001 Ͻ.001 F-GM 111.0 (24.9) 103.5 (15.2) 108.5 (11.6) .614 .592 .965 .858 O-GM 122.8 (19.8) 120.8 (17.5) 127.6 (26.1) .791 .965 .879 .773 F-WM 46.7 (16.5) 46.3 (7.9) 73.9 (7.1) Ͻ.001 .995 Ͻ.001 Ͻ.001 O-WM 63.6 (15.4) 65.6 (10.4) 90.3 (13.7) .001 .980 .001 .001 a Data are mean (standard deviation).

1914 Kim Oct 2018 www.ajnr.org Table 2: Intragroup comparison between rCBF in the frontal and occipital lobesa Group F-GM O-GM P Value F-WM O-WM P Value Preterm 111.0 (24.9) 122.8 (19.8) .730 46.7 (16.5) 63.6 (15.4) .004 TEA 103.5 (15.2) 120.8 (17.5) .026 46.3 (7.9) 65.6 (10.4) Ͻ.001 Infants 108.5 (11.6) 127.6 (26.1) .190 73.9 (7.1) 90.3 (13.7) .046 a Data are mean (standard deviation) (%).

FIG 2. Intragroup comparison of rCBF in the frontal and occipital lobes. A single asterisk indicates P Ͻ .05; double asterisks, P Ͻ .001.

FIG 3. Relationship between postmenstrual age and CBF. A single asterisk indicates P Ͻ .05; double asterisks, P Ͻ .001.

Intragroup Comparison There was a positive relationship between the rCBF of the F-WM Results of the intragroup comparison are summarized in Table 2 and PMA ([rCBF of F-WM] ϭϪ107.48 ϩ 43.13 ϫ log[PMA]). and Fig 2. When the rCBF of the F-WM and O-WM was com- For the rCBF of the O-WM and PMA, the linear regression pared, all 3 age groups showed significantly higher rCBF values in model was the best fit with an R2 value of 0.452 (P Ͻ .001), and the O-WM compared with the F-WM. In infants, the O-WM there was a positive relationship between the 2 variables showed a rCBF value 1.2 times higher than the rCBF value of the ([rCBF of O-WM] ϭ 24.15 ϩ 1.096 ϫ [PMA]) (Fig 3). F-WM. In preterm and TEA neonates, the O-WM showed a rCBF value 1.4 times higher than the F-WM. Regarding GM, the O-GM DISCUSSION showed a significantly higher rCBF value compared with the This study showed that multidelay ASL could be applied to neo- F-GM in TEA neonates (120.8% versus 103.5%, P ϭ .026). In nates and infants by optimizing PLD and T1 values. By multidelay preterm neonates and infants, the O-GM (122.8%–127.6%) ASL, absolute cerebral perfusion values in neonates and infants showed a higher rCBF value than the F-GM (108.5%–111.0%), were estimated and a significantly higher wbCBF was observed in but there was no statistical significance. the older age groups. The rCBF of brain regions significantly dif- Relationship between Postmenstrual Age and CBF fered according to age group. Intragroup analysis of rCBF showed There was a significant positive correlation between PMA and higher perfusion in the occipital lobes compared with the frontal wbCBF (r ϭ 0.732, P Ͻ .001) (Fig 3). For PMA and the rCBF of the lobes. There was a negative correlation between PMA and the thalamus, there was a significant negative correlation between rCBF of the thalamus but a positive correlation between PMA and values (r ϭϪ0.449, P ϭ .010). There was a significant positive the rCBF of the WM. correlation between PMA and the rCBF of the F-WM (r ϭ 0.402, The benefits of translating neonatal perfusion imaging into the P ϭ .023) and O-WM (r ϭ 0.410, P ϭ .020). Correlation between clinical area are numerous because neonatal brain imaging can PMA and the rCBF of the caudate (r ϭϪ0.152, P ϭ .406), F-GM help predict neurodevelopmental outcomes.22-24 Because brain (r ϭϪ0.045, P ϭ .806), and O-GM (r ϭϪ0.157, P ϭ .392) MR imaging is now more commonly used in preterm neonates showed no statistical significance. Among nonlinear regression and in patients suspected of having brain injuries, identifying MR models, the logarithmic regression model was the best fit for the imaging parameters that correlate with neurodevelopmental out- rCBF of the F-WM and PMA with an R2 value of 0.298 (P ϭ .001). comes is becoming more important. AJNR Am J Neuroradiol 39:1912–18 Oct 2018 www.ajnr.org 1915 Because functional activity and brain maturation can be indi- ologic evolution of CBF within 1 year of birth. Deeper insight into rectly measured by the cerebral metabolic rate, numerous at- this field could be possible with future studies using noninvasive tempts have been made to measure CBF in neonates. In the past, ASL. PET25-28and xenon-enhanced CT29 were used to estimate CBF in Deep GM showed the highest rCBF values among the brain neonates. With these methods, a relatively low wbCBF was iden- regions, and there was a negative trend of values with age, which is tified in neonates compared with adults. However, these past in line with findings in prior studies.7,27,29,33 Prior studies on studies had inherent limitations due to radiation use and the in- brain metabolism in neonates showed that resting metabolism is vasive properties of their methods. The subsequent development not identical across the brain.7 Metabolism is generally low in the and application of ASL in neonates showed results that were in cortex at term but increases in the parietal, temporal, and occipital agreement with former PET results.30 Still, an accurate estimation lobes by 3 months. The last region to increase its metabolism is the of CBF is challenging with ASL in neonates due to their low blood frontal cortex.37 A previous neonatal ASL study showed higher content and restricted water extraction in brain tissue along with occipital cortical rCBF compared with frontal cortical rCBF in physiologic low CBF, which all contribute to a low signal-to-noise both preterm (68% versus 56%) and TEA neonates (92% versus ratio.11 74%), though the study did not evaluate statistical significance.7 The absolute wbCBF found in our study (15.5–18.3 mL/ Still, 1 study using PET showed comparable occipital and frontal 100 g/min) was higher than that in a previous ASL study of neo- cortical CBF values in preterm (5.8 versus 6.5 ␮mol/100 g/min) nates (7–12 mL/100 g/min).7 We speculate that this difference is and term neonates (8.7 versus 7.6 ␮mol/100 g/min).27 Except for due to arterial transit-time correction with multidelay ASL. ASL in TEA neonates, the rCBF of the cortical GM in our study did not incorporates a similar fundamental mechanism with PET, but it show significant difference between the frontal and occipital uses protons in the blood as tracers and is easily affected by arterial lobes. These discrepancies could be attributed to the small sample transit time. For example, the CBF of WM calculated from ASL size of our study. was underestimated due to a longer arterial transit time compared WM showed lower CBF values compared with GM, which is with GM.31 In a similar manner, in occlusive diseases such as consistent with previous studies on neonates.33 In all age groups, Moyamoya disease, arterial transit time is elongated in the af- the O-WM showed higher rCBF values compared with the fected hemispheres and results in underestimated CBF.13 With F-WM. One possible explanation for this finding is the higher multiple PLDs, we could generate an arterial transit time map that development of the O-WM compared with the F-WM. A number was corrected for CBF values. It was possible to overcome delayed of studies have performed regional WM evaluation using MR im- transit time in the adult population using this transit time–cor- aging, and the most widely accepted tools are DWI1,2 and DTI.3 rected CBF map made with multidelay ASL.32 While we did ac- One DWI study evaluating the development of WM in different quire arterial transit-time maps and showed median (interquar- brain regions showed lower ADC values in the occipital peritrigo- tile range) values according to the age groups, analyzing regional nal WM compared with the F-WM.1 In addition, a DTI study and age-dependent arterial transit times was beyond the scope of showed lower isotropic diffusion values in the O-WM (1.46 ϫ this study. Still, investigating arterial transit times obtained by 10Ϫ3mm2/s) compared with the F-WM (1.56 ϫ 10Ϫ3mm2/s).3 multidelay ASL and comparing single-delay and multidelay ASL Because lower ADC and lower isotropic diffusion values suggest in the neonatal population would be an interesting topic for fu- higher WM development, the higher rCBF in the O-WM in our ture studies. study is in accordance with these studies. Still, we think that this A prior neonatal ASL study measured a wbCBF of 16–21 mL/ result needs to be interpreted with caution. A meta-analysis of 100 g/min,33 which is relatively higher than the wbCBF of our DWI studies performed on neonate brains showed similar ADC study. However, the study did not optimize the T1 value for neo- values in the O-WM (146.4–164.2 ϫ 10Ϫ5mm2/s) and F-WM nates, which could result in overestimated CBF values. Hemato- (147.9–161.9 ϫ 10Ϫ5mm2/s), both of which were measured in the crit levels vary in neonates according to gestational age and con- cortical WM.2 In that study, a relatively lower ADC value was tinue to change,34 with the T1b being influenced by the observed in the subcortical WM (105.4–149.5 ϫ 10Ϫ5mm2/s). hematocrit levels.9 Without augmentation of the T1 value for ne- Because we manually drew ROIs and the spatial resolution of ASL onates, ASL could result in overestimated CBF.10,35 is generally low, including the subcortical WM of the occipital The higher wbCBF values observed in the older age groups and lobe might have resulted in significantly higher CBF compared the positive correlation between PMA and wbCBF found in our with F-WM. study were consistent with those in previous studies.7,27,36 Unlike There are several limitations to this study. First, there were a in the older age groups, which show a reversed pattern of decreas- limited number of subjects in each age group. The insignificant ing CBF according to age,6 infants show a rapid increase in CBF results found for some of the parameters might have been due to according to age. By means of PET, infants with a PMA of 32–60 the small number of included subjects. In addition, the significant weeks were reported to have CBF from 5.5 to 18.7 ␮mol/100 correlation between PMA and rCBF values could have been highly g/min.27 This prior study showed a significant positive correlation influenced by the infant group. Future studies with a larger num- between wbCBF and PMA as was seen in our results.27 In another ber of subjects would offer additional information on the age- study using phase-contrast MR imaging, a CBF of 18–30 mL/100 related changes of CBF in this young population. Second, the mL/min at birth rapidly increased with age and reached 60 mL/ study population cannot fully represent healthy neonates and in- 100 mL/min in 1 year.36 Compared with the number of studies in fants because there were subjects with clinical suspicion of brain older populations, there have been very few studies on the physi- injury. Still, we tried to minimize the effect of abnormal CBF by 1916 Kim Oct 2018 www.ajnr.org excluding subjects with visible abnormalities. In addition, TEA arterial spin labeling MRI in neonates. Eur J Radiol 2013;82:538–43 neonates cannot represent full-term neonates. We know that CrossRef Medline PMA rather than postnatal age is the chief factor in glucose 8. Massaro AN, Bouyssi-Kobar M, Chang T, et al. Brain perfusion in 14,27 encephalopathic newborns after therapeutic hypothermia. AJNR use. However, preterm birth might disrupt the maturational Am J Neuroradiol 2013;34:1649–55 CrossRef Medline program in a regionally specific manner, so TEA neonates cannot 9. Varela M, Hajnal JV, Petersen ET, et al. A method for rapid in vivo completely represent healthy full-term neonates.37 Third, the ASL measurement of blood T1. NMR Biomed 2011;24:80–88 CrossRef signal could have been affected by head positioning in the coil. Medline The infant brain has a larger variability of labeling efficiency due 10. Varela M, Petersen ET, Golay X, et al. Cerebral blood flow measure- ments in infants using look-locker arterial spin labeling. J Magn to variability in positioning within the standard head coil.38 We Reson Imaging 2015;41:1591–600 CrossRef Medline tried to position heads as centrally as possible in the coil during 11. Proisy M, Bruneau B, Rozel C, et al. Arterial spin labeling in clinical scanning. Still, the coil itself was not dedicated to neonates, and pediatric imaging. Diagn Interv Imaging 2016;97:151–58 CrossRef there was inevitable variability in head positioning. Fourth, seda- Medline tion during scanning may have affected the CBF. Although chloral 12. Bokkers RP, Bremmer JP, van Berckel BN, et al. Arterial spin labeling hydrate does not affect brain cortex activity,39 no thorough inves- perfusion MRI at multiple delay times: a correlative study with H(2)(15)O positron emission tomography in patients with symp- tigation has been performed on human CBF. On the other hand, tomatic carotid artery occlusion. J Cereb Blood Flow Metab 2010;30: 40 midazolam causes dose-related changes in CBF. Natural sleep 222–29 CrossRef Medline during scanning using the “feed and wrap” technique on neonates 13. Hara S, Tanaka Y, Ueda Y, et al. Noninvasive evaluation of CBF and would be a better way of assessing normal development using perfusion delay of Moyamoya disease using arterial spin-labeling CBF. Last, we used a single longitudinal relaxation value for brain MRI with multiple postlabeling delays: comparison with 15O-gas PET and DSC-MRI. AJNR Am J Neuroradiol 2017;38:696–702 tissue. The intrinsic longitudinal relaxation value, which is used CrossRef Medline for CBF estimation, differs by region, age, and subject. In neo- 14. Wang R, Yu S, Alger JR, et al. Multi-delay arterial spin labeling per- nates, the longitudinal relaxation value for WM is higher than for fusion MRI in moyamoya disease–comparison with CT perfusion GM, and both WM and GM values decrease with PMA.10 Re- imaging. Eur Radiol 2014;24:1135–44 CrossRef Medline gion-, age-, and subject-specific CBF estimations would be ideal; 15. Wells JA, Lythgoe MF, Gadian DG, et al. In vivo Hadamard encoded continuous arterial spin labeling (H-CASL). Magn Reson Med 2010; however, these are not yet applicable in this study as well as in 63:1111–18 CrossRef Medline most other clinical studies. 16. Alsop DC, Detre JA, Golay X, et al. Recommended implementation of arterial spin-labeled perfusion MRI for clinical applications: a CONCLUSIONS consensus of the ISMRM perfusion study group and the European Multidelay ASL results in transit time-corrected CBF maps, which Consortium for ASL in Dementia. Magn Reson Med 2015;73:102–16 CrossRef Medline can show the developmental changes and differences occurring in 17. Dai W, Robson PM, Shankaranarayanan A, et al. Reduced resolution neonates and infants. PMA was significantly correlated with re- transit delay prescan for quantitative continuous arterial spin gional changes in CBF. Because multidelay ASL could overcome labeling perfusion imaging. Magn Reson Med 2012;67:1252–65 transit delay in this young population, it could be a promising tool CrossRef Medline for imaging brain maturation processes. 18. Dai W, Shankaranarayanan A, Alsop DC. Volumetric measurement of perfusion and arterial transit delay using Hadamard encoded continuous arterial spin labeling. Magn Reson Med 2013;69: Disclosures: Hyun Gi Kim—RELATED: Grant: National Research Foundation of Korea 1014–22 CrossRef Medline grant was received. Detailed information is in the submitted work. 19. Dai W, Garcia D, de Bazelaire C, et al. Continuous flow-driven in- version for arterial spin labeling using pulsed radio frequency and gradient fields. Magn Reson Med 2008;60:1488–97 CrossRef Medline REFERENCES 20. Maleki N, Dai W, Alsop DC. Optimization of background suppres- 1. Bu¨ltmann E, Mußgnug HJ, Zapf A, et al. Changes in brain micro- sion for arterial spin labeling perfusion imaging. MAGMA 2012;25: structure during infancy and childhood using clinical feasible 127–33 CrossRef Medline ADC-maps. Childs Nerv Syst 2017;33:735–45 CrossRef Medline 21. Luh WM, Wong EC, Bandettini PA, et al. QUIPSS II with thin-slice 2. Coats JS, Freeberg A, Pajela EG, et al. Meta-analysis of apparent TI1 periodic saturation: a method for improving accuracy of quan- diffusion coefficients in the newborn brain. Pediatr Neurol 2009;41: titative perfusion imaging using pulsed arterial spin labeling. Magn 263–74 CrossRef Medline Reson Med 1999;41:1246–54 CrossRef Medline 3. Schneider JFL, Il’yasov KA, Hennig J, et al. Fast quantitative diffusion- 22. Guo T, Duerden EG, Adams E, et al. Quantitative assessment of tensor imaging of cerebral white matter from the neonatal period to white matter injury in preterm neonates: association with out- adolescence. Neuroradiology 2004;46:258–66 CrossRef Medline comes. Neurology 2017;88:614–22 CrossRef Medline 4. Nossin-Manor R, Chung AD, Whyte HE, et al. Deep gray matter 23. Bassi L, Chew A, Merchant N, et al. Diffusion tensor imaging in maturation in very preterm neonates: regional variations and pa- preterm infants with punctate white matter lesions. Pediatr Res thology-related age-dependent changes in magnetization transfer 2011;69:561–66 CrossRef Medline ratio. Radiology 2012;263:510–17 CrossRef Medline 24. Smyser CD, Snyder AZ, Shimony JS, et al. Effects of white matter 5. Proisy M, Mitra S, Uria-Avellana C, et al. Brain perfusion imaging in injury on resting state fMRI measures in prematurely born infants. neonates: an overview. AJNR Am J Neuroradiol 2016;37:1766–73 PLoS One 2013;8:e68098 CrossRef Medline CrossRef 25. Altman DI, Volpe JJ. Positron emission tomography in newborn 6. Biagi L, Abbruzzese A, Bianchi MC, et al. Age dependence of cerebral infants. Clin Perinatol 1991;18:549–62 CrossRef Medline perfusion assessed by magnetic resonance continuous arterial spin 26. Nama V, Kozlowski JK, Hamvas A. Measurement of protein flux labeling. J Magn Reson Imaging 2007;25:696–702 CrossRef Medline with positron emission tomography in neonates. Arch Dis Child Fe- 7. De Vis JB, Petersen ET, de Vries LS, et al. Regional changes in brain tal Neonatal Ed 1999;80:F26–29 CrossRef Medline perfusion during brain maturation measured non-invasively with 27. Kinnala A, Suhonen-Polvi H, Aa¨rimaa T, et al. Cerebral metabolic AJNR Am J Neuroradiol 39:1912–18 Oct 2018 www.ajnr.org 1917 rate for glucose during the first six months of life: an FDG positron 34. Jopling J, Henry E, Wiedmeier SE, et al. Reference ranges for hemat- emission tomography study. Arch Dis Child Fetal Neonatal Ed 1996; ocrit and blood hemoglobin concentration during the neonatal 74:F153–57 CrossRef Medline period: data from a multihospital health care system. Pediatrics 28. Chugani HT, Phelps ME, Mazziotta JC. Positron emission tomogra- 2009;123:e333–37 Medline phy study of human brain functional development. Ann Neurol 35. De Vis JB, Hendrikse J, Groenendaal F, et al. Impact of neonate 1987;22:487–97 CrossRef Medline haematocrit variability on the longitudinal relaxation time of 29. Chiron C, Raynaud C, Mazie`reB,etal.Changes in regional cerebral blood: implications for arterial spin labelling MRI. Neuroimage Clin blood flow during brain maturation in children and adolescents. 2014;4:517–25 CrossRef Medline J Nucl Med 1992;33:696–703 Medline 36. Varela M, Groves AM, Arichi T, et al. Mean cerebral blood flow 30. Andersen JB, Henning WS, Lindberg U, et al. Positron emission to- measurements using phase contrast MRI in the first year of life. mography/magnetic resonance hybrid scanner imaging of cerebral NMR Biomed 2012;25:1063–72 CrossRef Medline blood flow using (15)O-water positron emission tomography and 37. Peterson BS. Brain imaging studies of the anatomical and func- arterial spin labeling magnetic resonance imaging in newborn pig- tional consequences of preterm birth for human brain develop- lets. J Cereb Blood Flow Metab 2015;35:1703–10 CrossRef Medline ment. Ann N Y Acad Sci 2003;1008:219–37 CrossRef Medline 31. Ye FQ, Berman KF, Ellmore T, et al. H(2)(15)O PET validation of steady-state arterial spin tagging cerebral blood flow measurements in 38. Wang Z, Ferna´ndez-Seara M, Alsop DC, et al. Assessment of func- humans. Magn Reson Med 2000;44:450–56 CrossRef Medline tional development in normal infant brain using arterial spin 32. Yun TJ, Sohn CH, Yoo RE, et al. Transit time corrected arterial spin labeled perfusion MRI. Neuroimage 2008;39:973–78 CrossRef labeling technique aids to overcome delayed transit time effect. Medline Neuroradiology 2018;60:255–65 CrossRef Medline 39. Thoresen M, Henriksen O, Wannag E, et al. Does a sedative dose of 33. Miranda MJ, Olofsson K, Sidaros K. Noninvasive measurements of chloral hydrate modify the EEG of children with epilepsy? Electro- regional cerebral perfusion in preterm and term neonates by mag- encephalogr Clin Neurophysiol 1997;102:152–57 CrossRef Medline netic resonance arterial spin labeling. Pediatr Res 2006;60:359–63 40. Wolff J. Cerebrovascular and metabolic effects of midazolam and CrossRef Medline flumazenil. Acta Anaesthesiologica Scandinavica 1990;92:75–77 Medline

1918 Kim Oct 2018 www.ajnr.org ORIGINAL RESEARCH PEDIATRICS

The Impact of Persistent Leukoencephalopathy on Brain White Matter Microstructure in Long-Term Survivors of Acute Lymphoblastic Leukemia Treated with Chemotherapy Only

X N.D. Sabin, X Y.T. Cheung, X W.E. Reddick, X D. Bhojwani, X W. Liu, X J.O. Glass, X T.M. Brinkman, X S.N. Hwang, X D. Srivastava, X C.-H. Pui, X L.L. Robison, X M.M. Hudson, and X K.R. Krull

ABSTRACT BACKGROUND AND PURPOSE: Survivors of acute lymphoblastic leukemia are at risk for neurocognitive deficits and leukoencephalop- athy. We performed a longitudinal assessment of leukoencephalopathy and its associations with long-term brain microstructural white matter integrity and neurocognitive outcomes in survivors of childhood acute lymphoblastic leukemia treated on a modern chemother- apy-only protocol.

MATERIALS AND METHODS: One hundred seventy-three survivors of acute lymphoblastic leukemia (49% female), treated on a chemo- therapy-only protocol, underwent brain MR imaging during active therapy and repeat imaging and neurocognitive testing at follow-up (median, 13.5 years of age; interquartile range, 10.7–17.6 years; median time since diagnosis, 7.5 years; interquartile range, 6.3–9.1 years). Persistence of leukoencephalopathy was examined in relation to demographic and treatment data and to brain DTI in major fiber tracts and neurocognitive testing at follow-up.

RESULTS: Leukoencephalopathy was found in 52 of 173 long-term survivors (30.0%) and persisted in 41 of 52 (78.8%) who developed it during therapy. DTI parameters were associated with leukoencephalopathy in multiple brain regions, including the corona radiata (frac- tional anisotropy, P ϭ .001; mean diffusivity, P Ͻ .001), superior longitudinal fasciculi (fractional anisotropy, P ϭ .02; mean diffusivity, P Ͻ .001), and superior fronto-occipital fasciculi (fractional anisotropy, P ϭ .006; mean diffusivity, P Ͻ .001). Mean diffusivity was associated with neurocognitive impairment including in the genu of the corpus callosum (P ϭ .04), corona radiata (P ϭ .02), and superior fronto- occipital fasciculi (P ϭ .02).

CONCLUSIONS: Leukoencephalopathy during active therapy and neurocognitive impairment at long-term follow-up are associated with microstructural white matter integrity. DTI may be more sensitive than standard MR imaging for detection of clinically consequential white matter abnormalities in childhood acute lymphoblastic leukemia survivors treated with chemotherapy and in children undergoing treatment.

ABBREVIATION: ALL ϭ acute lymphoblastic leukemia

ontemporary treatment for pediatric acute lymphoblastic adverse effects such as neurocognitive deficits, metabolic and en- Cleukemia (ALL) has largely eliminated the use of prophylactic docrine disorders, and subsequent neoplasms.1,3,4 Parallel to this cranial radiation therapy1,2 and has substantially reduced many advance has been the use of risk-directed systemic and intrathecal chemotherapy, which has increased survival rates to Ͼ90% in some studies.1 Although the neurocognitive function of survivors

Received March 26, 2018; accepted after revision July 19. treated with chemotherapy only is better preserved than in those From the Departments of Diagnostic Imaging (N.D.S., W.E.R., J.O.G., S.N.H.), Epide- who underwent cranial radiation therapy, these survivors do miology and Cancer Control (Y.T.C., T.M.B., L.L.R., M.M.H., K.R.K.), Biostatistics (W.L., demonstrate deficiencies.2,5 D.S.), Psychology (T.M.B., K.R.K.), and Oncology (C.-H.P., M.M.H.), St. Jude Children’s Research Hospital, Memphis, Tennessee; and Children’s Center for Cancer and High doses of methotrexate, one of the primary chemothera- Blood Diseases (D.B.), Children’s Hospital Los Angeles, Los Angeles, California. peutic agents used for consolidation treatment of ALL, are asso- Noah D. Sabin and Yin Ting Cheung contributed equally to this work. ciated with leukoencephalopathy (ie, white matter hyperintensi- This work was supported by the National Institute of Mental Health Grant No. R01 ties on brain MR imaging).6,7 These changes may be transient or MH085849, National Cancer Institute Grant No. R01 CA090246, and National Can- cer Institute Cancer Center Support Grant No. P30 CA021765. Paper previously presented, in part, at: Annual Meeting of the American Society of Indicates open access to non-subscribers at www.ajnr.org Clinical Oncology, May 29 to June 2, 2015; Chicago, Illinois. Indicates article with supplemental on-line tables. Please address correspondence to Noah D. Sabin, MD, JD, Department of Diagnos- Indicates article with supplemental on-line photo. tic Imaging, St. Jude Children’s Research Hospital, 262 Danny Thomas Place, MS 220, Memphis, TN 38105-3678; e-mail; [email protected] http://dx.doi.org/10.3174/ajnr.A5791

AJNR Am J Neuroradiol 39:1919–25 Oct 2018 www.ajnr.org 1919 persist over multiple MR imaging examinations conducted dur- Treatment ing and after chemotherapy. Leukoencephalopathy is sometimes The details of protocol therapy have been previously reported.1,17 associated with clinical findings of neurotoxicity such as stroke- Briefly, central nervous system–directed therapy comprised like symptoms, seizures, or aphasia.6,8 13–18 triple intrathecal treatments with methotrexate, hydrocor- DTI is an MR imaging technique to evaluate microstructural tisone, and cytarabine and 4 doses of high-dose intravenous 2 white matter integrity based on the diffusion of water in the brain. methotrexate at an average of 2.5 g/m per dose for patients with Water diffusion in intact white matter is anisotropic (ie, diffusion low-risk ALL. Patients with standard- or high-risk ALL received is limited in direction, presumably due to cell membranes and 16–25 triple intrathecal treatments and 4 doses of high-dose 2 myelin sheaths in parallel axonal fibers associated with white mat- methotrexate at an average of 5.0 g/m per dose. No patient re- ter tracts).9 Fractional anisotropy is a measure of unidirectionality ceived prophylactic central nervous system radiation therapy, of water diffusion on a scale of 0–1, with 0 representing no limi- even in the presence of central nervous system leukemia at tation in direction and 1 indicating diffusion completely confined diagnosis. to 1 direction. Mean diffusivity represents the average amount of MR Imaging diffusion regardless of direction.10-12 Injury to a white matter MR imaging examinations for the follow-up study included non- tract is associated with lower fractional anisotropy and higher contrast 3D sagittal T1WI, axial T2WI, and axial proton-density mean diffusivity compared with healthy white matter.13,14 and axial T2 FLAIR images obtained on a 1.5T platform. DTI of In a previous study of long-term survivors of childhood the brain was successfully obtained on a 1.5T platform using a ALL treated with chemotherapy only, we reported associations be- double spin-echo EPI pulse sequence with 12 noncollinear, non- tween neurobehavioral outcomes and on-treatment leukoencepha- coplanar diffusion-gradient directions. Imaging sets were ac- lopathy and DTI parameters in the frontostriatal tract, a tract we a quired with a spatial resolution of 1.7 ϫ 1.7 ϫ 3.0 mm and 4 15 priori predicted to be associated with executive function problems. acquisitions to ensure the highest signal-to-noise ratio possible We also examined pharmacologic determinants of neurocognitive within a limited amount of time. Voxelwise tensor calculations performance, cortical thickness, and functional MR imaging in this were performed with the Diffusion II toolkit under SPM8 16 cohort. In the current study, we more broadly examined the asso- (https://www.fil.ion.ucl.ac.uk/spm/). DTI measures of fractional ciation between white matter integrity and neurocognitive outcomes anisotropy and mean diffusivity were extracted from voxels in survivors by evaluating persistent leukoencephalopathy and DTI within fiber tracts throughout the brain. parameters in all major white matter tracts in the brain. We hypoth- All active therapy and long-term follow-up MR imaging ex- esized that survivors who demonstrated persistent leukoencephalop- aminations were reviewed for leukoencephalopathy by a board- athy would demonstrate poor microstructural white matter integrity certified neuroradiologist with a Certificate of Added Qualifica- as measured by DTI parameters and that poor white matter integrity tion in neuroradiology and graded according to the radiographic would be associated with neurocognitive impairment. criteria of the Common Terminology Criteria for Adverse Events (Version 4.0; https://www.eortc.be/services/doc/ctc/CTCAE_ 4.03_2010-06-14_QuickReference_5x7.pdf). The neuroradiolo- MATERIALS AND METHODS gist was blinded to patient risk stratification, neurocognitive test Study Participants performance, and DTI results. Hyperintensity on the T2- Four hundred eight children with ALL were treated at St. Jude weighted and T2 FLAIR images in the supratentorial white matter Children’s Research Hospital on a single protocol (June 2000 to was considered leukoencephalopathy if located in the supraven- October 2010; Total Therapy XV, Total Therapy Study XV for tricular white matter and/or periventricular white matter and Newly Diagnosed Patients with Acute Lymphoblastic Leukemia; more prominent than expected for terminal zones of myelination, clinicaltrials.gov No. NCT00137111) without prophylactic cra- normal developmental areas of T2 hyperintensity, for each sub- nial radiation therapy. Of these, 369 underwent MR imaging ex- ject’s age.18 A second board-certified neuroradiologist with a Cer- aminations of the brain at 4 time points during therapy to assess tificate of Added Qualification in neuroradiology separately development, progression, and/or resolution of leukoencepha- graded 30 of the MR imaging examinations to estimate interrater lopathy.6 Survivors were recruited during a long-term follow-up reliability, yielding a ␬ statistic of 0.75 (95% CI, 0.64–0.85), indi- clinical evaluation when they were at least 8 years of age and Ն5 cating substantial agreement. years from diagnosis. Exclusion criteria included the following: relapse, diagnosis of a subsequent neoplasm, lack of proficiency in Neurocognitive Testing English, or the presence of an unrelated neurologic disorder Neurocognitive testing was performed by certified examiners un- associated with cognitive impairment. Among the 295 survi- der the supervision of a board-certified clinical neuropsycholo- vors who fulfilled the eligibility criteria, 189 (64%) partici- gist. Testing procedures followed standard clinical guidelines, pated in long-term outcome studies between June 2009 and with fixed test order and a schedule to reduce the effects of inter- October 2014. Follow-up MR imaging of the brain was success- ference and fatigue. Testing evaluated executive function (cogni- fully performed for 173 of the survivors (On-line Figure). This tive flexibility, cognitive fluency, working memory, organization, study was approved by the institutional review board, and and problem-solving abilities),19 intelligence (intelligence quo- written informed consent was obtained from the patients, their tient),20 processing speed,21,22 attention,23 memory,24 and fine- parents, or guardians, as appropriate. motor dexterity.25 Raw scores for these neurocognitive tests were 1920 Sabin Oct 2018 www.ajnr.org frequencies were reported for categoric (173 ؍ Table 1: Demographics and treatment characteristics (N No. (%) Mean (SD) Median (IQR) variables. The Mann-Whitney U test Demographics (continuous variables) and the ␹2 test Sex (categoric variables) were used to evalu- Male 89 (51) ate whether demographic and clinical Female 84 (49) Race/ethnicity characteristics were different between White 124 (72) survivors with and without persistent Asian 3 (2) leukoencephalopathy. Associations be- Black 21 (12) tween persistent leukoencephalopathy Hispanic 18 (10) and white matter integrity (fractional Others 7 (4) Current age (yr) 14.4 (4.6) 13.5 (10.7–17.6) anisotropy and mean diffusivity) were Patient’s highest education (yr) 7.7 (3.9) 7.0 (4.0–11.0) examined using general linear model- Maternal education (yr) 13.6 (2.5) 13.0 (12.0–16.0) ing, adjusting for current age. P values Paternal education (yr) 13.6 (3.1) 12.0 (12.0–16.0) were adjusted by controlling for the false Treatment characteristics discovery rate within the global white Age at diagnosis (yr) 6.7 (4.3) 5.3 (3.5–8.6) Time since diagnosis (yr) 7.7 (1.8) 7.5 (6.3–9.1) matter tracts. Only tracts that were sig- Treatment risk stratum nificantly associated with leukoenceph- Low 102 (59) alopathy were tested for associations Standard 71 (41) with methotrexate exposure using gen- a Chemotherapy doses eral linear modeling. Neurocognitive Oral dexamethasone (mg/m2) 1096.4 (303.2) 1099.9 (985.3–1246.1) IV high-dose cytarabine (g/m2) 8.5 (3.5) 8.0 (8.0–8.0) scores were transformed into age-ad- IV leucovorin (mg/m2) 343.5 (207.1) 300.0 (220.0–390.0) justed z scores (␮ ϭ 0, ␴ ϭ 1.0) using IV high-dose methotrexateb (g/m2) 15.4 (6.7) 14.2 (11.4–19.0) nationally representative norms. One- IT MHA (No. of counts) 14.4 (4.0) 13.0 (12.0–16.0) sample t tests were applied to compare Note:—IQR indicates interquartile range; IT MHA, intrathecal injection of methotrexate plus hydrocortisone plus the group average with the expected cytarabine. population value (␮ ϭ 0) for the specific a Except for IT MHA, all drug doses are presented as cumulative doses (g/m2 or mg/m2). b High-dose IV methotrexate was defined as a daily dose of Ͼ1 g/m2 of IV methotrexate. tests. Impairment on an individual neu- rocognitive test was defined as a score falling below the tenth percentile of the norm (z score Յ Ϫ1.286). Global neurocognitive impairment was defined as having Ն2 neurocognitive test scores that fall Ͼ1.5 SDs or 1 score that falls Ͼ2 SDs below the mean. Associations between neurocognitive scores and persistent leukoencephalopathy were evaluated using general linear modeling, adjusting for age at evaluation. Associa- tions between global neurocognitive impairment and white mat- ter integrity were assessed using general linear modeling, adjusted for age at evaluation. All analyses were conducted in SAS (SAS 9.4; SAS Institute, Cary, North Carolina). Statistical significance was defined as a P value Ͻ .05, and all statistical tests were 2-sided.

FIG 1. Examples of leukoencephalopathy grading. Left, mild hyper- intensity in the bilateral periatrial white matter in this survivor of RESULTS childhood ALL is compatible with grade 1 leukoencephalopathy Demographic and treatment-related data are provided in Table 1, according to the Common Terminology Criteria for Adverse and their associations with leukoencephalopathy are shown in On- Events (Version 4.0). Right, more extensive and confluent hyperin- tensity in the periventricular white matter that extends into the line Table 1. Race/ethnicity for each subject was obtained from the bilateral supraventricular frontoparietal white matter is consid- medical record as self-reported by the patients’ parents at the time of ered grade 2 leukoencephalopathy. treatment. Of the 41 survivors with persistent leukoencephalopathy transformed into age-adjusted z scores based on nationally repre- (leukoencephalopathy that was present on both active therapy and sentative normative data (␮ ϭ 0, ␴ ϭ 1.0). Applying our previous long-term follow-up MR imaging examinations), 37 had grade 1 and methods,16 we selected neurocognitive measures on which survi- 4 had grade 2 according to the Common Terminology Criteria for vors’ performance differed from normative samples. Survivors Adverse Events (Version 4.0) (Fig 1). Because of the small number of with Ն2 neurocognitive test results Ͼ1.5 SDs or 1 test result Ͼ2 participants with grade 2 abnormality, leukoencephalopathy was SDs below age-adjusted normative data were considered to have treated as a binary variable (present or absent). There were no signif- global neurocognitive impairment. icant differences in the cumulative dosages of known neurotoxic chemotherapeutic agents between patients with or without per- Statistical Analysis sistent leukoencephalopathy (On-line Table 1). For demographic and clinical data, means, SDs, medians, and After correcting for the false discovery rate, there were signif- interquartile ranges were calculated for continuous variables and icant associations between age-adjusted DTI parameters in mul- AJNR Am J Neuroradiol 39:1919–25 Oct 2018 www.ajnr.org 1921 FIG 2. Leukoencephalopathy and white matter integrity. MD indicates mean diffusivity. Higher MD is indicative of worse white matter integrity. Point estimates represent the differences in white matter integrity between survivors with persistent leukoencephalopathy and survivors without a history of leukoencephalopathy within the global tracts listed on the x-axis. Error bars represent the 95% confidence intervals. The P values compare the MD between survivors with and without acute leukoencephalopathy using general linear modeling, adjusted for current age. All models have been corrected for the false discovery rate within the global tracts. The image shows that survivors with persistent leukoencephalopathy had reduced white matter integrity, demonstrated by higher global MD, in the corpus callosum, corona radiata, posterior thalamic radiations, superior longitudinal fasciculi, and superior fronto-occipital fasciculi. Details on the associations between leukoencepha- lopathy and lateralized diffusion tensor imaging measures within the specific tracts can be found in On-line Table 2. tiple fiber tracts and the presence of persistent leukoencephalop- cephalopathy continuing to demonstrate leukoencephalopathy at athy (Fig 2 and On-line Table 2). Lower fractional anisotropy and long-term follow-up. Persistent leukoencephalopathy was associ- higher mean diffusivity were demonstrated in the corpus callo- ated with microstructural indices of poor white matter integrity sum, corona radiata, superior longitudinal fasciculi, and superior on DTI, though only the DTI parameters were associated with fronto-occipital fasciculi; higher mean diffusivity was detected in clinically relevant neurocognitive outcomes. Such findings sug- the posterior thalamic radiations in survivors with leukoenceph- gest that DTI may be a better method for evaluation of white alopathy. The total number of intrathecal administrations of matter pathology in patients and survivors of childhood ALL. methotrexate, hydrocortisone, and cytarabine was associated A study of 54 patients treated on 2 Pediatric Oncology Group with higher mean diffusivity values in several fiber tracts (Table ALL protocols reported a prevalence of leukoencephalopathy of 2). High-dose intravenous methotrexate was not associated with 22% for one protocol and 68% for the second at roughly 7 years DTI parameters. following diagnosis.26 The investigators attributed the higher Survivors had lower performance scores than population norms prevalence in the second treatment group to the greater use of on measures of executive function, memory, processing speed, and triple intrathecal therapy, increased frequency of intrathecal ther- global intelligence (almost all P Ͻ .001). There were no significant apy, and lack of leucovorin rescue compared with the first group. associations between neurocognitive performance measures and the The approximately 24% prevalence of leukoencephalopathy ob- presence of persistent leukoencephalopathy (On-line Table 3). How- served in the present investigation is similar to that reported in ever, higher mean diffusivity was associated with overall neurocog- survivors in the first group of the Pediatric Oncology Group nitive impairment (Fig 3 and On-line Table 4). study. We detected no significant associations between specific DISCUSSION chemotherapy cumulative doses and leukoencephalopathy. To our knowledge, this is the largest study of persistent leukoen- The association between poor DTI parameters in multiple supra- cephalopathy and tractography-based examination of white mat- tentorial white matter tracts and the presence of leukoencephalopa- ter integrity in long-term survivors of childhood ALL treated with thy in our survivors confirms the persistence of white matter injury chemotherapy only. With serial imaging of the brain conducted well into long-term follow-up. While our DTI results are similar to during therapy and Ͼ5 years following diagnosis in 173 survivors, those previously reported by others,13,27-33 our novel linkage of DTI we identified the prevalence of persistent leukoencephalopathy to to leukoencephalopathy suggests that white matter integrity is nega- be 23.7%, with 78.8% of survivors who developed acute leukoen- tively impacted during the early course of chemotherapy treatment. 1922 Sabin Oct 2018 www.ajnr.org Table 2: Association between methotrexate and white matter integritya Intrathecal MHAb High-Dose Methotrexateb Mean Diffusivity Fractional Anisotropy Mean Diffusivity Fractional Anisotropy Tracts Est. SE P Est. SE P Est. SE P Est. SE P Corpus callosum 0.0017 0.0008 .04c Ϫ0.0009 0.0007 .21 Ϫ0.0003 0.0005 .60 0.0006 0.0004 .17 Genu 0.0016 0.0009 .07 Ϫ0.0015 0.0012 .21 Ϫ0.0001 0.0005 .82 Ϫ0.0002 0.0007 .76 Body 0.0008 0.0010 .45 0.0000 0.0009 .99 Ϫ0.0008 0.0006 .21 0.0011 0.0005 .05 Splenium 0.0022 0.0011 .04c Ϫ0.0011 0.0007 .13 0.0000 0.0006 .95 0.0007 0.0004 .09 Corona radiata 0.0014 0.0007 .04c Ϫ0.0004 0.0005 .36 Ϫ0.0003 0.0004 .44 0.0002 0.0003 .57 Anterior (left) 0.0017 0.0009 .06 Ϫ0.0010 0.0005 .06 Ϫ0.0002 0.0005 .73 Ϫ0.0001 0.0003 .75 Anterior (right) 0.0011 0.0009 .19 Ϫ0.0005 0.0006 .41 Ϫ0.0005 0.0005 .31 0.0001 0.0003 .78 Superior (left) 0.0013 0.0005 .02c 0.0003 0.0007 .70 Ϫ0.0001 0.0003 .69 0.0004 0.0004 .33 Superior (right) 0.0015 0.0005 .005c Ϫ0.0003 0.0007 .66 Ϫ0.0002 0.0003 .58 0.0002 0.0004 .53 Posterior (left) 0.0009 0.0007 .23 0.0003 0.0007 .65 Ϫ0.0007 0.0004 .08 0.0007 0.0004 .12 Posterior (right) 0.0016 0.0007 .02c Ϫ0.0002 0.0007 .79 Ϫ0.0003 0.0004 .42 0.0002 0.0004 .58 Posterior thalamic radiation 0.0015 0.0007 .03c Ϫ0.0003 0.0007 .63 0.0000 0.0004 .94 0.0007 0.0004 .11 Left 0.0011 0.0010 .29 0.0000 0.0008 .98 Ϫ0.0003 0.0006 .65 0.0009 0.0005 .05 Right 0.0018 0.0006 .003c Ϫ0.0006 0.0007 .42 0.0001 0.0004 .69 0.0005 0.0004 .24 Superior longitudinal 0.0007 0.0005 .16 0.0006 0.0005 .25 Ϫ0.0001 0.0003 .75 0.0005 0.0003 .14 fasciculus Left 0.0006 0.0005 .22 0.0008 0.0006 .16 Ϫ0.0001 0.0003 .61 0.0004 0.0003 .30 Right 0.0007 0.0005 .13 0.0005 0.0006 .41 0.0000 0.0003 .87 0.0006 0.0003 .10 Superior fronto-occipital 0.0020 0.0009 .02c Ϫ0.0004 0.0009 .65 Ϫ0.0003 0.0005 .60 0.0009 0.0005 .09 fasciculus Left 0.0021 0.0010 .04c Ϫ0.0004 0.0009 .63 Ϫ0.0005 0.0006 .41 0.0009 0.0005 .10 Right 0.0019 0.0007 .01c Ϫ0.0003 0.0010 .73 0.0000 0.0005 .97 0.0008 0.0006 .14 Note:—Est. indicates parameter estimate; SE, standard error; MHA, methotrexate plus hydrocortisone plus cytarabine. a General linear modeling was applied for the test of strength of association between treatment variables with mean diffusivity and fractional anisotropy for each tract, adjusted for age at evaluation. Higher mean diffusivity and lower fractional anisotropy are indicative of worse white matter integrity. b Intrathecal MHA was defined as the number of intrathecal injections of methotrexate plus hydrocortisone plus cytarabine; high-dose IV methotrexate was defined as a daily dose of Ͼ1 g/m2, presented as cumulative doses (g/m2). c Significant.

et al26 reported a significant association between leukoencephalopathy and poorer performance on tests of attention in a group of 43 ALL survivors treated with chemotherapy only. Although there were no significant differences in the mean full-scale intelligence quo- tient, verbal intelligence quotient, or performance intelligence quotient be- tween patients with or without leukoen- cephalopathy, survivors with leukoen- cephalopathy composed 80% of those scoring Ͼ1 SD below the mean for ver- bal intelligence quotient; 64%, for per- formance intelligence quotient; and FIG 3. Association between white matter integrity and global neurocognitive impairment. Black 89%, for full-scale intelligence quo- circles represent survivors with global neurocognitive impairment. White circles represent survi- tient.26 In contrast, we did not see simi- vors without global neurocognitive impairment. Global neurocognitive impairment is defined as Ͼ having Ն2 neurocognitive tests (listed in On-line Table 3) that fall Ͼ1.5 SDs or 1 test that falls Ͼ2 lar trends in 150 survivors who com- SDs below the age-adjusted population normative data. The y-axis represents mean diffusivity. pleted both MR imaging examinations Higher mean diffusivity is indicative of worse white matter integrity. Error bars represent the 95% and neurocognitive testing. However, confidence intervals. P values compare the mean diffusivity between survivors with neurocogni- tive impairment (closed circle, n ϭ 63) and without neurocognitive impairment (open circle, n ϭ our DTI measures were associated with 87) using general linear modeling, adjusted for current age. The image shows that survivors with neurocognitive function. neurocognitive impairment had reduced white matter integrity demonstrated by higher global Mean diffusivity in the genu of the mean diffusivity in the corona radiata and superior fronto-occipital fasciculus. Details on the associations between global neurocognitive impairment and lateralized diffusion tensor imaging corpus callosum, corona radiata, and measures within the specific tracts can be found in On-line Table 4. the superior fronto-occipital fasciculi was associated with global neurocogni- Despite associations with lower fractional anisotropy and tive impairment. These tracts all involve the frontal lobes, which higher mean diffusivity, leukoencephalopathy was not signifi- play a role in executive function16 and which demonstrate pro- cantly associated with neurocognitive test performance. Duffner longed development during childhood, potentially placing them AJNR Am J Neuroradiol 39:1919–25 Oct 2018 www.ajnr.org 1923 at extended risk for neurotoxicity. Other investigators have also ing sequences. While the precise significance of the loss of struc- found associations between abnormal DTI parameters and defi- tural integrity that leukoencephalopathy represents is unclear, cits in neurocognitive performance in survivors of ALL treated continued refinement of chemotherapy treatment regimens for with chemotherapy only.13,27,28,32 That only mean diffusivity and ALL should nevertheless aim to reduce the occurrence of white not fractional anisotropy was associated with neurocognitive def- matter abnormalities. Potential treatments and interventions to icits suggests that the amount of diffusion, rather than the degree enhance white matter development and function also need to be of anisotropy, in our survivors’ white matter had a greater effect explored. on their neurocognitive abilities. The larger number of associa- tions between mean diffusivity in multiple fiber tracts and the CONCLUSIONS total number of intrathecal administrations of chemotherapeutic Leukoencephalopathy that develops in patients treated for child- agents, compared with fractional anisotropy, also indicates that hood ALL and long-term neurocognitive impairment are associ- chemotherapy has a greater effect on the amount of diffusion in ated with microstructural white matter integrity. DTI may be those tracts than the degree to which it is anisotropic. This is an more sensitive than standard MR imaging sequences for identify- area that requires further investigation. ing clinically consequential white matter abnormalities in this Limitations to our study include distinguishing leukoenceph- group of cancer survivors. alopathy from terminal zones of myelination, which can be diffi- cult in some cases. The grades of leukoencephalopathy were ACKNOWLEDGMENTS compared over multiple MR imaging examinations to try to stan- We would like to acknowledge Cara Kimberg, PhD, of Psycholog- dardize the visual thresholds used across the studies; however, it is ical Care Associates, Woburn, Massachusetts; and Ms Cynthia impossible, on the basis of visual inspection alone, to be certain Jones, Ms Deborah Stewart, and Ms Adrienne Studaway of the that some cases of leukoencephalopathy were not classified as Department of Epidemiology and Cancer Control, St. Jude Chil- terminal zones or that terminal zones may have been incorrectly dren’s Research Hospital, for administering the neurocognitive considered leukoencephalopathy. Given the size of our cohort, tests; and Ms Joycelynn Butler, of the Department of Epidemiol- though, the instances of misclassification likely did not have a ogy and Cancer Control, St. Jude Children’s Research Hospital, large effect on our results. In addition, the strong correlation be- for extracting and cleaning the data. tween the presence of leukoencephalopathy and DTI parameters suggests that misclassification did not significantly affect our data. Disclosures: Noah D. Sabin—RELATED: Grant: National Cancer Institute, National Another limitation is that this investigation lacked a control Institute of Mental Health.* Ching-Hon Pui—RELATED: Grant: National Institutes of Health CA 21765*. Leslie L. Robison—RELATED: Grant: National Institutes of Health*. group for comparison with the DTI findings. The differences in Melissa M. Hudson—RELATED: Grant: National Cancer Institute, Comments: U01 the DTI parameters between survivors with and without leukoen- 195547*. Kevin R. Krull—RELATED: Grant: National Institute of Mental Health, National Cancer Institute*; UNRELATED: Grants/Grants Pending: National Cancer Institute*; Roy- cephalopathy and the linkage of the DTI measures to standardized alties: Wolters Kluwer Publishers, Comments: royalties for chapters written on atten- neurocognitive testing, however, demonstrate the importance of tion-deficit/hyperactivity disorder for UpToDate. Chapters are unrelated to content of our DTI results. The limited research DTI sequence used in this current work. *Money paid to the institution. study was designed and implemented Ͼ8 years ago and was not optimal for tracking all possible connections. However, the con- REFERENCES nections reported were reproducible and reliable. Future studies 1. Pui CH, Campana D, Pei D, et al. Treating childhood acute lympho- will likely include higher spatial and angular resolution acquisi- blastic leukemia without cranial irradiation. N Engl J Med 2009;360: tions. We had no pretreatment DTI of the brain, which prevented 2730–41 CrossRef Medline us from determining whether DTI parameters were already ab- 2. Cheung YT, Krull KR. Neurocognitive outcomes in long-term sur- normal due to disease. Because treatment typically begins within vivors of childhood acute lymphoblastic leukemia treated on con- temporary treatment protocols: a systematic review. Neurosci 24 hours of diagnosis for ALL and because the imaging in this Biobehav Rev 2015;53:108–20 CrossRef Medline study was acquired solely for research, we were not able to obtain 3. Hudson MM, Ness KK, Gurney JG, et al. Clinical ascertainment of pretreatment imaging. Last, no neurocognitive testing was per- health outcomes among adults treated for childhood cancer. JAMA formed before active therapy; thus, some neurocognitive deficits 2013;309:2371–81 CrossRef Medline 4. Sabin ND, Santucci AK, Klimo P Jr, et al. Incidental detection of late in our subjects may have existed before treatment. The lack of a subsequent intracranial neoplasms with magnetic resonance imag- baseline study, however, is not likely to account for most of the ing among adult survivors of childhood cancer. J Cancer Surviv problems in neurocognitive performance we detected. 2014;8:329–35 CrossRef Medline Children successfully treated for ALL with chemotherapy 5. Krull KR, Brinkman TM, Li C, et al. Neurocognitive outcomes de- alone are at risk for persistent leukoencephalopathy. Although cades after treatment for childhood acute lymphoblastic leukemia: a report from the St Jude Lifetime Cohort study. J Clin Oncol 2013; leukoencephalopathy is an MR imaging biomarker of white mat- 31:4407–15 CrossRef Medline ter abnormality, it was not associated with neurocognitive deficits 6. Bhojwani D, Sabin ND, Pei D, et al. Methotrexate-induced neuro- in our survivors. Differences in DTI parameters, a more precise toxicity and leukoencephalopathy in childhood acute lymphoblas- measure of microstructural white matter integrity, however, were tic leukemia. J Clin Oncol 2014;32:949–59 CrossRef Medline associated with global neurocognitive impairment, suggesting 7. Reddick WE, Glass JO, Helton KJ, et al. Prevalence of leukoenceph- alopathy in children treated for acute lymphoblastic leukemia with that DTI may serve as a more sensitive imaging measure of clini- high-dose methotrexate. AJNR Am J Neuroradiol 2005;26:1263–69 cally consequential loss of white matter integrity in ALL survivors Medline than the presence of leukoencephalopathy on standard MR imag- 8. Fisher MJ, Khademian ZP, Simon EM, et al. Diffusion-weighted MR 1924 Sabin Oct 2018 www.ajnr.org imaging of early methotrexate-related neurotoxicity in children. 22. Wechsler D. Wechsler Adult Intelligence Scale. 4th ed. San Antonio: AJNR Am J Neuroradiol 2005;26:1686–89 Medline Pearson; 2008 9. Choudhri AF, Chin EM, Blitz AM, et al. Diffusion tensor imaging of 23. Conners CK. Conners’ Continuous Performance Test II. North cerebral white matter: technique, anatomy, and pathologic pat- Tonawanda: Multi-Health Systems; 2001 terns. Radiol Clin North Am 2014;52:413–25 CrossRef Medline 24. Meyers JE, Meyers KR. Rey Complex Figure and Recognition Trial. 10. Raffin E, Dyrby TB. Diagnostic approach to functional recovery: Odessa: Psychological Assessment Resources; 1996 diffusion-weighted imaging and tractography. Front Neurol Neuro- 25. Strauss E, Sherman EM, Spreen O. A Compendium of Neuropsycho- sci 2013;32:26–35 CrossRef Medline logical Tests: Administration, Norms, and Commentary. 3rd ed. New 11. Yang E, Nucifora PG, Melhem ER. Diffusion MR imaging: basic York: Oxford University Press; 2006 principles. Neuroimaging Clin N Am 2011;21:1–25, vii CrossRef 26. Duffner PK, Armstrong FD, Chen L, et al. Neurocognitive and neu- Medline roradiologic central nervous system late effects in children treated 12. Roberts RM, Mathias JL, Rose SE. Relationship between diffusion on Pediatric Oncology Group (POG) P9605 (standard risk) and tensor imaging (DTI) findings and cognition following pediatric P9201 (lesser risk) acute lymphoblastic leukemia protocols TBI: a meta-analytic review. Dev Neuropsychol 2016;41:176–200 (ACCL0131): a methotrexate consequence? A report from the Chil- CrossRef Medline dren’s Oncology Group. J Pediatr Hematol Oncol 2014;36:8–15 13. Edelmann MN, Krull KR, Liu W, et al. Diffusion tensor imaging and CrossRef Medline neurocognition in survivors of childhood acute lymphoblastic leu- 27. Aukema EJ, Caan MW, Oudhuis N, et al. White matter fractional kaemia. Brain 2014;137(Pt 11):2973–83 CrossRef Medline anisotropy correlates with speed of processing and motor speed in 14. Sidaros A, Engberg AW, Sidaros K, et al. Diffusion tensor imaging young childhood cancer survivors. Int J Radiat Oncol Biol Phys 2009; during recovery from severe traumatic brain injury and relation to 74:837–43 CrossRef Medline clinical outcome: a longitudinal study. Brain 2008;131(Pt 2):559–72 28. ElAlfy M, Ragab I, Azab I, et al. Neurocognitive outcome and white CrossRef Medline matter anisotropy in childhood acute lymphoblastic leukemia sur- 15. Cheung YT, Sabin ND, Reddick WE, et al. Leukoencephalopathy and vivors treated with different protocols. Pediatr Hematol Oncol 2014; long-term neurobehavioural, neurocognitive, and brain imaging 31:194–204 CrossRef Medline outcomes in survivors of childhood acute lymphoblastic leukaemia 29. Genschaft M, Huebner T, Plessow F, et al. Impact of chemotherapy treated with chemotherapy: a longitudinal analysis. Lancet Haema- for childhood leukemia on brain morphology and function. PLoS tol 2016;3:e456–66 CrossRef Medline One 2013;8:e78599 CrossRef Medline 16. Krull KR, Cheung YT, Liu W, et al. Chemotherapy pharmacody- 30. Khong PL, Leung LH, Fung AS, et al. White matter anisotropy in namics and neuroimaging and neurocognitive outcomes in long- post-treatment childhood cancer survivors: preliminary evidence term survivors of childhood acute lymphoblastic leukemia. J Clin of association with neurocognitive function. J Clin Oncol 2006;24: Oncol 2016;34:2644–53 CrossRef Medline 884–90 CrossRef Medline 17. Pui CH, Relling MV, Sandlund JT, et al. Rationale and design of Total 31. Porto L, Preibisch C, Hattingen E, et al. Voxel-based morphometry Therapy Study XV for newly diagnosed childhood acute lymphoblas- and diffusion-tensor MR imaging of the brain in long-term survi- tic leukemia. Ann Hematol 2004;83(Suppl 1):S124–26 Medline vors of childhood leukemia. Eur Radiol 2008;18:2691–700 CrossRef 18. Welker KM, Patton A. Assessment of normal myelination with mag- Medline netic resonance imaging. Semin Neurol 2012;32:15–28 CrossRef 32. Schuitema I, Deprez S, Van Hecke W, et al. Accelerated aging, de- Medline creased white matter integrity, and associated neuropsychological 19. Delis DC, Kaplan E, Kramer JH. Delis-Kaplan Executive Function Sys- dysfunction 25 years after pediatric lymphoid malignancies. J Clin tem. San Antonio: Psychological Corporation; 2001 Oncol 2013;31:3378–88 CrossRef Medline 20. Wechsler D. Wechsler Abbreviated Scale of Intelligence. San Antonio: 33. Zou L, Su L, Xu J, et al. Structural brain alteration in survivors of Psychological Corporation; 1999 acute lymphoblastic leukemia with chemotherapy treatment: a 21. Wechsler D. Wechsler Intelligence Scale for Children. 4th ed. San voxel-based morphometry and diffusion tensor imaging study. Antonio: Harcourt Assessment; 2003 Brain Res 2017;1658:68–72 CrossRef Medline

AJNR Am J Neuroradiol 39:1919–25 Oct 2018 www.ajnr.org 1925 ORIGINAL RESEARCH PEDIATRICS

Apparent Diffusion Coefficient Levels and Neurodevelopmental Outcome in Fetuses with Brain MR Imaging White Matter Hyperintense Signal

X E. Katorza, X G. Strauss, X R. Cohen, X M. Berkenstadt, X C. Hoffmann, X R. Achiron, X E. Barzilay, and X O. Bar-Yosef

ABSTRACT BACKGROUND AND PURPOSE: One of the perplexing findings of fetal brain MR imaging is white matter T2 hyperintense signal. The aims of our study were initially to determine the main etiologies associated with white matter T2 hyperintense signal, then to examine whether the different etiologies have different ADC values, and, last, to assess the association of white matter T2 hyperintense signal with developmental outcome.

MATERIALS AND METHODS: This was a prospective cohort study of 44 MR imaging scans of fetal brains obtained for suspected brain pathologies at a tertiary medical center during 2011–2015. Clinical data were collected from electronic medical charts. ADC values were measured and averaged in the frontal, parietal, occipital, and temporal lobes. Neurodevelopmental assessments were performed with the Vineland Adaptive Behavior Scales II.

RESULTS: Half of the cases of MRI hyperintense T2 signal of the fetal brain were associated with congenital cytomegalovirus infection. The other half were mainly idiopathic. Thus, the study group was divided to subgroups positive and negative for cytomegalovirus. Both groups had hyperintense signal in the temporal lobe. The group positive for cytomegalovirus had involvement of the parietal lobe. Only this group had increased ADC values in the temporal and parietal lobes. There was no association between the neurodevelopment outcome and the etiologies or ADC values.

CONCLUSIONS: T2 hyperintense signal in fetal brain MRI associated with positive cytomegalovirus infection has increased ADC values in the temporal and parietal lobes, suggestive of brain edema in these areas. However, the association between this finding and neurode- velopment outcome requires further evaluation.

ABBREVIATIONS: CMV ϭ cytomegalovirus; fbMRI ϭ fetal brain MRI; ICC ϭ interclass correlation coefficient; VABS ϭ Vineland Adaptive Behavior Scales II; WMHS ϭ white matter hyperintense signal

etal brain MR imaging (fbMRI) has been increasingly used in hand, the validity and relevance of this finding have been Frecent years as a means of tracking normal and pathologic fetal questioned.5 brain maturation.1 One of the perplexing findings of fbMRI is In recent years, DWI and its ADC metric have become a quan- white matter T2 hyperintense signal (WMHS). On the one hand, titative method for evaluation of fetal brain maturation.6 Previous it has been associated with in utero brain pathologies, such as studies showed that ADC values of the developing fetal brain ischemia and cytomegalovirus (CMV) infection.2-4 On the other correlate with fetal brain maturation.7-11 Deviation from normal ADC values has been shown to be associated with brain patho- logies such as ischemia, CMV infection, and ventriculo- megaly.2,12,13 Postmortem studies of animal and human fetuses Received February 26, 2018; accepted after revision July 19. with hypoxic-ischemic brain injury have demonstrated a transi- From the Antenatal Diagnostic Unit (E.K., G.S., R.C., R.A., E.B.), Department of Ob- stetrics and Gynecology, Pediatric Neurology Unit (O.B.-Y.), The Danek Gertner tion from low ADC values after the injury to increased values 7 Institute of Human Genetics (M.B.), and Neuroradiology Unit (C.H.), Department of days after the injury. This transition was associated with histologic Diagnostic Radiology, Chaim Sheba Medical Center, Tel-Hashomer, Israel; and Sackler School of Medicine (E.K., G.S., R.C., M.B., C.H., R.A., E.B., O.B.-Y.), Tel Aviv findings changing from initial cytotoxic edema and swollen University, Tel Aviv, Israel. astrocytes to vasogenic edema, astrogliosis, and abundance of E. Katorza and G. Strauss contributed equally to this work. macrophages.2,14 Thus, ADC and its association with histopa- Please address correspondence to Omer Bar-Yosef, MD, PhD, Pediatric Neurology Unit, Department of Pediatrics, Chaim Sheba Medical Center, Tel-Hashomer thology could be used to test the validity and meaning of T2 5262100, Ramat Gan, Israel; e-mail: [email protected] hyperintensity. http://dx.doi.org/10.3174/ajnr.A5802 The aims of our study were initially to determine the main

1926 Katorza Oct 2018 www.ajnr.org etiologies associated with WMHS, then to examine whether the different etiologies have different ADC values, and, last, to assess the association of WMHS with developmental outcome.

MATERIALS AND METHODS Subjects This was a prospective cohort study of women who were referred for fetal brain MR imaging to our tertiary medical center, Sheba Medical Center, Ramat Gan, Israel, between 2011 and 2015. The cohorts for this study were chosen on the basis of identification of a hyperintense signal on the T2-weighted sequences. Demo- graphic and clinical data were collected from the electronic re- cords of each patient. Data obtained from the records included maternal history, prenatal screening tests, imaging results from anatomic sonogra- FIG 1. ADC calculation was performed on 8 circular ROIs:2onthe phy and MR imaging, maternal CMV status, and perinatal his- white matter of both frontal (1, 2), parietal (3, 4), temporal (5, 6), and occipital lobes (7, 8). A circular ROI was placed over the desired ana- tory. Fetal cytomegalovirus infection was confirmed by either am- tomic area, ranging from 75 to 98 mm2. niocentesis performed during pregnancy or by the presence of CMV DNA in neonatal urine or saliva. The study was approved by Sheba Medical Center, Ramat Gan, White Matter T2 Hyperintense Signal Israel, institutional ethics board. Informed consent was obtained WMHS is a subjective interpretation of the signal from the fetal from each participant in the study prospectively. brain white matter on T2 sequences, made by the radiologist. This diagnosis is established when specific areas of the brain MR Imaging Scans white matter appear hyperintense in comparison with other Fetal brain MR imaging was performed using a 1.5T system areas or with that expected according to the gestational age 16 (Optima MR450w with GEM Suite; GE Healthcare, Milwaukee, (Fig 2). Wisconsin). Single-shot fast spin-echo T2-weighted sequences in 3 orthogonal planes were performed using a half-Fourier tech- Interobserver Validity of ADC Measurements nique (NEX ϭ 0.53) with the following parameters: section thick- To validate the consistency of measurements and reliability of ness, 3 or 4 mm; no gap; flexible coil (8-channel cardiac coil). results, 2 observers evaluated 10 fetuses. Interobserver variability FOV was determined by the size of the fetal head with a range was assessed by the interclass correlation coefficient (ICC). We of 24 ϫ 24 to 30 ϫ 30 cm; acquisition time was between 40 and considered an ICC value of Ն0.8 as excellent agreement. 45 seconds (matrix, 320 ϫ 224; TE, 90 ms; TR, 1298 ms; pixel bandwidth, 122 Hz/pixel; specific absorption rate values, 1.1– Neurodevelopmental and Hearing Outcome 1.7 W/kg). A DWI sequence in 1, 2, or 3 orthogonal planes was Children were assessed by the Vineland Adaptive Behavior Scales, then performed, with a 40-cm FOV, b-values of 0 and 700 ms, 2nd edition (VABS), which is a structured parent interview assess- and a slice thickness of 4 mm with no gap. All MR images were ing 4 different domains of behavior: communication, daily living obtained by the same protocol at our institution and assessed skills, socialization, and motor skills. All 4 domains are included by a specialist in fetal sonography and a neuroradiologist ex- in an adaptive composite score.17,18 VABS assessment was con- pert in MR imaging as previously published.15 ducted by a phone interview by 2 medical students trained and supervised by a pediatric neurologist and child development ex- Target Variables pert experienced in conducting VABS (O.B.-Y.). Validation of the ADC calculation was performed on 8 circular ROIs: 2 on the white phone interviews were assessed by correlation of the ICC to the matter of both frontal, parietal, temporal, and occipital lobes. A VABS scores of 15 children (divided into 7 and 8 children between circular ROI was used, placed over the desired anatomic area, ranging the students) evaluated by the pediatric neurologist (O.B.-Y.). from 75 to 98 mm2, similar to that accepted in the recent literature.11 The ICC was Ͼ0.83 for the 4 VABS domains and the composite Examples of ROI placements over the various anatomic areas can be score. Scores of children were considered abnormal if the stan- seen in Fig 1. The ROIs were placed with the following anatomic dard score was Ͻ70. considerations: Frontal lobe ROIs were placed in the anterior part of Hearing outcome was assessed 2–3 days after delivery by Tran- the frontal lobes above the lateral ventricles, parietal lobe ROIs were sient Evoked Otoacoustic Emissions. Neonates positive for CMV placed in the posterior part of the parietal lobes above the lateral were tested routinely before 1 month of age by brain stem audi- ventricles, temporal lobe ROIs were placed anterior to the temporal tory evoked potential, then at 3, 6, and 12 months by Transient horns, and occipital lobe ROIs were placed posterior to the lateral Evoked Otoacoustic Emissions and behavioral assessment. ventricles. The ADC was calculated using IntelliSpace workstation software (Phillips Healthcare, Best, the Netherlands; https://www. Statistical Analysis usa.philips.com/healthcare/solutions/clinical-informatics/ Categoric variables were expressed as number and percentage. enterprise-imaging-pacs). Distribution of continuous variables was assessed using a histo- AJNR Am J Neuroradiol 39:1926–31 Oct 2018 www.ajnr.org 1927 FIG 2. T2 MR images (single-shot fast spin-echo T2-weighted sequences in 3 orthogonal planes using a half-Fourier technique, NEX ϭ 0.53) of 2 fetuses at 33 weeks of gestational age. The CMV-positive fetus has diffuse WMHS (white arrowheads), unilateral ventriculomegaly, and intraventricular adhesions (black asterisks), suggesting ventriculitis (A–C). The CMV-negative fetus has WMHS located in the white matter in the temporal lobes (white arrowhead)(D–F).

Table 1: Demographic and clinical characteristics of fetuses with white matter T2 test, or McNemar test as appropriate. a hyperintense signal Continuous variables were compared CMV-Positive CMV-Negative P using the Student t test or Mann-Whit- Characteristic (n = 22) (n = 22) Value ney test as appropriate. A 2-tailed P Ͻ Maternal age (yr) 30.5 (26.7–34.2) 31 (28.7–34.2) .74 Gestational age at MRI (wk) 33 (32–34) 33.5 (32–35) .69 .05 was considered statistically signifi- Previous pregnancies 2 (2–3) 2 (1–4) .94 cant. Analyses were performed with Previous labors 1 (1–1.75) 1 (0–2.25) .95 SPSS (Version 24.0, 2016; IBM, Ar- Abnormal outcome in previous labors 6 (27%) 7 (32%) .74 monk, New York). Abnormal maternal medical background 0 6 (27%) .02 Mode of conception 19 (86%) 22 (100%) .35 Spontaneous RESULTS IVF 2 (9%) 0 (0%) Demographic and Clinical Induced pregnancy 1 (5%) 0 (0%) Characteristics of the Study Sex (male) 12 (54%) 8 (38%) .36 Population Normal nuchal translucency scan results 21 (95%) 22 (100%) .99 Normal first trimester biochemical test results 21 (94%) 21 (94%) Ͼ.99 The study subjects comprised 43 single- Normal second trimester biochemical test results 22 (100%) 22 (100%) Ͼ.99 ton pregnant women and 1 single twins Normal early anatomic scan findings 22 (100%) 16 (72%) .02 pregnant woman who underwent fb- Normal late anatomic scan findings 21 (94%) 11 (60%) .009 MRI scans. All 44 fetuses included in this Note:—IVF indicates in vitro fertilization. study were in their third trimester of a Data are presented as median (interquartile range) for continuous variables or number (percentage) for categoric variables. pregnancy at the time of the fetal MR imaging, with a median gestational age gram and Q-Q plot. Continuous variables were described using of 32 weeks (interquartile range, 28.25–35 weeks). The average median and interquartile range or mean and SD as appropriate. maternal age was 33 years (interquartile range, 25–32 years). The Categoric variables were compared using the ␹2 test, Fisher exact indications for MR imaging included the following: maternal cy- 1928 Katorza Oct 2018 www.ajnr.org Table 2: Radiologic MR imaging findingsa In the CMV-positive group, there CMV-Positive CMV-Negative P Value were significantly less early and late Gestational age at MRI (wk) 33 (32–34) 33.5 (32–35) .6 anatomic scan findings (P ϭ .02, P ϭ Radiologic finding .009, respectively) (Table 1). Frontal T2 hyperintense signal 13/22 (59%) 6/22 (27.3%) .067 Parietal T2 hyperintense signal 14/22 (63.6%) 3/22 (13.6%) .002 Temporal T2 hyperintense signal 22/22 (100%) 21/22 (95.5%) Ͼ.999 Distribution of Hyperintense Signal Occipital T2 hyperintense signal 0/22 (0%) 0/22 (0%) Ͼ.999 and Additional MR Imaging T2 hyperintense signal at 3 lobes (frontal, 13/22 (59%) 3/22 (13/6%) .004 Findings parietal, and temporal) Both groups had similar rates of Additional findings 1/22 (4.5%) 7/22 (31.8%) .046 WMHS in the temporal lobe (P Ͼ a Data are presented as median (interquartile range) for continuous variables or number (percentage) for categoric .999). WMHS was not depicted in the variables. occipital lobe in both groups. How- Table 3: Comparison of ADC values of the study group with the control groupa ever, the CMV-positive group had sta- Lobe, Side Study Group ADC ADC Control P Value tistically higher rates of parietal hyper- Frontal intense signal (P ϭ .002) and a trend Ϫ Right CMV 1793 (191) 1809 (165) .754 toward higher rates in the frontal lobe Right CMVϩ 1872 (160) .160 (P ϭ .067). The CMV-positive group Left CMVϪ 1741 (141) .115 Left CMVϩ 1858 (193) .302 had more extended WMHS involving Parietal all 3 lobes (P ϭ .004). The CMV-nega- Right CMVϪ 1659 (200) 1748 (192) .099 tive fetuses had statistically significantly ϩ Right CMV 1840 (140) .057 higher rates (P ϭ .046) of minor addi- Left CMVϪ 1690 (188) .268 tional findings on imaging, including Left CMVϩ 1852 (138) .03 ϭ Temporal subarachnoid cyst (n 2), slightly en- Right CMVϪ 1702 (169) 1692 (164) .843 larged subarachnoid space (n ϭ 1), and Right CMVϩ 1835 (161) .002 lateral ventricle asymmetry (n ϭ 4) Ϫ Left CMV 1689 (193) .933 (Table 2 and Fig 2). Left CMVϩ 1846 (122) Ͻ.001 Occipital Right CMVϪ 1659 (224) 1731 (167) .171 Interobserver Validity of ADC Right CMVϩ 1723 (129) .830 Measurements Ϫ Left CMV 1675 (169) .224 ADC measurements showed excellent Left CMVϩ 1725 (111) .882 interobserver agreement for all regions, Note:—CMVϩ indicates CMV-positive; CMVϪ, CMV-negative. a Data are presented as mean (SD). ADC units are 106 mm2/s. with the ICC ranging between 0.81 and 0.97. tomegalovirus infection (n ϭ 31), family history of central ner- ADC Value Measurements vous system pathology (n ϭ 3), and abnormal findings during ADC values for each group were compared with the normal ADC fetal sonography (n ϭ 10). The abnormal sonographic findings values as published by Hoffmann et al.11 Fetuses in the CMV- included the following: lateral ventricular asymmetry (n ϭ 4), positive group were found to have statistically higher ADC values mega cisterna magna (n ϭ 2), small head circumference (n ϭ 2), in the temporal lobe bilaterally (P ϭ .002 and P Ͻ .001) and the fisted hands (n ϭ 1), and hyperechogenic bowel (n ϭ 1). Twenty- left parietal lobe (P ϭ .033) and a trend in the right parietal lobe two (50%) of the fetuses were found to be positive for CMV by (P ϭ .057) (Table 3). amniocentesis or by saliva or urinary CMV DNA testing after birth. ADC values were compared for each region between the 2 Thus, to examine the association between ADC values and etiology, study groups. Fetuses in the CMV-positive group were found to we divided the main study group into a CMV-positive group consist- have higher ADC values in the left frontal lobe (P ϭ .026), the ing of 22 fetuses with confirmed CMV infection and a CMV-negative parietal lobe bilaterally (right, P ϭ .001; left, P ϭ .002), and the group consisting of 22 fetuses with WMHS from an unknown temporal lobe bilaterally (right, P ϭ .011; left, P ϭ .002) (Table 4). etiology. These 2 distinct patient populations are described separately Delivery Data, Hearing, and Neurodevelopmental and compared by their demographic, clinical, imaging, and neu- Assessment rodevelopment characteristics. Neurodevelopmental assessment (VABS) and delivery data were The median gestational age, pregnancy history, and sex did analyzed for 20 children in the CMV-positive group and 18 chil- not differ statistically between the groups. All mothers in the dren in the CMV-negative group (Table 5). Six patients (2 from CMV-positive group had normal medical background, com- the CMV-positive group and 4 from the CMV-negative group) pared with 6 (27%) mothers in the CMV-negative group who refused to participate in the neurodevelopmental assessment. De- had significant abnormal medical background including livery data and age at developmental assessment were not statisti- thrombophilia, Raynaud disease, and a history of cerebellar cally different between the groups. Comparison of the VABS score infarction (P ϭ .02). for the 4 domains and the adaptive composite score showed only AJNR Am J Neuroradiol 39:1926–31 Oct 2018 www.ajnr.org 1929 a trend for motor skills (P ϭ .07). Four children in the CMV- and left frontal lobes. The ADC values of the CMV-negative group positive group and 1 child in the CMV-negative group had at least did not differ from those of the control group. Thus, although the 1 VABS score Ͻ70; however, this was not statistically significant. incidences of WMHS were similar in the temporal lobes, the Only 1 child had a hearing deficit; this child was CMV-positive median ADC values were different. Because ADC is a quanti- with a low VABS score. tative measure and not a subjective interpretation, it is possible that some WMHS of the CMV-negative group is actually DISCUSSION overinterpretation. WMHS is a puzzling finding in fetal MR imaging, and its signifi- As for the CMV-positive group, the specific involvement of cance is not completely understood. The aims of our study were the temporal lobes was previously described in fbMRI. WMHS is initially to explore the etiologies associated with this finding, then part of a spectrum of findings localized to this area, including to examine whether the different etiologies are characterized by cysts, dilation of the temporal horns, and reduced temporal lobe different ADC values, and, last, to investigate the association of volume.4,11,21,22 The reason for the specific vulnerability of the WMHS with neurodevelopmental outcome. temporal lobe is unclear. Most interesting, in addition to the pres- Diverse etiologies were found to be associated with WMHS in ence of WMHS in the temporal lobe, most of the fbMRIs in our this study. The major one (half of the study group), in accordance study depicted a similar signal in the parietal lobe. This finding with fetal MR imaging literature, is congenital CMV infec- was previously described in children with congenital CMV.23-25 tion.13,19,20 The precise etiologies of the second half of the group Thus, the combination of temporal and parietal hyperintense T2 are not completely clear. Ten fetuses had further imaging find- signal is potentially a sign of CMV infection involving the brain. ings; thus, the WMHS might be a different aspect of the overall Further involvement of WMHS including the frontal lobes may brain pathology. serve as an additional sign of congenital CMV infection. Comparison between the 2 groups demonstrated a difference The combination of WMHS and high ADC values in the in the distribution of the WMHS. Both groups had the same prev- CMV-positive group suggests a possible etiology for this finding. alence of WMHS in the temporal lobes, but the CMV-positive Hyperintense signal was associated at postmortem examination group had a higher rate in the parietal lobes and a trend in the with astrogliosis and extracellular brain edema in the subacute frontal lobes. In general, the CMV-positive group had more dif- stage of hypoxic-ischemic injury.2 Although hypoxic-ischemic in- fuse WMHS. jury pathogenesis is different from CMV infection, the simplest The median ADC values were higher in the CMV-positive explanation of the T2 finding is an increase in extracellular fluid group than in the CMV-negative group in the temporal, parietal, content in the white matter. Kotovich et al26 found low ADC values in fetuses with congenital CMV infection without T2 hy- Table 4: Comparison of ADC values between fetuses with CMV-positive infection and fetuses with isolated white matter perintense signal. In a previous study of the same group, Yaniv et a hyperintense signal al4 described postmortem histology consisting of edema and cel- Lobe, Side CMV-Positive CMV-Negative P Value lular infiltration of plasma cells, lymphocytes, and microglia. The Frontal difference in T2 hyperintense signal and ADC values between the Right 1872 (160) 1793 (192) .147 2 latter studies and ours might reflect a different ratio between Left 1858 (193) 1741 (141) .026 Parietal the edema and cellular component in the brain tissue. When the Right 1840 (140) 1659 (200) .001 microgliosis and edema component are more prominent, the Left 1852 (138) 1690 (188) .002 ADC values are larger than those of controls; whereas when Temporal the component of cellular degeneration and immune cell infil- Right 1835 (161) 1702 (169) .011 tration is more prominent, the ADC values are lower than Left 1846 (122) 1689 (193) .002 2,14,27-29 Occipital those of controls. Right 1723 (129) 1659 (224) .259 The association between WMHS and abnormal neurodevel- Left 1725 (111) 1675 (169) .254 opmental outcome was not found in this study. In the CMV- a Data are presented as mean (SD). ADC units are 106 mm2/s. positive group, 4 children (20%) had abnormal neurodevelop- mental findings or hearing loss. Based a Table 5: Delivery data, hearing, and VABS assessment on neurodevelopmental data from pre- CMV-Positive CMV-Negative vious studies, the expected range of Characteristic (n = 20) (n = 18) P Value these abnormal findings in children with Birth week 39.2 (38.5–39.9) 37.9 (39.7–40.6) .55 asymptomatic congenital CMV (normal Birth weight (percentile) 49 (31–73) 49 (13–68) .53 Apgar score at 5 min Ն9 20 (100%) 18 (100%) Ͼ.99 prenatal imaging findings) is 10%– 30,31 Hearing deficitb 1 (5%) 0 (0%) Ͼ.99 15%. Thus, it is difficult to assess Age (mo) at VABS test 30 (17–49) 27.5 (17.2–28.5) .75 whether the WMHS and the increased VABS motor skills 92.5 (81.75–104) 103 (95.25–108) .07 ADC values predict worse prognosis. VABS daily living skills 103 (99.25–117) 109 (99.25–117) .28 Combining the results from our VABS socialization 101 (98.5–105.75) 100 (96.75–115.75) .89 study and others2,11-14,26 investigating VABS communication 107.5 (100.25–113) 108 (99.25–117) .84 VABS adaptive score composite 102 (90–18) 107 (98–115) .16 the association of ADC and brain pa- Child with any VABS score Ͻ70 4 (20%) 1 (5.5%) .34 thologies suggests the following practi- a Data are presented as median (interquartile range) for continuous variables or as number (percentage) for categoric variables. cal advice for clinicians: The combina- b Hearing deficit was assessed by either transient evoked otoacoustic emissions or brain stem evoked potential. tion of WMHS and normal ADC 1930 Katorza Oct 2018 www.ajnr.org suggests that the source of WMHS is most likely without clinical brain diffusivity in fetal isolated mild ventriculomegaly. Eur Radiol significance. As for cases of abnormal ADC either below or above 2016;26:840–48 CrossRef Medline the expected value for gestational age, they raise the suspicion of 14. Baud O, Daire JL, Dalmaz Y, et al. Gestational hypoxia induces white matter damage in neonatal rats: a new model of periventricular leu- brain pathology such as CMV infection, ventriculomegaly, or komalacia. Brain Pathol 2004;14:1–10 CrossRef Medline ischemia. However, the association of abnormal ADC with neu- 15. Gat I, Hoffmann C, Shashar D, et al. Fetal brain MRI: novel classifi- rodevelopmental outcome is unclear. cation and contribution to sonography. Ultraschall Med 2016;37: Our study is limited by the size of the study group. A larger 176–84CrossRef Medline number of patients could be used to correlate the distribution of 16. Lipitz S, Hoffmann C, Feldman B, et al. Value of prenatal ultrasound and magnetic resonance imaging in assessment of congenital pri- T2 hyperintense signal or ADC values with neurodevelopment. mary cytomegalovirus infection. Ultrasound Obstet Gynecol 2010;36: On the other hand, the small number of children in each group 709–17 CrossRef Medline strengthens the significance of the difference in T2 hyperintense 17. Limperopoulos C, Majnemer A, Steinbach CL, et al. Equivalence re- signal and ADC values between the CMV-positive and -negative liability of the Vineland Adaptive Behavior Scale between in-per- groups. son and telephone administration. Phys Occup Ther Pediatr 2006;26: 115–27 Medline 18. Sparrow SS. Cicchetti D, Balla DA. Vineland Adaptive Behavior Scales. CONCLUSIONS 2nd ed. San Antonio; Pearson: 2005 ADC measurements supported the validity of T2 hyperintense 19. Oosterom N, Nijman J, Gunkel J, et al. Neuro-imaging findings in signal in fbMRI only in fetuses with CMV infection. The associa- infants with congenital cytomegalovirus infection: relation to tri- tion of this finding with neurodevelopmental outcome requires mester of infection. Neonatology 2015;107:289–96 CrossRef Medline further investigation. 20. de Vries LS, Gunardi H, Barth PG, et al. The spectrum of cranial ultrasound and magnetic resonance imaging abnormalities in congenital cytomegalovirus infection. Neuropediatrics 2004;35: REFERENCES 113–19 CrossRef Medline 1. Griffiths PD, Bradburn M, Campbell MJ, et al MERIDIAN collabor- 21. Cannie MM, Devlieger R, Leyder M, et al. Congenital cytomegalovi- ative group. Use of MRI in the diagnosis of fetal brain abnormalities rus infection: contribution and best timing of prenatal MR imag- in utero (MERIDIAN): a multicentre, prospective cohort study. ing. Eur Radiol 2016;26:3760–69 CrossRef Medline Lancet 2017;389:538–46 CrossRef Medline 22. Doneda C, Parazzini C, Righini A, et al. Early cerebral lesions in 2. Guimiot F, Garel C, Fallet-Bianco C, et al. Contribution of diffusion- cytomegalovirus infection: prenatal MR imaging. Radiology 2010; weighted imaging in the evaluation of diffuse white matter isch- 255:613–21 CrossRef Medline emic lesions in fetuses: correlations with fetopathologic findings. 23. van der Knaap MS, Vermeulen G, Barkhof F, et al. Pattern of white AJNR Am J Neuroradiol 2008;29:110–115 CrossRef Medline matter abnormalities at MR imaging: use of polymerase chain 3. Malinger G, Lev D, Lerman-Sagie T. Imaging of fetal cytomegalovi- reaction testing of Guthrie cards to link pattern with congenital rus infection. Fetal Diagn Ther 2011;29:117–26 CrossRef Medline cytomegalovirus infection. Radiology 2004;230:529–36 CrossRef 4. Yaniv G, Hoffmann C, Weisz B, et al. Region-specific reductions Medline in brain apparent diffusion coefficient in cytomegalovirus- 24. Manara R, Balao L, Baracchini C, et al. Brain magnetic resonance infected fetuses. Ultrasound Obstet Gynecol 2016;47:600–07 findings in symptomatic congenital cytomegalovirus infection. Pe- CrossRef Medline diatr Radiol 2011;41:962–70 CrossRef Medline 5. Fazekas F, Kleinert R, Offenbacher H, et al. Pathologic correlates of 25. Hart CK, Wiley S, Choo DI, et al. Developmental disabilities and incidental MRI white matter signal hyperintensities. Neurology intracranial abnormalities in children with symptomatic cytomeg- 1993;43:1683–89 CrossRef Medline alovirus and cochlear implants. ISRN Otolaryngol 2012;2012:502746 6. Schneider JF, Confort-Gouny S, Le Fur Y, et al. Diffusion-weighted CrossRef Medline imaging in normal fetal brain maturation. Eur Radiol 2007;17: 26. Kotovich D, Guedalia JS, Hoffmann C, et al. Apparent diffusion co- 2422–29 CrossRef Medline efficient value changes and clinical correlation in 90 cases of cyto- 7. Sartor A, Arthurs O, Alberti C, et al. Apparent diffusion coefficient megalovirus-infected fetuses with unremarkable fetal MRI results. measurements of the fetal brain during the third trimester of AJNR Am J Neuroradiol 2017;38:1443–48 CrossRef Medline pregnancy: how reliable are they in clinical practice? Prenat Diagn 27. Vannucci RC, Christensen MA, Yager JY. Nature, time-course, and 2014;34:357–66 CrossRef Medline extent of cerebral edema in perinatal hypoxic-ischemic brain dam- 8. Neil JJ, Shiran SI, McKinstry RC, et al. Normal brain in human age. Pediatr Neurol 1993;9:29–34 CrossRef Medline newborns: apparent diffusion coefficient and diffusion anisotropy 28. Lodygensky GA, West T, Moravec MD, et al. Diffusion characteris- measured by using diffusion tensor MR imaging. Radiology 1998; tics associated with neuronal injury and glial activation following 209:57–66 CrossRef Medline hypoxia-ischemia in the immature brain. Magn Reson Med 2011;66: 9. Girard N, Confort-Gouny S, Schneider J, et al. MR imaging of brain 839–45CrossRef Medline maturation. J Neuroradiol 2007;34:290–310 CrossRef Medline 29. Miller SP, Vigneron DB, Henry RG,et al. Serial quantitative diffu- 10. Toft PB, Leth H, Peitersen B, et al. The apparent diffusion coefficient sion tensor MRI of the premature brain: development in newborns of water in gray and white matter of the infant brain. J Comput Assist with and without injury. J Magn Reson Imaging 2002;16:621–62 Tomograph 1996;20:1006–11 CrossRef Medline CrossRef Medline 11. Hoffmann C, Weisz B, Lipitz S, et al. Regional apparent diffusion 30. Naing ZW, Scott GM, Shand A, et al. Congenital cytomegalovirus coefficient values in 3rd trimester fetal brain. Neuroradiology 2014; infection in pregnancy: a review of prevalence, clinical features, di- 56:561–67 CrossRef Medline agnosis and prevention. AustNZJObstet Gynaecol 2016;56:9–18 12. Erdem G1, Celik O, Hascalik, et al. Diffusion-weighted imaging eval- CrossRef Medline uation of subtle cerebral microstructural changes in intrauterine 31. Lipitz S, Yinon Y, Malinger G, et al. Risk of cytomegalovirus-associated fetal hydrocephalus. Magn Reson Imaging 2001;25:417–22 Medline sequelae in relation to time of infection and findings on prenatal im- 13. Yaniv G, Katorza E, Bercovitz R, et al. Region-specific changes in aging. Ultrasound Obstet Gynecol 2013;508–14 CrossRefMedline

AJNR Am J Neuroradiol 39:1926–31 Oct 2018 www.ajnr.org 1931 ORIGINAL RESEARCH PEDIATRICS

Longitudinal Findings of MRI and PET in West Syndrome with Subtle Focal Cortical Dysplasia

X Y. Sakaguchi, X H. Kidokoro, X C. Ogawa, X Y. Okai, X Y. Ito, X H. Yamamoto, X A. Ohno, X T. Nakata, X T. Tsuji, X T. Nakane, X H. Kawai, X K. Kato, X S. Naganawa, and X J. Natsume

ABSTRACT BACKGROUND AND PURPOSE: Despite the development of neuroimaging, identification of focal cortical dysplasia remains challenging. The purpose of this study was to show the longitudinal changes of MR imaging and FDG-PET in patients with West syndrome and subtle focal cortical dysplasia.

MATERIALS AND METHODS: Among 52 consecutive patients with West syndrome, 4 were diagnosed with subtle focal cortical dysplasia on 3T MR imaging. MR imaging and PET findings were evaluated longitudinally at onset and at 12 and 24 months of age.

RESULTS: At the onset of West syndrome, MR imaging demonstrated focal signal abnormalities of the subcortical white matter in 2 patients. In the other 2 patients, focal subcortical high-intensity signals became visible on follow-up T2WI as myelination progressed. PET at onset showed focal cortical hypometabolism in 3 patients, with 1 of these patients also having focal hypermetabolism and 1 having normal findings. On PET at 24 months, hypometabolism persisted in 2 patients and disappeared in 1, and hypermetabolism disappeared in 1. In 1 patient with normal MR imaging and PET findings at onset, focal hyperintensity and hypometabolism first appeared at 24 months of age. The findings on MR imaging and PET in these patients evolved differently with brain maturation and the clinical course.

CONCLUSIONS: Subtle focal cortical dysplasia can be undetectable on MR imaging at the onset of West syndrome and is not always accompanied by hypometabolism or hypermetabolism on PET. Longitudinal MR imaging and PET studies may be useful for detecting such lesions. Even in West syndrome with a congenital structural abnormality, PET findings evolve differently with brain maturation and the clinical condition.

ABBREVIATIONS: EEG ϭ electroencephalography; FCD ϭ focal cortical dysplasia; WS ϭ West syndrome

est syndrome (WS) is an age-dependent epileptic enceph- the cause is undetermined in 20% of patients.1 The presumed Walopathy characterized by a triad of epileptic spasms, hyp- etiology in such cases includes genetic abnormalities and focal sarrhythmia on electroencephalography (EEG), and neurodevel- cortical dysplasia (FCD). opmental regression. WS is attributed to various etiologies, but Despite recent developments in neuroimaging techniques, identification of FCD remains challenging. Subtle FCD can be missed on MR imaging at the onset of WS.2 It has also been re- Received April 23, 2018; accepted after revision July 8. From the Departments of Pediatrics (Y.S., H. Kidokoro, C.O., Y.O., Y.I., H.Y., A.O., ported that MR imaging findings of subtle FCD can disappear T. Nakata, J.N.), Radiology (T. Nakane, H. Kawai, S.N.), Radiological and Medical Lab- with white matter maturation.3 FDG-PET may suggest FCD even oratory Sciences (K.K.), and Developmental Disability Medicine (J.N.), Nagoya Uni- 4 versity Graduate School of Medicine, Nagoya, Japan; Brain and Mind Research if initial MR imaging findings are normal. However, the findings Center (H. Kidokoro, Y.I., H.Y., H. Kawai, S.N., J.N.), Nagoya University, Nagoya, Ja- of PET at onset evolve along with the patient’s epilepsy.5-9 On pan; and Department of Pediatrics (T.T.), Okazaki City Hospital, Okazaki, Japan. This work was supported by a Grant-in-Aid for Scientific Research (C) 23591492 and MR imaging evaluation of infants, the status of myelination a Grant-in-Aid for Scientific Research on Innovative Areas (Comprehensive Brain affects the findings, and MR imaging findings of FCD cannot Science Network) from the Ministry of Education, Culture, Sports, Science and 2 Technology of Japan. be detected or differ from those in older patients. Thus, lon- Please address correspondence to Jun Natsume, MD, PhD, Department of Pediat- gitudinal MR imaging and PET studies may be useful to diag- rics, Nagoya University Graduate School of Medicine, 65 Tsurumai-cho, Showa-ku, Nagoya, Aichi 466-8550, Japan; e-mail: [email protected] nose FCD accurately. Indicates open access to non-subscribers at www.ajnr.org We hypothesized that serial evaluation of MR imaging and Indicates article with supplemental on-line tables. PET from the onset to 2 years of age shows the evolution of Indicates article with supplemental on-line photo. subtle FCD findings and leads to the accurate diagnosis and http://dx.doi.org/10.3174/ajnr.A5772 location of the lesions in patients with WS. To verify this hy-

1932 Sakaguchi Oct 2018 www.ajnr.org pothesis, we reviewed our series of patients with WS in whom hydrate, intravenous midazolam or ketamine was administered MR imaging and PET were longitudinally performed during 45 minutes after the FDG injection. Patients were scanned for 60 2007 and 2013. minutes after intravenous administration of FDG. EEGs were not monitored during the PET scans. Instead, pediatric neurologists MATERIALS AND METHODS observed all patients throughout the scan and confirmed that This study was approved by the Research Ethics Committee at there were no seizure manifestations during the examination. Nagoya University Graduate School of Medicine. Informed con- Thirty-four axial images (4-mm-thick) were obtained, and the sent was obtained from patients who participated in clinical images from 45 to 60 minutes after FDG injection were used for investigations. the evaluation.

Patient Selection Assessment of MR Imaging and PET Findings Between 2007 and 2013, fifty-two children admitted to Nagoya MR imaging and PET findings were assessed with visual inspec- University Hospital were newly diagnosed as having WS. In a tion by 3 pediatric neurologists (J.N., Y.S., and H. Kidokoro) and systematic review of these 52 patients, 23 were diagnosed as hav- 2 radiologists (T. Nakane for MR imaging and K.K. for PET). One ing an unknown etiology at the onset of WS, meeting the follow- of the radiologists was a pediatric neuroradiologist (T. Nakane) ing criteria (previously termed “cryptogenic”): 1) normal birth regularly involved in the MR imaging evaluation of patients with and absence of any etiologic factors related to WS; 2) normal epilepsy. Another radiologist (K.K.) specialized in nuclear medi- development before onset and absence of neurologic abnormali- cine and evaluating PET findings of patients with epilepsy. MR ties at onset; 3) the occurrence of clusters of spasms without any imaging was independently evaluated by 3 pediatric neurologists other types of seizures before the onset of spasms; and 4) normal (J.N., Y.S., and H. Kidokoro) and 1 neuroradiologist (T. Nakane). laboratory and MR imaging findings at onset. Among the 23 pa- PET was evaluated independently by 3 pediatric neurologists tients with an unknown etiology at onset, 2 patients were later (J.N., Y.S., and H. Kidokoro) and 1 radiologist specialized in nu- diagnosed as having FCD on MR imaging. In the remaining 29 clear medicine (K.K.). Four reviewers agreed on the presence or patients with genetic, structural, or metabolic etiologies, 2 pa- absence of abnormal findings on MR imaging or PET, and they tients had subtle focal abnormalities of FCD on 3T MR imaging at decided on the distribution of the abnormalities by discussion. onset. Two other patients had diffuse or bilateral FCD that was On MR imaging, whether there were structural abnormalities, clearly recognized on MR imaging. These 2 patients were not in- abnormal high or low intensity signals, and delayed myelination cluded in this study because the focus was on patients with focal, was evaluated. On PET, focal hypometabolism or hypermetabo- subtle lesions that can be missed on conventional MR imaging. lism was defined as a regional decrease or increase, respectively, in Thus, the present study included 4 patients in whom subtle FCD FDG accumulation in Ն2 gyri on Ն2 slices. To assess localization was seen at onset or during the follow-up period. of PET abnormalities, we performed an MR imaging–PET coreg- istration technique using SPM12b software (http://www.fil.ion. Neuroimaging Protocol ucl.ac.uk/spm/software/spm12) and the image visualization soft- MR Imaging. MR imaging was performed longitudinally at the ware Register (Brain Imaging Center, Montreal Neurological In- diagnosis of WS before adrenocorticotropic hormone therapy stitute, Montreal, Quebec, Canada). and at 12 and 24 months of age. Additional MR imaging was performed according to clinical need. MR imaging was performed Clinical Data using a 3T scanner (Magnetom Trio, a Tim System; Siemens, Other clinical variables were collected from medical records. EEG Erlangen, Germany) with a 32-channel phased array head coil. All recordings were performed at onset, during adrenocorticotropic patients were sedated with oral chloral hydrate (80 mg/kg) before hormone therapy, and at least at 12 and 24 months of age. The the examination. When patients did not appear sufficiently se- EEG records were assessed by 3 pediatric neurologists (J.N., Y.S., dated after chloral hydrate intake, intravenous midazolam or ket- and H. Kidokoro). All patients with WS were followed by pediat- amine was administered. We acquired the following images: axial ric neurologists at our outpatient clinics until 5–7 years of age. T1WI (TR/TE, 400/11 or 700/6.8 ms; slice thickness, 5 mm); The developmental quotient or intelligence quotient was as- T2WI (TR/TE, 5210/72 or 5200/131 ms; slice thickness, 5 mm); sessed at the last follow-up using the Kinder Infant Develop- FLAIR images (TR/TE, 9000/139 ms; TI, 2500 ms; slice thickness, ment Scale, a developmental test standardized in Japan in 5 mm); sagittal T1WI (TR/TE, 500/6.8 ms; slice thickness, 5 mm); 1989.10 If the patient was older than 7 years of age, the coronal T2WI (TR/TE, 5210/72 or 5210/131 ms; slice thickness, Wechsler Intelligence Scale for Children, Fourth Edition, was 5 mm); and T1-weighted sagittal MPRAGE (TR/TE, 1570/2.2 ms; administered because the Kinder Infant Development Scale slice thickness, 1 mm) with reconstructed axial and coronal was appropriate for children younger than 7 years of age. De- images. velopmental status was considered as follows: 1) normal, when FDG-PET. FDG-PET was performed as part of the clinical routine patients had a developmental quotient of Ն80; 2) borderline, to search for underlying pathologies using a Headtome V scanner when patients had a developmental quotient between 70 and (Shimadzu, Kyoto, Japan). All patients were sedated with a chloral 79; 3) mild delay, when patients had a developmental quotient hydrate suppository (50 mg/kg) before the PET examination. between 50 and 69; and 4) severe delay, when patients had a When patients did not appear sufficiently sedated after chloral developmental quotient of Ͻ50. AJNR Am J Neuroradiol 39:1932–37 Oct 2018 www.ajnr.org 1933 FIGURE. MR imaging and PET findings of 4 patients. Longitudinal findings of T2-weighted MR imaging and PET of patient 1 (upper row, A–E), patient 2 (second row, F–I), patient 3 (third row, J–N), and patient 4 (lower row, O–S). Patient 1: note a focal hyperintense lesion in the subcortical white matter of the left frontal lobe on axial MR imaging at onset (arrow, A). PET hypometabolism is seen in the left frontotemporal lobe at 11 months of age (arrows, B), in the right frontotemporal lobe at 24 months of age (arrows, C), and in the left frontal lobe at 36 months of age (arrow, D). PET at 24 months of age also shows possible hypermetabolism in the left temporal lobe (arrowheads, C). A coregistered coronal image of PET on MR imaging at 36 months of age shows hypometabolism in the left orbitofrontal area that corresponds to the MR imaging lesion (arrow, E). Patient 2: a hyperintense area in the subcortical white matter of the left frontal lobe on axial MR imaging at onset (arrow, F). Multifocal hypometabolism in the frontal and temporal lobes is seen on PET at onset (arrows, G) that disappears at 24 months of age (H). No PET hypometabolism at 24 months of age (H) is seen. Hypometabolism in the left frontal lobe corresponds to the MR imaging lesion on a coronal MR imaging–PET coregistration image (arrows, I). Patient 3: normal MR imaging findings at onset (J). Hyperintense subcortical white matter is seen in the right precentral area on MR imaging at 12 months of age (arrow, K). Left parietotemporal hypometabolism and right frontal hypermetabolism on PET is seen at onset (arrow, L). Right frontal hypermetabolism disappears at 12 months of age (M). Left temporal hypometabolism is seen on PET at 24 months of age, whereas MR imaging shows no abnormality (arrow, N). Patient 4: both MR imaging (O) and PET (Q) findings are normal at onset. A hyperintense area in the right orbitofrontal white matter is seen on axial MR imaging at 24 months of age (arrow, P). Hypometabolism in the right orbitofrontal lobe corresponds to the MR imaging lesion (arrow, R and S) and is seen in the left temporal lobe on PET (arrows, R). RESULTS cortical hypometabolism in 3 patients, and one of them (patient The profiles of the 4 patients are summarized in On-line Table 1. 3) also had focal hypermetabolism. The longitudinal findings in The age at onset of spasms ranged from 2 to 18 months. The age at each patient are detailed below (Figure and On-line Figs 1–4). diagnosis of WS ranged from 4 to 19 months. All patients had Patient 1. This patient began having epileptic spasms at 7 months normal development before the onset of spasms and normal neu- of age. Her initial MR imaging at 11 months of age showed a focal rologic examination findings at onset. All patients received adre- nocorticotropic hormone therapy. After adrenocorticotropic hyperintense lesion on both T1WI and T2WI in the left frontal hormone therapy, 3 patients were free of seizures. Another one white matter at the level of the foramen of Monro (Fig 1A). She (patient 1) had a relapse of spasms at 17 months of age, and oral underwent gadolinium-enhanced MR imaging and methionine antiepileptic drugs were administered. The patient became free of PET to rule out a brain tumor, and both had negative findings. seizures from 2 years of age. The developmental quotient or intel- FDG-PET at onset showed hypometabolism in the left frontotem- ligence quotient at the last follow-up was normal in 3 patients and poral area (Fig 1B), which was more widespread than the lesion on mildly delayed in 1 patient. MR imaging. At 24 months of age, PET showed hypometabolism in the right frontotemporal lobe (arrows) contralateral to the MR MR Imaging and PET Findings imaging lesion and possible hypermetabolism in the left temporal MR imaging, PET, and EEG findings are summarized in On-line lobe (arrowheads) (Fig 1C). At 36 months of age, the area of Table 2. The initial PET scans at the onset of WS showed focal hypometabolism was localized to the left frontal lobe overlying 1934 Sakaguchi Oct 2018 www.ajnr.org the MR imaging lesion (Fig 1D, -E). The EEG at 24 months of age DISCUSSION showed repetitive focal spikes, polyspikes, and slow waves in the Recent developments in neuroimaging technology have made it left frontal area, concordant with the MR imaging lesion. The possible to understand the causes of WS, such as FCD. However, paroxysmal epileptic discharges were less frequent at 36 months it is still difficult to accurately diagnose microscopic lesions. of age. At the last follow-up at 5 years of age, the patient was In the present study, longitudinal MR imaging and PET find- free of seizures with antiepileptic drugs (levetiracetam and ings of patients with WS and subtle MR imaging lesions of FCD topiramate). She had mild a developmental delay. Her devel- are presented. In 2 cases (patients 3 and 4), no MR imaging ab- opmental quotient by the Kinder Infant Development Scale at normality was visible in the early infantile period, but an MR 5 years of age was 57. imaging abnormality became visible later with progress of myeli- nation in the surrounding white matter. PET showed additional Patient 2. The patient had late-onset spasms at 18 months of age, regional hypometabolism remote from the MR imaging lesions. with otherwise no previous medical history of neurological ab- In patient 3, PET at the onset of WS also showed hypermetabo- normalities. T2WI at onset showed slight hyperintensity in the left lism in the MR imaging lesion that was visible later at 12 months frontal white matter (Fig 1F). PET at onset showed multifocal of age. hypometabolism in the left frontal, left temporal, and right frontal Two of the 4 patients had negative MR imaging findings in the areas (Fig 1G, -I). At 24 months of age, PET showed no abnor- early infantile period, and they were regarded as having WS of mality, even at the site of the MR imaging lesion (Fig 1H). EEG unknown etiology at onset. Abnormal white matter hyperinten- at 24 months of age showed sporadic spikes, polyspikes, and sity became visible on the T2-weighted follow-up MR imaging. sharp waves in the left frontal area, concordant with the MR Signal change of the white matter on MR imaging occurs during imaging lesion. At the last follow-up at 5 years of age, the the first 2 years of life as a result of myelination. Myelination is patient was free of seizures without antiepileptic drugs, except recognized as a high signal on T1WI and a low signal on T2WI. In for febrile seizures. Her psychomotor development was nor- general, myelination progresses from caudal to cephalad and mal. Her developmental quotient by the Kinder Infant Devel- from dorsal to ventral.11 Therefore, the subcortical white matter opment Scale at 5 years of age was 95. matured last (other than the in calcarine and Rolandic areas), proceeding from the occipital region anteriorly to the anterior Patient 3. The patient had onset of spasms at 2 months of age. MR frontal and temporal lobes. Changes of myelination on T2WI ap- imaging findings at onset were normal (Fig 1J), but MR imaging pear at 14–18 months of age in the midfrontal subcortical white at 12 months of age showed a hyperintense area in the right frontal matter and at 24–30 months of age in the anterior frontal subcor- subcortical white matter (Fig 1K). The hyperintense lesion was tical white matter.11 In the present study, the lesion in the precen- more clearly recognized at 24 months of age with progress of tral area (patient 3) became visible at 12 months of age, and the myelination in the surrounding white matter. PET at 4 months of lesion in the orbitofrontal area (patient 4) became visible at 24 age showed left parietotemporal hypometabolism and right fron- months of age. The age at detection of the lesion on MR imaging tal hypermetabolism (Fig 1L). At 12 and 24 months of age, hyper- matched the period of myelination. Before the maturational metabolism in the right frontal lobe disappeared (Fig 1M), while change of subcortical white matter, it is difficult to detect the PET showed left temporal hypometabolism and MR imaging characteristic findings of FCD, such as blurring of the gray-white showed no abnormality (Fig 1N). The EEG at 24 months of age matter junction and signal changes in subcortical white mat- showed sporadic polyspikes in the left temporal lobe that were ter.12,13 Thus, all patients with WS of unknown etiology at diag- concordant with PET hypometabolism. At the last follow-up at 7 nosis should be re-evaluated by MR imaging at 18–24 months of years of age, the patient was free of seizures without antiepileptic age. drugs. His intelligence quotient by the Wechsler Intelligence Scale Concerning FDG-PET findings, it has been reported that mat- for Children, Fourth Edition, was 84. His intelligence level was urational change occurs in regional cerebral glucose metabolism, within normal limits, but he had an attention deficit/hyperactivity especially during the infantile period.14,15 During this period, the disorder. cerebral glucose uptake pattern progresses from the subcortical gray matter to the cortex and from posterior-to-anterior cortical Patient 4. The patient began having spasms at 3 months of age. areas.14,15 In early infancy, PET shows lower uptake in the frontal She had no abnormalities on both MR imaging and PET at 4 and and temporal lobes than in the parietal and occipital lobes, and 12 months of age (Fig 1O, -Q). The right orbitofrontal MR imag- frontotemporal pathologic hypometabolism can be missed in this ing lesion and hypometabolism became evident at 24 months of period. In the present study, the frontal hypometabolism at onset age (Fig 1P, -R ,-S). PET at 24 months of age showed hypometab- in patient 4 was unclear, possibly due to the nature of the glucose olism in the left temporal lobe at a different site from the MR metabolism pattern in the infantile period. imaging lesion (Fig 1R). EEG at 24 months of age showed sharp FDG-PET is useful for detecting FCD in WS. A previous study waves in the right frontal and left temporal areas that were con- showed that PET was more sensitive than MR imaging for patients cordant with 2 areas of PET hypometabolism. At the last fol- with a mild degree of cortical dysplasia, though MR imaging was low-up at 5 years of age, the patient was free of seizures without performed using a 1.5T scanner.16 More recent reports with 1.5T antiepileptic drugs. She had normal psychomotor development. or 3T MR imaging scanners have also shown the usefulness of PET Her developmental quotient by the Kinder Infant Development in patients with MR imaging negative for FCD, especially with Scale was 91 at 5 years of age. PET/MR imaging coregistration.17 Chugani et al18 showed that AJNR Am J Neuroradiol 39:1932–37 Oct 2018 www.ajnr.org 1935 PET hypometabolism was consistent with EEG localization of ep- in brain tumors26 or an abnormal gyral pattern, increased cortical ileptogenic foci in patients with MR imaging negative for WS. thickness, or irregularity of the cortical–white matter junction as Asano et al19 also reported that the ictal electrocorticographic seen in polymicrogyria.27,28 It has been reported that FCD has a findings associated with spasm originated within a PET hypo- variety of features on MR imaging studies and does not always metabolic region. In the present study, PET showed hypometab- show the characteristic signs, such as cortical thickening, blurring olism or hypermetabolism that corresponded to the MR imaging of the gray-to-white matter surface, and signal changes in the lesions, but also hypometabolism in areas remote from the MR underlying white matter.29 From these observations, the MR im- imaging lesions. In 2 patients (patients 3 and 4), the remote hy- aging findings in the present study suggest FCD even without pometabolic areas were accompanied by focal independent epi- confirmation by histopathology. Second, a strong limitation of leptiform discharges on the interictal EEG. The remote hypo- the present study is that the sample size was very small. Studies metabolic areas with focal epileptiform discharges may indicate with a large number of patients are needed to draw conclusions on latent dysplastic lesions or secondary epileptogenic foci. the usefulness of MR imaging and PET. Third, MR imaging and Even in the region of structural abnormalities such as FCD, PET were assessed by visual inspection without using quantitative PET findings evolve with time from hypometabolism to normal, or statistical techniques (such as volumetric MR imaging analysis hypermetabolism to normal, or normal to hypometabolism dur- or quantification of glucose metabolism). Using these techniques ing the first years of life. Previous studies have demonstrated that may increase sensitivity and objectivity. FDG-PET showed hypometabolism in the region of FCD, usually corresponding to, but somewhat more extensive than, the MR CONCLUSIONS imaging lesion.20 In contrast, cases with FCD lesions showing A case series of patients with WS with FCD who showed subtle or hypermetabolism have also been reported.21 While subclinical unrecognized MR imaging abnormalities at diagnosis was pre- seizures during the scan possibly cause hypermetabolism on PET, sented. The findings on MR imaging and PET in these patients it has been reported that hypermetabolism can be seen in regions evolved differently with brain maturation and the clinical course with frequent epileptiform discharges, even in the interictal state, of WS. Their seizure outcomes were consistently favorable com- especially with FCD.22 While several studies have shown that PET pared with previous reports. Serial MR imaging and PET studies has an advantage for detecting subtle FCD, Sankar et al2 reported may be useful to accurately diagnose the etiology and make the that PET was not as sensitive as follow-up MR imaging in patients appropriate prediction of seizure outcomes in WS. with WS. The present observation of normal PET findings at the site of FCD at some points is consistent with the results of Sankar REFERENCES et al. Conversely, this serial PET study showed hypometabolism 1. Riikonen R. Epidemiological data of West syndrome in Finland. or hypermetabolism at the site of FCD on PET scans at least at 1 Brain Dev 2001;23:539–41 CrossRef Medline 2. Sankar R, Curran JG, Kevill JW, et al. Microscopic cortical dysplasia point in all patients. Evaluation of serial PET with follow-up MR in infantile spasms: evolution of white matter abnormalities. AJNR imaging has a complementary effect in detecting subtle lesions. Am J Neuroradiol 1995;16:1265–72 Medline Generally, patients with WS with FCD have poor seizure con- 3. Eltze CM, Chong WK, Bhate S, et al. Taylor-type focal cortical trol by antiepileptic drugs and are candidates for resective sur- dysplasia in infants: some MRI lesions almost disappear with gery.23,24 However, all patients in the present study had favorable maturation of myelination. Epilepsia 2005;46:1988–92 CrossRef Medline seizure outcomes without surgery, at least until preschool age, and 4. Chugani HT, Shields WD, Shewmon DA, et al. Infantile spasms, I: the developmental outcome was also favorable in 3 patients. This PET identifies focal cortical dysgenesis in cryptogenic cases difference may be explained by the different characteristics of the for surgical treatment. Ann Neurol 1990;27:406–13 CrossRef MR imaging and PET findings of FCD in the present cases. Typ- Medline ical MR imaging findings of FCD are a focal area of cortical thick- 5. Itomi K, Okumura A, Negoro T, et al. Prognostic value of positron emission tomography in cryptogenic West syndrome. Dev Med ness and blurring of the gray-white matter junction, with or with- Child Neurol 2002;44:107–11 CrossRef Medline out signal change in subcortical white matter on T2WI and 6. Maeda N, Watanabe K, Negoro T, et al. Transient focal cortical hy- FLAIR.25 All patients in the present study, however, showed only pometabolism in idiopathic West syndrome. Pediatr Neurol 1993;9: a hyperintense signal on T2WI in the subcortical white matter, 430–34 CrossRef Medline 7. Maeda N, Watanabe K, Negoro T, et al. Evolutional changes of without cortical thickness or blurring of the gray-white matter cortical hypometabolism in West’s syndrome. Lancet 1994;343: junction. Additionally, PET hypometabolism was not persistently 1620–23 CrossRef Medline observed at the MR imaging lesion of FCD. The different neuro- 8. Natsume J, Maeda N, Itomi K, et al. PET in infancy predicts long- imaging characteristics and, probably, prompt initiation of adre- term outcome during adolescence in cryptogenic West syndrome. nocorticotropic hormone with a short treatment lag after onset AJNR Am J Neuroradiol 2014;35:1580–85 CrossRef Medline 9. Natsume J, Watanabe K, Maeda N, et al. Cortical hypometabolism might be associated with the favorable outcomes in the present and delayed myelination in West syndrome. Epilepsia 1996;37: study. 1180–84 CrossRef Medline The present study has some limitations. First, the diagnosis of 10. Aoki S, Hashimoto K, Ikeda N, et al. Comparison of the Kyoto Scale FCD was not confirmed pathologically because the patients of Psychological Development 2001 with the parent-rated Kinder showed favorable seizure outcomes. The differential diagnosis of Infant Development Scale (KIDS). Brain Dev 2016;38:481–90 CrossRef Medline the MR imaging lesion may include benign tumor and polymi- 11. Barkovich AJ, Raybaud C. Pediatric Neuroimaging. 5th ed. Philadelphia: crogyria. However, none of the patients showed less cortical Lippincott Williams & Wilkins; 2012 thickening and intravenous contrast enhancement than observed 12. Ballesteros MC, Hansen PE, Soila K. MR imaging of the developing 1936 Sakaguchi Oct 2018 www.ajnr.org human brain, Part 2: postnatal development. Radiographics 1993;13: 21. Namer IJ, Valenti-Hirsch MP, Scholly J, et al. Hypermetabolism dur- 611–22 CrossRef Medline ing resting-state FDG-PET suggesting intrinsic epileptogenicity in 13. Barkovich AJ, Kjos BO, Jackson DE Jr, et al. Normal maturation of focal cortical dysplasia. Clin Nucl Med 2014;39:993–95 CrossRef the neonatal and infant brain: MR imaging at 1.5 T. Radiology 1988; Medline 166:173–80 CrossRef Medline 22. Bansal L, Miller I, Hyslop A, et al. PET hypermetabolism in medi- 14. Chugani HT, Phelps ME. Maturational changes in cerebral function cally resistant childhood epilepsy: incidence, associations, and sur- in infants determined by 18FDG positron emission tomography. gical outcome. Epilepsia 2016;57:436–44 CrossRef Medline Science 1986;231:840–43 CrossRef Medline 23. Kang JW, Rhie SK, Yu R, et al. Seizure outcome of infantile spasms 15. Kinnala A, Suhonen-Polvi H, Aa¨rimaa T, et al. Cerebral metabolic with focal cortical dysplasia. Brain Dev 2013;35:816–20 CrossRef rate for glucose during the first six months of life: an FDG positron Medline emission tomography study. Arch Dis Child Fetal Neonatal Ed 1996; 24. Yum MS, Ko TS, Lee JK, et al. Surgical treatment for localization- 74:153–57 Medline related infantile spasms: excellent long-term outcomes. Clin Neurol 16. Kim SK, Na DG, Byun HS, et al. Focal cortical dysplasia: comparison Neurosurg 2011;113:213–17 CrossRef Medline of MRI and FDG-PET. J Comput Assist Tomogr 2000;24:296–302 25. Widdess-Walsh P, Diehl B, Najm I. Neuroimaging of focal cortical CrossRef Medline dysplasia. J Neuroimaging 2006;16:185–96 CrossRef Medline 17. Desarnaud S, Mellerio C, Semah F, et al. 18F-FDG PET in drug- 26. Bronen RA, Vives KP, Kim JH, et al. Focal cortical dysplasia of Tay- resistant epilepsy due to focal cortical dysplasia type 2: additional value of electroclinical data and coregistration with MRI. Eur J Nucl lor, balloon cell subtype: MR differentiation from low-grade tu- Med Mol Imaging 2018;45:1449–60 CrossRef Medline mors. AJNR Am J Neuroradiol 1997;18:1141–51 Medline 18. Chugani HT, Shewmon DA, Shields WD, et al. Surgery for intracta- 27. Battal B, Ince S, Akgun V, et al. Malformations of cortical ble infantile spasms: neuroimaging perspectives. Epilepsia 1993;34: development: 3T magnetic resonance imaging features. World J Ra- 764–71 CrossRef Medline diol 2015;7:329–35 CrossRef Medline 19. Asano E, Juha´sz C, Shah A, et al. Origin and propagation of epileptic 28. Leventer RJ, Guerrini R, Dobyns WB. Malformations of cortical spasms delineated on electrocorticography. Epilepsia 2005;46:1086–97 development and epilepsy. Dialogues Clin Neurosci 2008;10:47–62 CrossRef Medline Medline 20. Lee N, Radtke RA, Gray L, et al. Neuronal migration disorders: positron 29. Bast T, Ramantani G, Seitz A, et al. Focal cortical dysplasia: preva- emission tomography correlations. Ann Neurol 1994;35:290–97 CrossRef lence, clinical presentation and epilepsy in children and adults. Acta Medline Neurol Scand 2006;113:72–81 CrossRef Medline

AJNR Am J Neuroradiol 39:1932–37 Oct 2018 www.ajnr.org 1937 ORIGINAL RESEARCH PEDIATRICS

Imaging of Clival Hypoplasia in CHARGE Syndrome and Hypothesis for Development: A Case-Control Study

X C.M. de Geus, X J.E.H. Bergman, X C.M.A. van Ravenswaaij-Arts, and X L.C. Meiners

ABSTRACT BACKGROUND AND PURPOSE: We present the largest case series to date on basiocciput abnormalities in CHARGE syndrome (Coloboma of the eye, Heart defects, Atresia of the choanae, Retardation of growth and/or development, Genital and/or urinary abnormalities, and Ear abnormalities and/or deafness). We aimed to show that basiocciput abnormalities are common and may aid in diagnosis. We furthermore explored whether clivus size correlates with the type of chromodomain-helicase-DNA binding protein 7 gene (CHD7) mutation, which causes CHARGE syndrome, and with clinical criteria according to Blake et al and Verloes.

MATERIALS AND METHODS: We retrospectively analyzed the clivus of 23 patients with CHARGE syndrome with CHD7 mutations on MR imaging or CT. We recorded the size of the clivus, the Welcher angle, basilar invagination, and Chiari I malformations. We compared the clival size and Welcher angle of patients with CHARGE syndrome with those of 72 age-matched controls. Additionally, we tested for correlations between clivus size and mutation type or clinical criteria.

RESULTS: Eighty-seven percent of the patients with CHARGE syndrome had an abnormal clivus; 61% had a clivus Ͼ2.5 SD smaller than that of age-matched controls. An abnormally large Welcher angle was observed in 35%. Basiocciput hypoplasia was found in 70%, and basilar invagination, in 29%. None of the patients had a Chiari I malformation. At the group level, patients with CHARGE syndrome had a smaller clivus and larger Welcher angle than controls. No significant correlation between clivus size and mutation type or clinical criteria was found.

CONCLUSIONS: Most patients with CHARGE syndrome have an abnormal clivus. This suggests that clivus abnormalities may be used as an additional diagnostic tool. Our results provide evidence that CHD7, which is expressed in the presomitic mesoderm during somitogen- esis, plays an important role in the formation of the clivus.

ABBREVIATIONS: Ba-Xs ϭ exosphenobasion; Ba-Es ϭ endosphenobasion; CHARGE ϭ Coloboma of the eye, Heart defects, Atresia of the choanae, Retardation of growth and/or development, Genital and/or urinary abnormalities, and Ear abnormalities and deafness; CHD7 ϭ chromodomain-helicase-DNA binding protein 7 gene. The animal homologue is Chd7. Non-italicized CHD7 (human) and Chd7 (animal) refer to the protein.

oloboma of the eye, Heart defects, Atresia of the choanae, syndrome is a complex disorder with multiple congenital anomalies CRetardation of growth and/or development, Genital and/or uri- that occurs in approximately 6 in 100,000 live births.1 First de- nary abnormalities, and Ear abnormalities and deafness (CHARGE) scribed independently by Hall2 and Hittner et al3 in 1979, the acronym CHARGE was coined by Pagon et al in 1981.4 Many more features are associated with the syndrome, such as semicir- Received April 26, 2018; accepted after revision July 19. cular canal dysplasia, facial nerve palsy, anosmia with or without From the Departments of Genetics (C.M.d.G., J.E.H.B., C.M.A.v.R.) and Radiology olfactory bulb hypoplasia, delayed puberty, and cleft lip/palate. (L.C.M.), University of Groningen, University Medical Center Groningen, Groningen, 5 the Netherlands. Clinical criteria have been published by Blake et al in 1998, Ver- 6 7 C.M. de Geus was supported by the CHARGE Syndrome Foundation (CHARGE Syn- loes in 2005, and Hale et al in 2015, which aid in the clinical drome Foundation pilot grant) and a personal grant of the University Medical Cen- diagnosis (On-line Table 1). Guidelines for cranial imaging were ter Groningen. J.E.H. Bergman was supported by the Netherlands Organization for 8 Health Research and Development (ZonMW 92003460). published by de Geus et al in 2017. In 2004, CHARGE syndrome Please address correspondence to L.C. Meiners, MD, Department of Radiology, was found to be caused by mutations or deletions of the chro- University Medical Centre Groningen, University of Groningen, PO Box 30.001, NL-9700 RB Groningen, the Netherlands; e-mail: [email protected] modomain-helicase-DNA binding protein 7 gene (CHD7) gene, Indicates open access to non-subscribers at www.ajnr.org and molecular confirmation currently plays a pivotal role in the 9 Indicates article with supplemental on-line appendix and tables. diagnosis. Pathogenic mutations in the CHD7 gene usually occur Indicates article with supplemental on-line photo. de novo, though familial occurrence has been described.10 http://dx.doi.org/10.3174/ajnr.A5810 There is great phenotypic diversity in CHARGE syndrome, which

1938 de Geus Oct 2018 www.ajnr.org FIG 1. Clival abnormalities in CHARGE syndrome. A, Sagittal T1 scan of a 4.5-year-old boy without CHARGE syndrome. White lines show the measurement of the Ba-Es and Ba-Xs. B, A 22-month-old boy with CHARGE syndrome (patient 14). He has a hypoplastic sclerotome of the clivus (arrow) with a large Welcher angle. Although the top of the odontoid process of the dens does not extend cranially to the Chamberlain line, there is a slight angulation of the medulla oblongata without impression. C and D, Clival size versus age. White dots show length of the Ba-Es (C) and Ba-Xs (D) of the individual controls; black dots show the same parameters of individual patients with CHARGE syndrome.

FIG 2. Platybasia in patients with CHARGE syndrome. A, Sagittal T1 scan of a 4.5-year-old boy without CHARGE syndrome. The Welcher line is shown in white. B, A 22-month-old boy with CHARGE syndrome (patient 14). Note the large Welcher angle on the midsagittal T1 scan. C, Welcher angle versus age. White dots show the Welcher angle of the individual controls. Black dots show the Welcher angles of individual patients with CHARGE syndrome. complicates early diagnosis. CT and MR imaging in CHARGE may larger deletions were categorized as truncating (ie, mutations play an important role in the diagnosis by demonstrating congenital leading to a nonfunctional protein or no protein at all). CHD7 abnormalities of the labyrinth, which are present in almost all pa- missense mutations were categorized as nontruncating (ie, muta- tients and can be assessed on CT and MR imaging.11 Olfactory bulb tions leading to production of an altered protein that may have hypoplasia, cerebellar dysplasia, and other congenital brain abnor- residual function). CHD7 splice site mutations may have truncat- malities may be demonstrated on MR imaging, but they are not in- ing or nontruncating effects, and these mutations could therefore variably present in all patients.12,13 Morphologic changes of the cli- not be further categorized. All patients were scored using the vus in CHARGE syndrome were first described on neuroimaging by Blake et al5 and Verloes6 criteria (On-line Table 1). Fujita et al in 2009.14 A smaller size, a malformed shape, platybasia, Neuroimaging was performed in the authors’ hospital in 12 pa- basilar invagination, and Chiari I malformations have since been de- tients on a 1.5T scanner (Siemens, Erlangen, Germany) using a reg- scribed by several authors.15-18 Normally, the body of the sphenoid ular head coil. The remainder of the patients were scanned at other occupies the upper portion of the clivus and is joined to the basilar hospitals using different scanners (1–1.5T) of different brands using occipital bone to form the complete clivus (Fig 1A).19 Steepness (of different protocols. Only patients with sagittal T1 2D TSE, 3D T1 the clivus) may be quantified by the Welcher angle formed by a line MPRAGE, or a sagittal 2D T2 TSE imaging were included in this through the frontal skull base and a line along the dorsal clivus (Fig 2A). study. Three patients assessed at an external hospital had a CT scan In the present study, we have elaborated on the previously with the possibility of sagittal reconstruction of a transverse CT scan published basiocciput findings by evaluating a large cohort of of either a head scan or a mastoid scan. patients with molecularly proved CHARGE syndrome in compar- All neuroimaging studies were assessed and measurements ison with age-matched controls. We further attempted to corre- were made by an experienced pediatric neuroradiologist late clival size with mutation type and clinical criteria. (L.C.M.).

MATERIALS AND METHODS Controls Patients Age-matched controls from 6 age groups (0–3 months and 1, 2, 6, The diagnosis of CHARGE syndrome was molecularly confirmed 10, and 16 years of age) had been scanned for various neurologic in all patients. CHD7 nonsense and frameshift mutations and and endocrine indications, not suspicious for CHARGE syn- AJNR Am J Neuroradiol 39:1938–42 Oct 2018 www.ajnr.org 1939 drome or skull base abnormalities, on a 1.5T MR imaging sys- Table 1: Characteristics for 23 patients with CHARGE syndrome in tem (Siemens, Erlangen, Germany), in the authors’ hospital our case series between 2002 and 2014. All controls had sagittal 2D or 3D Characteristics T1-weighted MR imaging included in the scanning protocol. Median age (range) 20 Mo (3 days to 16 yr) Males 15 The brain scan findings had been assessed as normal by various Females 8 experienced neuroradiologists, and at selection, this assess- Criteria of Blake et al5 satisfieda ment was confirmed by an experienced pediatric neuroradi- Typical CHARGE syndrome 12 ologist (L.C.M.). Negative 8 Missing data 3 Criteria of Verloes6 satisfieda Radiologic Analysis of the Clivus Typical CHARGE syndrome 12 Anatomic definitions and measurements of the clivus were used Partial CHARGE syndrome 0 as described by Fujita et al.14 Clivus size was quantified by measuring Atypical CHARGE syndrome 5 the exosphenobasion (Ba-Xs) and endosphenobasion (Ba-Es, Fig Negative 0 Missing data 6 1A). This was done by measuring from the basion, the point of the Criteria of Hale et al7 criteria satisfieda 23 clivus at the midpoint on the anterior margin of the foramen mag- Mutation type (n ϭ 23)b num, to the ventral (Ba-Xs) and dorsal (Ba-Es) margins of the visible Truncating (nonsense, frameshift, 17 synchondrosis.14 The Welcher angle is formed by the intersection deletion) between the nasion-tuberculum line and the tuberculum-basion Nontruncating (missense) 2 Splice site 4 line.20 Basioccipital hypoplasia has been defined as hypoplasia of Ն1 a See On-line Table 1. of the 5 clival segments (sclerotomes) of the clivus (simplified from b Truncating mutations (nonsense or frameshift) lead to a decrease in CHD7 protein. 14 Fujita et al ). Basilar invagination is commonly defined as cranial Nontruncating (missense) mutations lead to production of an altered CHD7 protein, displacement of Ͼ5 mm of the tip of the odontoid above to the which may have residual function. Splice site mutations may have truncating or nontruncating effects. Chamberlain line. This line extends from the posterior margin of the foramen magnum anteriorly along the hard palate. Type I Chiari RESULTS malformation is defined as herniation of at least 1 cerebellar tonsil In total, 23 patients with an age range of 3 days to 16 years (median Ն 5 mm below the foramen magnum. age, 20 months) were included in this study. Table 1 summarizes The exosphenobasion, the endosphenobasion, the Welcher the patients’ clinical criteria according to Blake et al, Verloes, and angle, basilar invagination, and type I Chiari malformation were Hale et al and their type of CHD7 mutation. The full spectrum was measured. represented in patients with both clinically typical and atypical CHARGE syndrome. Statistical Analysis Figure 1C, -D summarizes the lengths of the Ba-Es and Ba-Xs The scans of the patients with CHARGE syndrome were com- for patients and controls with increasing age. For 2 patients, the pared with the findings on sagittal T1-weighted MR images of presence of basilar invagination could not be reliably determined 72 controls in 6 age groups: 0–3 months and 1, 2, 6, 10, and 16 (patients 13 and 20 in On-line Table 3). Most patients with years of age. CHARGE syndrome (87%, 20/23) had a clivus that was small or Measurements of Ba-Es, Ba-Xs, and Welcher angle in controls had abnormal morphology or both. Fourteen patients had a clivus were used to calculate age-specific mean and SD values. Patient of Ͼ2.5 SDs smaller than in their age-matched controls (61%, measurements were compared with age-specific mean control 14/23). In 8 patients (35%, 8/23), an extra synchondrosis was values. A clivus was determined to be abnormally small if it was seen. Just more than two-thirds (70%, 16/23) of the patients with Ͼ 2.5 SDs below its age-specific control. The Welcher angle was CHARGE syndrome showed very short clivi with loss of the nor- determined to be abnormally large if it was Ͼ2.5 SDs above its mal triangular shape (basiocciput hypoplasia), which was further age-dependent control value. illustrated by 9 of them having a Welcher angle of Ն2.5 SDs above Because the Ba-Es, Ba-Xs, and the Welcher angle were highly that in controls, indicating platybasia (39%, 9/23; Fig 2C). The correlated (On-line Table 2), we orthogonalized the data with Welcher angle varied between 124° and 176° (mean 140°, SD factor analysis. The first factor was then modeled with nonlinear 11.2°). regression on the normal data. To control for bias due to the The results for the comparison at the group level are shown in difference in age distribution of patients with CHARGE syn- On-line Tables 2 and 4 and the On-line Figure. Most of the pa- drome and controls, we computed the observed minus the pre- tients with CHARGE syndrome showed a reduced value in factor dicted values, which were then tested with a 2-sample t test (for 1. Factor 1 had a positive correlation with the Ba-Es and Ba-Xs additional statistical methods, see the On-line Appendix). A and a negative correlation with the Welcher angle. Despite some 2-sided Fisher exact test was performed to examine correla- patients with CHARGE syndrome having clearly normal values, tions between the size of the clivus and the type of mutation as a group they had significantly lower values (P ϭ 6 ϫ 10Ϫ6), (truncating versus nontruncating), the presence of choanal which correspond with a smaller Ba-Es and Ba-Xs and larger atresia or coloboma, and satisfaction of the criteria of Verloes6 Welcher angles. or Blake et al5 (listed in On-line Table 1). No t test was per- Six (29%, 6/21, missing data n ϭ 2) patients with CHARGE formed for the criteria of Hale et al7 because all patients satis- syndrome showed basilar invagination. In 2 of these patients, a fied these criteria. minor impression of the craniovertebral junction on the medulla 1940 de Geus Oct 2018 www.ajnr.org Table 2: Correlations between clivus size and clinical parameters prevalence of clival abnormalities and/or skull base hypoplasia in Size these articles is similar to our numbers. We found that 20/23 of Clivus patients with CHARGE syndrome (87%) had an abnormal clivus, <2.5 SDs either in morphology or size. At the group level, patients with Compared P CHARGE syndrome had a smaller clivus and larger Welcher angle Variable Normal with Controls Value (On-line Figure). However, only 14/23 (61%) had an abnormally Truncating mutation small clivus, defined as Ͼ2.5 SDs smaller than that of age- (total n ϭ 19)a ϩ 5 12 .12 matched controls (On-line Table 3 and Fig 1). Basilar invagina- Ϫ 20 tion was seen in only 6 patients (29%, 6/21, missing data in 2 Presence of choanal atresia patients), with only 2 of these patients showing a minimal impres- ϩ 1 4 .33 sion on the ventral medulla oblongata. None of the patients in the Ϫ 810 Presence of coloboma present study had a Chiari I malformation. If we combined our (total n ϭ 22)b study and the above-mentioned 2 case series, 51 of 56 patients ϩ 4 8 .36 with CHARGE syndrome (91%) had clival abnormalities and/or Ϫ 55 skull base hypoplasia, underscoring the potential of this feature as Presence of cleft a diagnostic tool in CHARGE syndrome. (total n ϭ 22)b ϩ 3 2 .61 However, because all of these case series represent nonran- Ϫ 611 domly selected samples, there is a danger of overestimation. All Verloes6 criteria satisfiedc samples may be biased toward the more severe end of the clinical ϩ 4 8 1.00 spectrum because more severely affected patients may be more Ϫ 23 likely to undergo cerebral imaging. Our data do not show corre- Blake et al5 criteria satisfiedc ϩ 3 5 1.00 lation between the severity of the disorder (satisfaction of clinical Ϫ 48 criteria) and the presence of clival abnormalities. In fact, in our Note:—ϩ indicates yes; Ϫ, no. case series, a large number of patients were atypical on the Verloes a Four patients had a splice site mutation and could not be classified as either trun- criteria (5/17) or negative on the Blake et al criteria (8/20), yet cating or nontruncating. clival abnormalities were found in most. This finding underscores b For 1 patient, no data regarding presence of coloboma were available. For 1 patient, no information regarding cleft lip/palate was available. the importance of clival abnormalities on imaging in supporting c See also On-line Table 1. the diagnosis, especially in mildly affected patients. Basilar invagination in patients with CHARGE syndrome may oblongata was suggested (see Fig 1B for an example). None of the be of clinical importance because it may cause compression of the patients exhibited herniation of the cerebellar tonsils. medulla with ensuing clinical symptoms. No obvious neurologic Table 2 shows the correlation among clivus size, clinical crite- symptoms that could be attributed to the basilar invagination ria, and type of mutation. No significant correlation was found were reported in the clinical data of the 2 patients who showed between the size of the clivus in the patients with CHARGE syn- possible involvement of the medulla (patients 14 and 15 in On- 14 drome and the mutation type (truncating versus nontruncating), line Table 3). Only 1 of the patients in the series of Fujita et al occurrence of facial clefts, ocular coloboma, atresia of the cho- had neurologic sequelae, but she was reported to have syringomy- anae, or satisfaction of the Verloes or Blake et al criteria. elia in addition to basilar impression. In our study, hypoplasia of the clivus was suggested as an all- DISCUSSION or-nothing event: If hypoplasia was present, the degree of hyp- In this study, the presence of clival abnormalities was assessed in a oplasia was severe (Ba-Es of Ϫ4toϪ10 SDs smaller than in con- large group of patients with CHARGE syndrome and compared trols, Fig 1 and On-line Table 3). The high variability and with that in healthy controls. incomplete penetrance of specific features are well-known aspects Clival abnormalities in CHARGE syndrome have been pub- of CHARGE syndrome.21 CHARGE syndrome is exclusively lished previously. Fujita et al14 were the first to publish examples caused by mutations in CHD7. The chromodomain helicase of basioccipital hypoplasia in 7/8 patients (88%) and associated DNA-binding protein 7 (protein CHD7)9 is essential in embryo- basilar invagination in 5 (63%). Furthermore, one of their pa- logic development, and mutations in CHD7 result in a wide range tients exhibited a Chiari I malformation and syringomyelia. Na- of features with incomplete penetrance. Much of the clinical vari- tung et al15 described a case with a short clivus, fused cervical ability is still unexplained, though there is some correlation be- vertebrae, occipitalization of the atlas, and basilar invagination. tween clinical severity and mutation type: Patients with a mis- Hoch et al16 found skull base hypoplasia in 9/10 and a dorsally sense CHD7 mutation generally have a milder presentation of angulated clivus in 7/10 patients. Mahdi and Whitehead17,18 de- clinical features.22 In our cohort, no correlations were found scribed a child with a coronal clival cleft in 2017 and recently between clivus length and mutation type, satisfaction of all published a study consisting of 15 genetically and clinically con- CHARGE criteria, or specific symptoms. Milder missense muta- firmed CHARGE cases, in which they reported a coronal cleft in tions are, however, fairly rare in CHARGE syndrome. This is re- 13 (87%) patients and clival hypoplasia without a cleft in the flected in our study because only 2/23 patients had a missense remaining 2 patients. mutation. Although we did not particularly assess the coronal cleft, the The precise function of the CHD7 protein in the formation of AJNR Am J Neuroradiol 39:1938–42 Oct 2018 www.ajnr.org 1941 the skull base is unknown. CHD7 is expressed in the presomitic 5. Blake KD, Davenport SL, Hall BD, et al. CHARGE association: an mesoderm during somitogenesis,23 in which it plays a role in con- update and review for the primary pediatrician. Clin Pediatr (Phila) trolling left-right symmetry. Somitogenesis is an important pro- 1998;37:159–73 CrossRef Medline 6. Verloes A. Updated diagnostic criteria for CHARGE syndrome: a cess in the formation of the clivus, which is formed from 4 occip- proposal. Am J Med Genet A 2005;133A:306–08 Medline ital somites through a complicated process. In this process, fusion 7. Hale CL, Niederriter AN, Green GE, et al. Atypical phenotypes asso- of the first 3 somites creates the rostral basiocciput.24 After for- ciated with pathogenic CHD7 variants and a proposal for broadening mation of a transient sclerotome called the proatlas, parts of the CHARGE syndrome clinical diagnostic criteria. Am J Med Genet fourth somite then form the basion. Chd7 knockdown zebrafish A 2016;170A:344–54 23,25 8. de Geus CM, Free RH, Verbist BM, et al. Guidelines in CHARGE exhibit irregularly shaped vertebrae, supporting the role of syndrome and the missing link: cranial imaging. Am J Med Genet C CHD7 in somitogenesis. The altered anatomy of the clivus in Semin Med Genet 2017;175:450–64 CrossRef Medline many patients with CHARGE syndrome may therefore reflect er- 9. Vissers LE, van Ravenswaaij CM, Admiraal R, et al. Mutations in a rors in somitogenesis due to faulty CHD7 signaling. new member of the chromodomain gene family cause CHARGE This study has several limitations. It was based on retrospec- syndrome. Nat Genet 2004;36:955–57 CrossRef Medline 10. Jongmans MC, Hoefsloot LH, van der Donk KP, et al. Familial tive assessment of MR imaging and CT scans obtained at different CHARGE syndrome and the CHD7 gene: a recurrent missense mu- hospitals using different scanning protocols and image parame- tation, intrafamilial recurrence and variability. Am J Med Genet A ters. Nevertheless, a sagittal MR imaging scan or a sagittal CT 2008;146:43–50 Medline reconstruction, on which the clivus could be assessed, was always 11. Morimoto AK, Wiggins RH 3rd, Hudgins PA, et al. Absent semicir- available. However, in 2 cases, the clivus was difficult to assess. An cular canals in CHARGE syndrome: radiologic spectrum of find- ings. AJNR Am J Neuroradiol 2006;27:1663–71 Medline altered anatomy of the remainder of the skull base, also described 12. Yu T, Meiners LC, Danielsen K, et al. Deregulated FGF and homeotic 15 by Natung et al, made the definition of clival borders difficult in gene expression underlies cerebellar vermis hypoplasia in CHARGE several cases. Extreme clival hypoplasia also limited accurate syndrome. Elife 2013;2:e01305 CrossRef Medline measurements. 13. Leboucq N, Menjot de Champfleur N, Menjot de Champfleur S, et al. The olfactory system. Diagn Interv Imaging 2013;94:985–91 CrossRef Medline CONCLUSIONS 14. Fujita K, Aida N, Asakura Y, et al. Abnormal basioccipital develop- This is the largest case-control series on clivus abnormalities in ment in CHARGE syndrome. AJNR Am J Neuroradiol 2009;30: CHARGE syndrome, to our knowledge. Although the clinical rel- 629–34 CrossRef Medline evance of clival hypoplasia and platybasia in CHARGE syndrome 15. Natung T, Goyal A, Handique A, et al. Symmetrical chorioretinal is not yet clear, the results of this study confirm the suggestion by colobomata with craniovertebral junction anomalies in CHARGE syndrome: a case report with review of literature. J Clin Imaging Sci Mahdi and Whitehead18 that besides the well-known labyrinthine 2014;4:5 CrossRef Medline 26 anomalies and hypo- or aplasia of the olfactory bulb, clival ab- 16. Hoch MJ, Patel SH, Jethanamest D, et al. Head and neck MRI find- normalities may provide an important extra clue for the diagnosis ings in CHARGE syndrome. AJNR Am J Neuroradiol 2017;38: of CHARGE syndrome in neuroimaging studies. We hypothesize 2357–63 CrossRef Medline that CHD7 may play an important role in the development of 17. Mahdi E, Whitehead MT. Coronal clival cleft in CHARGE syn- drome. Neuroradiol J 2017;30:574–77 CrossRef Medline clival hypoplasia. 18. Mahdi ES, Whitehead MT. Clival malformations in CHARGE syn- drome. AJNR Am J Neuroradiol 2018;39:1153–56 CrossRef Medline ACKNOWLEDGMENTS 19. Chapman PR, Bag AK, Tubbs RS, et al. Practical anatomy of the We thank J. Senior for editing the manuscript and G.J. te Meer- central skull base region. Semin Ultrasound CT MR 2013;34:412–35 man, biostatistician, for his valuable help with the statistical CrossRef Medline 20. Smoker WR. Craniovertebral junction: normal anatomy, craniom- analysis. etry, and congenital anomalies. Radiographics 1994;14:255–77 21. Bergman JE, Janssen N, Hoefsloot LH, et al. CHD7 mutations and Disclosures: Christa M. de Geus—RELATED: Grant: CHARGE Syndrome Foundation CHARGE syndrome: the clinical implications of an expanding phe- pilot grant.* Grant: University Medical Center Groningen GrAGIKO grant. Jorieke notype. J Med Genet 2011;48:334–42 CrossRef Medline E. H. Bergman—RELATED: Grant: Netherlands Organization for Health Research and 22. Bergman JE, Janssen N, van der Sloot AM, et al. A novel classification Development, Comments: grant no. 92003460 to J.E.H. Bergman for a PhD (the PhD system to predict the pathogenic effects of CHD7 missense variants in was obtained between 2007 and 2011).* *Money paid to the institution. CHARGE syndrome. Hum Mutat 2012;33:1251–60 CrossRef Medline 23. Jacobs-McDaniels NL, Albertson RC. Chd7 plays a critical role in REFERENCES controlling left-right symmetry during zebrafish somitogenesis. 1. Janssen N, Bergman JE, Swertz MA, et al. Mutation update on the Dev Dyn 2011;240:2272–80 CrossRef Medline CHD7 gene involved in CHARGE syndrome. Hum Mutat 2012;33: 24. Mu¨ller F, O’Rahilly R. Segmentation in staged human embryos: the 1149–60 CrossRef Medline occipitocervical region revisited. J Anat 2003;203:297–315 CrossRef 2. Hall BD. Choanal atresia and associated multiple anomalies. J Pe- Medline diatr 1979;95:395–98 CrossRef Medline 25. Patten SA, Jacobs-McDaniels NL, Zaouter C, et al. Role of Chd7 in 3. Hittner HM, Hirsch NJ, Kreh GM, et al. Colobomatous microph- zebrafish: a model for CHARGE syndrome. PLoS One 2012;7:e31650 thalmia, heart disease, hearing loss, and mental retardation: a syn- CrossRef Medline drome. J Pediatr Ophthalmol Strabismus 1979;16:122–28 Medline 26. Blustajn J, Kirsch CF, Panigrahy A, et al. Olfactory anomalies in 4. Pagon RA, Graham JM Jr, Zonana J, et al. Coloboma, congenital heart CHARGE syndrome: imaging findings of a potential major diag- disease, and choanal atresia with multiple anomalies: CHARGE asso- nostic criterion. AJNR Am J Neuroradiol 2008;29:1266–69 CrossRef ciation. J Pediatr 1981;99:223–27 CrossRef Medline Medline

1942 de Geus Oct 2018 www.ajnr.org ORIGINAL RESEARCH PEDIATRICS

Multiple Brain Developmental Venous Anomalies as a Marker for Constitutional Mismatch Repair Deficiency Syndrome

X S.I. Shiran, X L. Ben-Sira, X R. Elhasid, X J. Roth, X U. Tabori, X M. Yalon, X S. Constantini, and X R. Dvir

ABSTRACT BACKGROUND AND PURPOSE: Biallelic constitutional mutations in DNA mismatch repair genes cause a distinct syndrome, consti- tutional mismatch repair deficiency syndrome (CMMRD), characterized by cancers from multiple organs, most commonly brain tumors, during childhood. Surveillance protocols include total and brain MR imaging among other modalities to enable early detection of tumors. Brain surveillance scans revealed prominent brain developmental venous anomalies (DVAs) in some patients. DVAs are benign vascular anomalies, and their incidence in the general population is 2.6%–6.4%. Most developmental venous anomalies are asymptomatic and are found incidentally. Our purpose was to assess the prevalence of DVAs in CMMRD patients and describe their phenotype.

MATERIALS AND METHODS: A retrospective descriptive analysis of brain MR imaging studies from 10 patients from 3 families with CMMRD was performed. Analysis included the number of developmental venous anomalies, location, draining vessels, and associated vascular anomalies (ie, cavernomas), with clinical correlation of symptoms and tumors.

RESULTS: All 10 patients had Ն2 developmental venous anomalies, and 2 had, in addition, non-therapy-induced cavernomas. There was no clinically symptomatic intracranial bleeding from developmental venous anomalies. Six patients had malignant brain tumors. The location of brain tumors was not adjacent to the developmental venous anomalies. No new developmental venous anomalies developed during follow-up.

CONCLUSIONS: The occurrence of multiple developmental venous anomalies in all our patients with CMMRD suggests that develop- mental venous anomalies may be a characteristic of this syndrome that has not been previously described. If confirmed, this quantifiable feature can be added to the current scoring system and could result in early implementation of genetic testing and surveillance protocols, which can be life-saving for these patients.

ABBREVIATIONS: CMMRD ϭ constitutional mismatch repair deficiency syndrome; DVA ϭ developmental venous anomaly

onstitutional mismatch repair deficiency syndrome young adulthood as well as nonmalignant features. The most C(CMMRD) is a cancer-predisposition syndrome character- common malignancies are brain tumors (predominantly malig- ized mainly by a high risk for developing cancer in childhood and nant gliomas, though other tumors are reported), lymphoid malignancies (most commonly non-Hodgkins lymphomas and

Received March 18, 2018; accepted after revision June 25. leukemia), and gastrointestinal cancers (Lynch syndrome–associ- 1,2 From the Departments of Radiology (S.I.S., L.B.-S.), Pediatric Hematology/Oncol- ated tumors, especially colorectal cancer). Nonmalignant man- ogy (R.E., R.D.), and Pediatric Neurosurgery (J.R., S.C.), Dana Children’s Hospital, Tel ifestations of the syndrome include benign tumors such as adeno- Aviv Sourasky Medical Center, Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel; Division of Hematology/Oncology (U.T.), Department of Pediatrics, mas and neurofibromas, features of neurofibromatosis type 1, The Hospital for Sick Children, University of Toronto, Toronto, Ontario, Canada; and Department of Pediatric Hematology/Oncology (M.Y.), Sheba Medical Center, predominantly cafe´au lait macules, other hypo- and hyperpig- Tel-Hashomer, Israel. mented skin alterations, pilomatricomas, and other features that Paper previously presented as a poster at: International Symposium on Pediatric Neuro-Oncology, June 28 to July 2, 2014; Singapore. were included in the suggested diagnostic criteria by the European 1 Please address correspondence to Shelly I. Shiran, MD, Department of Radiology, Consortium. Dana Children’s Hospital, Tel Aviv Sourasky Medical Center, Sackler Faculty of Because CMMRD is a cancer-predisposition syndrome with a Medicine, Tel Aviv University, Tel Aviv, Israel, 6 Weizman St, Tel Aviv, 64239 Israel; e-mail: [email protected] very high penetrance and most reported individuals are affected Indicates article with supplemental on-line table. during childhood, surveillance protocols were developed aiming http://dx.doi.org/10.3174/ajnr.A5766 at early detection and interventions of these cancers. MR imaging

AJNR Am J Neuroradiol 39:1943–46 Oct 2018 www.ajnr.org 1943 of the brain, every 6 months, starting at diagnosis or birth, is part On-line Table. There were 6 boys and 4 girls. The age at the earliest of the surveillance protocol recommended for these children by study available per child ranged from 1.0 to 12.0 years (mean, 6.5 the European Consortium Care for CMMR-D3 and the Interna- years). Seven children were treated for CNS tumors (4 glioblas- tional Biallelic Mismatch Repair Deficiency Consortium.2,4 It is toma multiforme, 1 medulloblastoma, 1 gliomatosis cerebri, and important to arrive at a definite diagnosis of CMMRD in a pedi- 1 cord primitive neuroectodermal tumor), 1 child was treated for atric or young adult patient with cancer as early as possible to acute myeloid leukemia (AML), and 2 children were treated for allow the recommended surveillance for the patient and genetic lymphoma. Colon polyposis was present in 6 children, and ade- counseling for family members. To facilitate CMMRD testing nocarcinoma of the colon, in 1 child. DVAs were noted in all in patients with cancer, the European Consortium Care for children, constituting 100% prevalence. Five children had 2 CMMR-D suggested a score based on diagnostic criteria.1 In the DVAs, 1 child had 3 DVAs (Fig 1), 3 children had 5 DVAs, and 1 suggested score, several features of the syndrome are assigned child had 7 DVAs. In total, 35 DVAs were observed; 83% were in points; a patient with a 3-point score or above is referred for the supratentorial brain and 64% drained to peripheral cortical genetic counseling. In Tel Aviv Sourasky medical center, we noted veins. The diameters of the collecting veins ranged from a mini- prominent developmental venous anomalies (DVAs) on routine mum of 0.8 mm to a maximum of 3.0 mm (average, 1.9 mm; brain MR imaging performed as a part of our surveillance proto- median, 1.8 mm; mode, 2.2 mm). The length of the collecting vein col in patients with CMMRD. ranged from a minimum of 5.0 mm to a maximum of 39.0 mm DVA is the most frequently encountered cerebral vascular (average, 24.9 mm; median, 27.0 mm; mode, 30.0 mm). There malformation, with an incidence of 2.6%–6.4% in different stud- were T2 signal changes in the brain parenchyma adjacent to the 5,6 ies. A DVA is characterized by a cluster of venous radicles that DVA in 26% of DVAs (Fig 2). converge into a collecting vein, resulting in the typical caput me- In children with brain tumors, there was no correlation be- 7 dusa appearance of the DVA. The collecting vein crosses a vari- tween the tumor locations and the location of the DVA. Caverno- able length of brain parenchyma to join either the superficial or mas were present in 2 patients. One patient with a brain tumor deep venous system. Two or more DVAs coexisting in separate had a single small cavernoma adjacent to the DVA before therapy; regions of the brain were observed in 7%–16% of described pa- on follow-up studies, after tumor treatment with radiation ther- 8,9 tients with DVAs. Most DVAs are asymptomatic and are found apy, he developed focal bleeding in the known cavernoma and incidentally. However, DVAs have been documented as a rare bleeding in additional therapy-induced multiple cavernomas that 7 cause of cerebrovascular bleeding and ischemic events. There is a were not apparent on the initial study. A second patient had 2 6,7 known association between DVAs and cavernomas. non-therapy-induced cavernomas present but distant from the Our aim was to assess the prevalence of DVAs in children with DVA. For 9 children, follow-up studies were available; the time CMMRD and to describe their phenotype to elucidate possible between first and last available studies ranged from 0.2 to 7.5 distinct features associated with CMMRD. years, and there was no change in the number of DVAs on fol- low-up studies. MATERIALS AND METHODS We performed detailed clinical and imaging analysis of all chil- DISCUSSION dren from 3 different families that are carriers of DNA mismatch In this study, we observed 100% prevalence of multiple DVAs in repair genes, each family with a distinct type of mutation. The first children with CMMRD. This is strikingly different from the prev- family has a PMS2 mutation (c.2458dupA), with 5 affected ho- alence of DVAs in the healthy population, which was reported as mozygous members. The second family has a MSH6 mutation 2.7% in a study based on postmortem evaluation,5 and 6.4% in a [c.2314cϾT (p.Arg772Trp)], with 2 affected homozygous mem- more recent study based on modern MR imaging.6 Go¨kc¸e et al6 bers. The third family has 2 different pathogenic mutations in evaluated 1165 patients with brain MR imaging studies and found MSH6 (c.3984_3987dupGTCA//c.3959_3962delCAAG), with 3 DVAs in 75 patients; of those, 10 had multiple DVAs, which is affected members who are compound heterozygotes. 0.9% of their total evaluated population and 13.3% of the group All children had surveillance brain MR imaging during the of patients who had DVAs. The phenotypic characteristics of the past 10 years. This retrospective study was approved by Tel Aviv DVAs in their study included 73% supratentorial location, 50% Sourasky medical center institutional review board. A retrospec- drainage to peripheral veins, and a collecting vein diameter range tive descriptive analysis of their brain MR imaging studies in- of 1.0–4.3 mm (median, 2.0 mm). These findings are not signif- cluded the number of DVAs, location, draining vessels (periph- icantly different from ours with 83% supratentorial location, 64% eral or central), length and diameter of the collecting vein, drainage to peripheral veins, and a collecting vein diameter range associated parenchymal changes (as reflected by increased T2WI of 0.8–3.0 mm (median, 1.8 mm). We found associated T2 signal signal intensity), and the presence of associated cavernomas. Cor- changes in the parenchyma adjacent to the DVA in 26% of the relation with coexisting brain tumor presence and location was DVAs, which is within the range of those reported in different also performed. Descriptive statistics were performed with SPSS studies: 7.8%,10 11.6%,8 and 28%.11 The phenotypic similarities (IBM, Armonk, New York). of the DVAs in our study group compared with those reported in the general population exclude morphologic unique features of RESULTS this vascular malformation in CMMRD patients. Brain MR imaging studies of 10 children with CMMRD were An increased prevalence of DVAs was reported in association evaluated. The complete patient information is presented in the with extensive head and neck venolymphatic malformations. In a 1944 Shiran Oct 2018 www.ajnr.org gray matter heterotopia, and CMMRD; these findings were not present in our study group. Jones et al18 reported an increased prevalence of DVAs in a population of patients with primary brain tumors compared with the healthy population; in their study, 10.2% of patients with brain tumors had a DVA present com- pared with 5.3% of the control group. Among patients with DVAs, only 4 had Ͼ1 DVA. They did not find any correla- tion between the tumor and DVA loca- tions, which is consistent with our re- sults of no correlation between tumor site and DVA location. In their study, they did not mention whether children FIG 1. Coronal T1-weighted postgadolinium injection images (A and B) demonstrate 3 prominent DVAs (arrows) in a girl who is homozygote for the PMS2 mutation. with CMMRD were included among the patients with brain tumors. Among studies describing patients with CM- MRD, in a French cohort of 31 pa- tients,19 there was mention of only 2 patients with intracranial angiomas; assuming the term “angioma” repre- sents a DVA, this finding is very differ- ent from the prevalence in our study. In the French cohort, the description of intracranial abnormalities is based on patients’ physician reports and not on dedicated evaluation of brain MR imaging by a neuroradiologist. Be- cause a DVA may be considered a nor- mal variant, its mention may be omit- ted from radiology reports; this may explain the discrepancy. Most studies FIG 2. MR imaging study of a boy who is homozygote for the PMS2 mutation and has 2 DVAs describing patients with CMMRD are demonstrates increased T2 signal in the left frontal white matter on axial T2WI (A). This is asso- ciated with a subtle DVA (arrow) depicted in the axial T1WI postgadolinium injection (B). dedicated to the clinical course and genetics; we did not find, to date, an series of 40 patients with facial venous malformations, 20% had additional description of associated intracranial vascular associated DVAs and 12.5% had multiple DVAs.12 In a series of 33 malformations. patients with orbital and periorbital lymphatic or venolymphatic In a child with a tumor, clinical suspicion of CMMRD arises malformations, up to 60% of patients had associated DVAs.13 We when coexistence of several features is observed. These include a did not encounter associated head and neck vascular malforma- consanguineous family, cafe´ au-lait spots (possibly mimicking tions in our study population, and we did not find an association neurofibromatosis type 1), and a family history of malignancies in between CMMRD and vascular malformations in the literature. It young relatives, especially brain tumors, lymphomas, or leuke- has been suggested that DVAs may be part of intracranial mani- mias. Indeed, these and additional clinical manifestations were festations of neurocutaneous disorders because there are case re- recently suggested to serve as criteria for a clinical diagnosis of ports describing an association between DVAs and blue rubber CMMRD.1 Our data suggest that the presence of multiple DVAs bleb nevus syndrome.14,15 Patients with CMMRD have not been in an MR imaging of the brain with an associated brain tumor or reported to have any association with blue rubber bleb nevus syn- other CMMRD-related malignancy can be another clue to the drome, but there is a known phenotypic overlap with neurofibro- diagnosis. This marker can be especially helpful to determine the matosis type 1, most commonly cafe´-au-lait macules.16 However, clinical diagnosis of CMMRD in individuals who are unaffected there are no known reports describing the prevalence of DVAs in by cancer and in whom the genetic tests are inconclusive. Most patients with neurofibromatosis type 1. From our experience (un- interesting, in 2 patients, in whom there was clinical suspicion of published data), the prevalence of DVAs in the neurofibromatosis CMMRD, the lack of DVAs was associated with negative molec- type 1 population is not as high as that seen in this study. Baas et ular evidence of CMMRD (unpublished data). al17 reported an association between corpus callosum agenesis, The main limitation of this study is the small group of patients. AJNR Am J Neuroradiol 39:1943–46 Oct 2018 www.ajnr.org 1945 Further assessment of larger patient cohorts from the interna- 7. Ruız DS, Yilmaz H, Gailloud P. Cerebral developmental venous tional consortia will further delineate the exact prevalence of anomalies: current concepts. Ann Neurol 2009;66:271–83 CrossRef DVAs and the characteristics of DVAs in this patient population. Medline 8. Linscott LL, Leach JL, Zhang B, et al. Brain parenchymal signal However, the presence of multiple DVAs in all patients in this abnormalities associated with developmental venous anomalies study has no parallel description in any other pathology, support- in children and young adults. AJNR Am J Neuroradiol 2014;35: ing the important role of DVAs for the clinical diagnosis of 1600–07 CrossRef Medline CMMRD. 9. Lee C, Pennington MA, Kenney CM. MR evaluation of developmen- tal venous anomalies: medullary venous anatomy of venous angio- CONCLUSIONS mas. AJNR Am J Neuroradiol 1996;17:61–70 Medline 10. Santucci GM, Leach JL, Ying J, et al. Brain parenchymal signal ab- Our cohort had 100% prevalence of multiple CNS DVAs in pa- normalities associated with developmental venous anomalies: de- tients with CMMRD. If confirmed, this quantifiable feature can tailed MR imaging assessment. AJNR Am J Neuroradiol 2008;29: be added to the current scoring system and result in early imple- 1317–23 CrossRef Medline mentation of genetic testing and surveillance protocols, which can 11. San Millan Ruíz D, Delavelle J, Yilmaz H, et al. Parenchymal abnor- be life-saving for these patients and families as well as allowing malities associated with developmental venous anomalies. Neuro- targeted tumor management.20,21 radiology 2007;49:987–95 CrossRef Medline 12. Boukobza M, Enjolras O, Guichard JP, et al. Cerebral developmental venous anomalies associated with head and neck venous malforma- ACKNOWLEDGMENTS tions. AJNR Am J Neuroradiol 1996;17:987–94 Medline We thank the International Biallelic Mismatch Repair Deficiency 13. Bisdorff A, Mulliken JB, Carrico J, et al. Intracranial vascular Consortium for their support in this work. This research was anomalies in patients with periorbital lymphatic and lymphati- further supported by Meagan’s Walk, the CIHR Joint Canada- covenous malformations. AJNR Am J Neuroradiol 2007;28: Israel Health Research Program, and an American Association 335–41 Medline 14. Gabikian P, Clatterbuck RE, Gailloud P, et al. Developmental venous for Cancer Research Stand Up 2 Cancer Catalyst TM Research anomalies and sinus pericranii in the blue rubber-bleb nevus Grant. syndrome: case report. J Neurosurg 2003;99:409–11 CrossRef Medline 15. Chung JI, Alvarez H, Lasjaunias P. Multifocal cerebral venous mal- REFERENCES formations and associated developmental venous anomalies in a case of blue rubber bleb nevus syndrome. Interv Neuroradiol 2003; 1. Wimmer K, Kratz CP, Vasen HFA, et al; EU-Consortium Care for 9:169–76 CrossRef Medline CMMRD (C4CMMRD). Diagnostic criteria for constitutional mis- 16. Wimmer K, Rosenbaum T, Messiaen L. Connections between con- match repair deficiency syndrome: suggestions of the European stitutional mismatch repair deficiency syndrome and neurofibro- consortium ‘Care for CMMRD’ (C4CMMRD). J Med Genet 2014;51: matosis type 1. Clin Genet 2017;91:507–19 CrossRef Medline 355–65 CrossRef Medline 2. Tabori U, Hansford JR, Achatz MI, et al. Clinical management 17. Baas AF, Gabbett M, Rimac M, et al. Agenesis of corpus callosum and and tumor surveillance recommendations of inherited mis- gray matter heterotopia in three patients with constitutional mis- match repair deficiency in childhood. Clin Cancer Res 2017;23: match repair deficiency syndrome. Eur J Hum Genet 2013;21:55–61 e32–37 CrossRef Medline CrossRef Medline 3. Vasen HF, Ghorbanoghli Z, Bourdeaut F, et al; EU-Consortium Care 18. Jones BV, Linscott L, Koberlein G, et al. Increased prevalence of for CMMR-D (C4CMMR-D. Guidelines for surveillance of individ- developmental venous anomalies in children with intracranial neo- uals with constitutional mismatch repair-deficiency proposed by plasms. AJNR Am J Neuroradiol 2015;36:1782–85 CrossRef Medline the European Consortium “Care for CMMR-D” (C4CMMR-D). 19. Lavoine N, Colas C, Muleris M, et al. Constitutional mismatch re- J Med Genet 2014;51:283–93 CrossRef Medline pair deficiency syndrome: clinical description in a French cohort. 4. Bakry D, Aronson M, Durno C, et al. Genetic and clinical determi- J Med Genet 2015;52:770–78 CrossRef Medline nants of constitutional mismatch repair deficiency syndrome: re- 20. Bouffet E, Larouche V, Campbell BB, et al. Immune checkpoint in- port from the constitutional mismatch repair deficiency consor- hibition for hypermutant glioblastoma multiforme resulting from tium. Eur J Cancer 2014;50:987–96 CrossRef Medline germline biallelic mismatch repair deficiency. J Clin Oncol 2016;34: 5. Sarwar M, McCormick WF. Intracerebral venous angioma: case re- 2206–11 CrossRef Medline port and review. Arch Neurol 1978;35:323–25 CrossRef Medline 21. Kline CN, Joseph NM, Grenert JP, et al. Targeted next-generation 6.c Go¨k¸e E, Acu B, Beyhan M, et al. Magnetic resonance imaging find- sequencing of pediatric neuro-oncology patients improves diagno- ings of developmental venous anomalies. Clin Neuroradiol 2014;24: sis, identifies pathogenic germline mutations, and directs targeted 135–43 CrossRef Medline therapy. Neuro Oncol 2017;19:699–709 CrossRef Medline

1946 Shiran Oct 2018 www.ajnr.org ORIGINAL RESEARCH PEDIATRICS

Spinal Imaging Findings of Open Spinal Dysraphisms on Fetal and Postnatal MRI

X U.D. Nagaraj, X K.S. Bierbrauer, X C.B. Stevenson, X J.L. Peiro, X F.Y. Lim, X B. Zhang, and X B.M. Kline-Fath

ABSTRACT BACKGROUND AND PURPOSE: Fetal MRI has become a valuable tool in the evaluation of open spinal dysraphisms making studies comparing prenatal and postnatal MRI findings increasingly important. Our aim was to determine the accuracy of predicting the level of the spinal dysraphic defect of open spinal dysraphisms on fetal MR imaging and to report additional findings observed when comparing fetal and postnatal MR imaging of the spine in this population.

MATERIALS AND METHODS: A single-center retrospective analysis was performed of fetal MRIs with open spinal dysraphisms from 2004 through 2016 with available diagnostic postnatal spine MR imaging. Images were reviewed by 2 board-certified fellowship-trained pediatric neuroradiologists. Corresponding clinical/operative reports were reviewed.

RESULTS: One hundred nineteen fetal MRIs of open spinal dysraphisms were included. The level of the osseous defect between fetal and postnatal MR imaging was concordant in 42.9% (51/119) of cases and was 1 level different in 39% (47/119) of cases. On postnatal MR imaging, type II split cord malformation was seen in 8.4% (10/119) of cases, with only 50% (5/10) of these cases identified prospectively on fetal MR imaging. Syrinx was noted in 3% (4/119) of prenatal studies, all cervical, all confirmed on postnatal MR imaging.

CONCLUSIONS: Fetal MR imaging is accurate in detecting the level of the spinal dysraphic defect, which has an impact on prenatal counseling, neurologic outcomes, and eligibility for fetal surgery. In addition, fetal MR imaging is limited in its ability to detect split cord malformations in patients with open spinal dysraphisms. Although rare, fetal MR imaging has a high specificity for detection of cervical spinal cord syrinx.

ABBREVIATION: OSD ϭ open spinal dysraphism

etal MR imaging has been well-established as a powerful tool for which fetal MR imaging has become an essential part of the work- Fin the prenatal evaluation of the neuroaxis and continues to up, guiding clinical management.3,4 play an increasing role in prenatal diagnosis, management, and Despite the heavy reliance on fetal MR imaging in the evalua- counseling.1,2 The Management of Myelomeningocele (MOMS) tion of open spinal dysraphisms and its use in the selection of trial remains the sentinel work driving more centers across the candidates for fetal surgery, the scientific literature examining country to offer prenatal repair of open spinal dysraphisms (OSDs), MR imaging of the fetal spine is limited, and there are very few studies that compare pre- and postnatal MR imaging findings.5-8 Determining the level of the defect is a key component of the inclusion criteria for fetal surgery and provides valuable informa- 8,9 Received April 1, 2018; accepted after revision June 25. tion for prognosis and prenatal counseling. From the Department of Radiology and Medical Imaging (U.D.N., B.Z., B.M.K.-F.), Our aim was to determine the accuracy of predicting the upper Department of Pediatric Neurosurgery (K.S.B., C.B.S.), Department of Pediatric Sur- level of the spinal dysraphic defect of open spinal dysraphisms on gery (J.L.P., F.Y.L.), and Department of Biostatistics and Epidemiology (B.Z.),Univer- sity of Cincinnati College of Medicine, Cincinnati Children’s Hospital Medical fetal MR imaging and to report additional findings observed when Center; Cincinnati, Ohio. comparing fetal and postnatal MR imaging of the spine. Preliminary results previously presented as an oral presentation at: Annual Meeting of the American Society of Neuroradiology and the Symposium of the ASNR Foundation, June 2–7, 2018; Vancouver, British Columbia, Canada. MATERIALS AND METHODS Please address correspondence to Usha D. Nagaraj, MD, University of Cincinnati Study Design College of Medicine, Department of Radiology and Medical Imaging, Cincinnati Children’s Hospital Medical Center, 3333 Burnet Ave, Cincinnati, OH 45229-3026; This study is a single-center, retrospective chart review. The case e-mail: [email protected]; @CincyKidsRad list was manually compiled from all the fetal MRIs performed at http://dx.doi.org/10.3174/ajnr.A5760 Cincinnati Children’s Hospital Medical Center in Cincinnati,

AJNR Am J Neuroradiol 39:1947–52 Oct 2018 www.ajnr.org 1947 Ohio, between 2004 and 2016. Inclusion criteria were fetuses with Table 1: Summary of cohort diagnostic-quality fetal MRIs for OSD (either myelomeningocele Characteristics Data (N = 119) or myelocele).10 Only fetuses with adequate available postnatal Sex neuroimaging and clinical/neurosurgical follow-up were in- Male 43.7% (52/119) Female 56.3% (67/119) cluded. Criteria for adequate postnatal neuroimaging for this Average GA at fetal MRI (wk) 23.9 Ϯ 3.6 study included a diagnostic-quality MR imaging of the spine Fetal Chiari grade within the first 3 months of life, which was used as the gold stan- III 81.5% (97/119) dard in this study. Determination of diagnostic-quality imaging II 11.8% (14/119) was made at the neuroradiologists’ discretion. The images were I 6.7% (8/119) Clubfoot viewed in the PACS. A chart review was performed to obtain None 76.5% (91/119) relevant clinical data. This study was compliant with the Health Unilateral 5.9% (7/119) Insurance Portability and Accountability Act and was approved Bilateral 17.6% (21/119) by the institutional review board. The requirement for informed In utero vs postnatal repair consent was waived. Open in utero repair 31.9% (38/119) Postnatal repair 68.1% (81/119) Average age at postnatal spine MRI (days) 19.9 Ϯ 20.2 Scanning Parameters Note:—GA indicates gestational age. All fetuses included in our study were scanned prenatally on a 1.5T magnet at Cincinnati Children’s Hospital Medical Center by Additional intracranial findings on fetal MR imaging were re- using a Ingenia 1.5T (Philips Healthcare, Best, the Netherlands) corded, including the degree of posterior fossa hindbrain hernia- or 1.5T Signa HDxt (GE Healthcare, Milwaukee, Wisconsin) sys- tion by Chiari grades 1–3, with grade 1 having a patent fourth tem. The spinal dysraphism protocol included axial, sagittal, and ventricle and cisterna magna, grade 2 having effacement of the coronal T2-single-shot FSE and balanced fast-field echo/FIESTA fourth ventricle with a patent cisterna magna, and grade 3 having images of the neuroaxis at 3- to 4-mm slice thickness with no skip. effacement of both the cisterna magna and the fourth ventri- Although this imaging protocol did not change during the study cle.12,13 Lateral ventricular size, by measuring the transverse period, the TR and TE varied on each scanner and were changed at atrial diameter on an axial or coronal image, and third ventric- times of scanner upgrades to optimize image quality. At least 2 ular size, by measuring the transverse diameter on a coronal stacks of images in each plane were obtained to the radiologists’ image, were also recorded.13-15 The presence or absence of a satisfaction. The smallest FOV possible was used. Postnatal MR clubfoot and, when present, whether it was unilateral or bilat- imaging was performed on 1 of 6 inpatient clinical magnets with eral were also routinely reported.16 evolving imaging protocols across the years. By 2014, T2-single- shot FSE images were largely replaced by T2-FSE images, and by Statistical Analysis 2015, true fast imaging with steady-state precession/FIESTA im- Descriptive analyses were performed to demonstrate the distribu- ages became a routine part of the postnatal spinal dysraphism tion of the imaging findings. Continuous variables were presented protocol. as means Ϯ SDs, and categoric variables were presented as num- ber (percentage). A 2-sample t test or 1-way ANOVA was used to Image Interpretation detect the differences in continuous variables among different All images were reviewed by 2 board-certified radiologists groups. The correlation among categoric variables was assessed by (U.D.N., B.M.K.-F.), both with added qualifications in pediatric the ␹2 or Fisher exact test when appropriate. All analyses were radiology and fellowship training in pediatric neuroradiology. performed using the Statistical Analysis System, Version 9.4 (SAS The readers were blinded to the pre- and postnatal imaging find- Institute, Cary, North Carolina). A P value Ͻ .05 was considered ings and reported fetal sonographic findings at the time of inter- statistically significant. pretation. Differences were resolved by consensus. Multiple imaging parameters were evaluated on fetal MR im- RESULTS aging of the spine. The superior level of the spinal dysraphic defect Description of Cohort was determined on fetal MR imaging by establishing the most A total of 119 fetuses (52 male, 67 female) met the criteria and caudal spinal hyperintense disc space as L5–S1 and the lowest were included in this analysis. The average gestational age at fetal Ϯ horizontal vertebral body as L5 and by counting vertebral bodies MR imaging was 23.9 3.6 weeks in the cohort as a whole. While superior to the highest level of the absence of the posterior ele- 31.9% (38/119) of fetuses underwent open in utero repair of ments at the bone/skin defect.8 On postnatal MR imaging, the OSD, the remaining 68.1% (81/119) underwent postnatal re- defect level was determined superior to inferior by examining the pair. The average age at postnatal spine MR imaging was Ϯ entire spine, with the odontoid being labeled C2 and assuming 7 19.9 20.2 days. These and other descriptors of the cohort are cervical vertebral bodies and 12 thoracic vertebral bodies. The summarized in Table 1. presence of an arachnoid cyst, defined as an intradural extramed- Imaging Findings ullary thin-walled CSF-intensity fluid collection, was docu- mented. The presence or absence of a measurable postoperative Level of the Defect. The upper level of the osseous defect on fetal fluid collection, visible spinal cord syrinx, or type II split cord MR imaging ranged from as high as T10 to as low as S3. Most malformation was also documented.11 patients had lumbar defects, 68.9% (82/119), with the largest per- 1948 Nagaraj Oct 2018 www.ajnr.org size. There was no significant correla- tion between defect level and Chiari grade, third ventricular size, and the presence of a clubfoot (Table 2). Split Cord Malformation and Syrinx. On fetal MR imaging, the possibility of a type II split cord malformation was raised in 13% (16/119) of patients; how- ever, only 31% (5/16) were confirmed on postnatal MR imaging. On postnatal MR imaging, split cord malformation was seen in 8.4% (10/119) of cases, with only of 50% (5/10) of these cases identi- fied prospectively on fetal MR imaging (Fig 3). Syrinx was noted in 3% (4/119) of prenatal studies, all cervical, all con- firmed on postnatal MR imaging; how- ever, fetal MR imaging was performed FIG 1. Distribution of defect levels on fetal MR imaging. after the second trimester in 75% (3/4) of these patients (Fig 4).

Postnatal Spine MR Imaging. Postnatal MR imaging was per- formed in all patients at an average age of 19.9 Ϯ 20.2 days; 31.9% (38/119) of patients had undergone prenatal repair; 65.5% (78/ 119) had undergone postnatal repair; 2.5% (3/119) of patients had spine MR imaging before postnatal repair; 39% (47/119) of postnatal spine MRIs had evidence of spinal cord syrinx, and of these subjects 29.8% (14/47) had undergone prenatal repair, 68% (32/47) had undergone postnatal repair, and 2% (1/47) was im- aged before postnatal repair. Measurable postoperative extraspi- nal fluid collections were seen in 32.8% (38/116) of postnatal spine MRIs, 5.3% (2/38) of these patients had undergone prenatal repair; 3.4% (4/119) of patients had intraspinal arachnoid cysts on postnatal spine MR imaging, 50% (2/4) of whom had under- gone prenatal repair; 36.2% (42/116) of patients did not have any FIG 2. Sagittal balanced fast-field echo/FIESTA image from fetal MR evidence of syrinx, fluid collection, or arachnoid cyst on postnatal imaging at 24 weeks’ 5 days’ gestational age (A) demonstrates findings spine MR imaging, 50% (21/42) of whom had undergone prenatal of a lumbosacral myelomeningocele with the upper level of the spinal repair (Table 3). dysraphic defect beginning at L4. Sagittal T2-FSE image from postna- tal MR imaging of the spine in the same patient at 20 days of age (B) shows postoperative changes from OSD closure, with the upper level DISCUSSION of the defect again beginning at L4 (same level), with L3 having an intact neural arch (arrow). We describe our experience with spine imaging findings of OSD on fetal and postnatal MR imaging and have made several obser- centage being at L4 (23.5%, 28/119);13.4% (16/119) had a defect vations: Concordance within 1 level of the osseous defect was seen at S2 or lower (Fig 1); 42.9% (51/119) had the same defect level in 82% of patients between pre- and postnatal MR imaging. There interpreted on postnatal spinal MR imaging compared with the was a significant correlation between the level of the defect and fetal MR imaging; 39.5% (47/119) were 1-level discrepant; 11.8% lateral ventricular size (p ϭ 0.001), with a higher defect level cor- (14/119) were discrepant at 2 levels; and 5.9% (7/119) were 3-lev- relating with larger lateral ventricular size. Fetal MR imaging was els discrepant (Fig 2). Of the discrepant levels, 42.6% (29/68) were limited in its ability to identify split cord malformation in this interpreted as lower than on the postnatal MR imaging and 57.4% population. Finally, syrinx was noted in only 3% of prenatal stud- (39/68) were higher on postnatal MR imaging; 13.4% (16/119) ies; however, all were cervical and all were confirmed on postnatal of patients were interpreted as having a defect at S2 or S3, which MR imaging. would preclude them from fetal surgery; 62.5% (10/16) of these Existing literature describing the accuracy of prenatal level as- patients were concordant to the exact level on postnatal MR im- signments of the OSD defect, defined as the highest open poste- aging and 25% (4/16) were higher and 12.5% (2/16) were lower. rior vertebral arch, is quite limited. One series described equal There was a significant correlation between the level of the accuracy of fetal sonography and fetal MR imaging in level assign- defect and lateral ventricular size (P ϭ .001) on fetal MR imaging, ments for myelomeningocele compared with postnatal imaging, with a higher defect level correlating with larger lateral ventricular though it was observed that prenatally assigned levels using either AJNR Am J Neuroradiol 39:1947–52 Oct 2018 www.ajnr.org 1949 Table 2: Significance of defect level on fetal MRI Chiari Lateral Ventricular Third Ventricular Clubfoot (Absent, Grade (I–III) Size (mm) Size (mm) Unilateral, Bilateral) Level of spinal dysraphic defect P ϭ .07 P ϭ .001 P ϭ .86 P ϭ .3

identification of the defect level prenatally of potential impor- tance in counseling.20 Despite previous sonographic literature de- scribing no significant relationship between defect level and ven- tricular size, our study illustrates how a higher spinal defect is associated with larger lateral ventricular size on fetal MR imaging in a larger sample size.21 This finding is of potential clinical sig- nificance because larger ventricles on prenatal imaging have been shown to be associated with an increased need for shunting in those undergoing fetal surgery.22 On the contrary, despite previ- ous literature describing a higher incidence of foot deformity with FIG 3. Axial T2-single-shot FSE image through a 23 weeks’ 3 days’ higher lesion levels, we did not observe a significant correlation gestational age fetus (A) with a myelocele demonstrates splitting of between the fetal spinal defect level and the presence of clubfoot the neural placode (arrow), consistent with a type II split cord mal- deformity in our cohort.23 This finding suggests that the patho- formation. Axial balanced fast-field echo/FIESTA image from postna- tal spine MR imaging of the same patient at 4 weeks of age confirms physiology of this disease process is complex and influenced by the prenatal findings of split cord malformation status post repair multiple factors, and further studies may be helpful in improving (arrow). prenatal counseling. Our study also adds the observation that there was no significant correlation between defect level and Chiari grade or third ventricle size. Prenatal diagnosis of split cord malformation, also known as diastematomyelia, has been described in a few isolated cases; how- ever, the diagnostic reliability of fetal MR imaging in a larger population is still unclear.24,25 Our series adds to the literature by demonstrating a relatively limited ability of fetal MR imaging to identify type II split cord malformation. The lack of our ability to consistently identify split cord malformation prospectively on fe- tal MR imaging can likely be explained by the very small size of the FIG 4. Sagittal balanced fast-field echo/FIESTA image from fetal MR fetus being imaged. However, it is less clear why cases in which imaging at 32 weeks’ gestational age (A) shows a spinal cord syrinx in the cervicothoracic spine (arrow). Sagittal balanced fast-field echo/ split cord malformation was questioned prenatally were not con- FIESTA image from postnatal spine MR imaging in the same patient at firmed on postnatal MR imaging. It is possible that clumped nerve 1 month of age (B) shows persistence of the syrinx (arrow). roots from the ventral neural placode or a prominent anterior technique may vary by Ն2 segments from the postnatal assigned median fissure of the spinal cord was mistaken for split cord mal- 26,27 levels in at least 20% of cases.17,18 Our study has similar findings formation on fetal MR imaging. Another possibility is that and adds to the literature our experience in a larger cohort in postoperative changes from OSD closure obscure the underlying which we found that the exact level of concordance between fetal split cord malformation on postnatal MR imaging. While the as- and postnatal MR imaging was only 42.9% (51/119) and was sociation between type II split cord malformation and open spinal within 1 level of concordance in 82% (98/119) of patients. The dysraphisms is known and does not preclude fetal surgery, know- reasons are multifactorial and may, in part, relate to the very small ing that our ability to diagnose this prenatally is limited may be size of the fetus being imaged, making it challenging to resolve helpful in counseling. There are notable clinical implications of millimetric-sized landmarks, in addition to the lack of ability to split cord malformation in this population: Not only do these resolve the disc spaces in the entire spine for accurate vertebral patients often require more extensive surgical untethering early in body counting on fetal MR imaging. In addition, determination life or at the time of primary repair, but they are also at increased of the defect level may be compromised postnatally by the post- risk of tethered cord syndrome and progressive scoliosis 28-31 operative changes, making a dysraphic defect at the surgical mar- postnatally. gin versus scar tissue challenging to differentiate. Also, normal Our study also adds to the literature a description of imaging anatomic variations in vertebral body counting, such as the pres- findings on postnatal spine MR imaging in patients with OSD ence of 6 lumbarized vertebral bodies, were not taken in to ac- after both prenatal and postnatal repair. Most notably, we ob- count on fetal MR imaging. This information is important, par- served that syrinx can be seen in association with open spinal ticularly when selecting candidates for fetal surgery because dysraphism on fetal MR imaging.32 We describe its incidence in a fetuses with a defect lower than S1 do not meet the selection cri- relatively large sample of fetuses with open spinal dysraphism teria per the MOMS trial.19 (3.4%, 4/119) and demonstrate a high specificity of fetal MR im- There is literature describing the correlation of the prenatal aging, particularly in the third trimester, because all fetuses with a defect level with the postnatal neurologic function, making the prenatally diagnosed syrinx in our series had the same findings on 1950 Nagaraj Oct 2018 www.ajnr.org Table 3: Postnatal spine MRI findings Percentage of Percentage of Percentage of Prenatal Repair Postnatal Repair Total (N = 119) Group (n = 38) Group (n = 78) P Value Extraspinal postoperative fluid collection 32.8% (38/116) 5.3% (2/38) 46.2% (36/78) P Ͻ .001 Arachnoid cyst 3.4% (4/119) 5.3% (2/38) 2.6% (2/78) P ϭ .46 Syrinx 39.5% (47/119) 36.8% (14/38) 41% (32/78) P ϭ .67 No syrinx, no fluid collection, and no arachnoid cyst 36.2% (42/116) 55.3% (21/38) 26.9% (21/78) P ϭ .003 postnatal MR imaging. The increased incidence of syrinx on post- Disclosures: Usha D. Nagaraj—UNRELATED: Royalties: Elsevier, Comments: for chap- natal MR imaging (39.5%, 47/119) compared with prenatal MR ters written for a diagnostic imaging textbook. Karin S. Bierbrauer—UNRELATED: Board Membership: Spina Bifida Coalition of Cincinnati. Beth M. Kline-Fath— imaging (3.4%, 4/119) is likely explained by ongoing associated UNRELATED: Payment for Lectures Including Service on Speakers Bureaus: lecture abnormalities of CSF flow dynamics in this population, including to radiologist at Mayo Clinic,* *Money paid to the institution. hindbrain herniation, hydrocephalus, and tethered cord rather 33 than a missed congenital syrinx. Although postoperative fluid REFERENCES collections were more frequently seen in the postnatal repair 1. Glenn OA, Barkovich J. Magnetic resonance imaging of the fetal brain group (46.2%, 36/78) compared with the prenatal repair group and spine: an increasingly important tool in prenatal diagnosis: Part 2. (5.3%, 2/38), this finding can likely be explained by the relative AJNR Am J Neuroradiol 2006;27:1807–14 Medline time of imaging after the operation. We also add our observed 2. Glenn OA, Barkovich AJ. Magnetic resonance imaging of the fetal incidence of intraspinal arachnoid cysts on postnatal spine MR brain and spine: an increasingly important tool in prenatal diagno- sis—Part 1. AJNR Am J Neuroradiol 2006;27:1604–11 Medline imaging in this patient population (3.4%, 4/119), which was seen 3. Adzick NS, Thom EA, Spong CY, et al. A randomized trial of prena- equally in those that underwent prenatal versus postnatal repair. tal versus postnatal repair of myelomeningocele. N Engl J Med 2011; Our study has some limitations. First, this is a retrospective 364:993–1004 CrossRef Medline study, which limits its internal validity. Also, given that this is a 4. Committee on Obstetric Practice Society for Maternal-Fetal Medicine. single-institution study performed within a certain timeframe, its Maternal-fetal surgery for myelomeningocele. 2017;130:164–67 Medline 5. Husler MR, Danzer E, Johnson MP, et al. Prenatal diagnosis and external validity may be limited as well. Along those same lines, postnatal outcome of fetal spinal defects without Arnold-Chiari II this study is also likely subject to some degree of selection bias malformation. Prenat Diagn 2009;29:1050–57. because the data were collected from one of the largest referral 6. Huang YL, Wong AM, Liu HL, et al. Fetal magnetic resonance im- centers in the country for prenatal repair of open spinal dysra- aging of normal spinal cord: evaluating cord visualization and co- phisms. Also, given that imaging studies were acquired during a nus medullaris position by T2-weighted sequences. Biomed J 2014; 37:232–36 CrossRef Medline 12-year period on multiple different clinical magnets with peri- 7. Duczkowska A, Bekiesinska-Figatowska M, Herman-Sucharska I, et odic upgrades, the heterogeneity of scanning parameters may af- al. Magnetic resonance imaging in the evaluation of the fetal spinal fect our results as well. In clinical practice, sonography is often canal contents. Brain Dev 2011;33:10–20 CrossRef Medline used in conjunction with fetal MR imaging to determine the de- 8. Chao TT, Dashe JS, Adams RC, et al. Fetal spine findings on MRI and fect level of an open spinal dysraphism, and in our practice, we associated outcomes in children with open neural tube defects. AJR Am J Roentgenol 2011;197:956–61 CrossRef Medline emphasize the sonographic findings for defect level in fetal coun- 9. Sutton LN. Fetal surgery for neural tube defects. Best Pract Res Clin seling because some believe that sonography can better delineate Obstet Gynaecol 2008;22:175–88 CrossRef Medline the osseous structures of the fetal spine than fetal MR imaging.34 10. Tortori-Donati P, Rossi A, Cama A. Spinal dysraphism: a review of However, given that sonography is highly operator-dependent, neuroradiological features with embryological correlations and which is problematic in clinical practice and in the context of a proposal for a new classification. Neuroradiology 2000;42:471–91 CrossRef Medline retrospective analysis, studies in fetal MR imaging of the spine are 11. Pang D, Dias M, Ahab-Barmada M. Split cord malformation, Part I: becoming increasingly important. a unified theory of embryogenesis for double spinal cord malfor- mations. Neurosurgery 1992;31:451–80 CrossRef Medline CONCLUSIONS 12. Sutton LN, Adzick NS, Bilaniuk LT, et al. Improvement in hindbrain We describe several observations at our institution on fetal and herniation demonstrated by serial fetal magnetic resonance imag- ing following fetal surgery for myelomeningocele. JAMA 1999;282: postnatal MR imaging of the spine in patients with OSD and add 1826–31 Medline to the existing literature several important findings. First, fetal 13. Nagaraj UD, Bierbrauer KS, Zhang B, et al. Hindbrain herniation in MR imaging is accurate in detecting the level of the dysraphic Chiari II malformation on fetal and postnatal MRI. AJNR Am J Neu- defect, which has an impact on prenatal counseling, neurologic roradiol 2017;38:1031–36 CrossRef Medline outcomes, and eligibility for fetal surgery. We also found that a 14. Levine D, Trop I, Mehta TS, et al. MR imaging appearance of fetal cere- bral ventricular morphology. Radiology 2002;223:652–60 CrossRef higher defect level correlated with increased fetal ventricular Medline size. Second, we describe additional findings of associated split 15. Garel C, Alberti C. Coronal measurement of the fetal lateral ventricles: cord malformation and spinal cord syrinx on fetal MR imaging comparison between ultrasonography and magnetic resonance imag- in this population. Fetal MR imaging was limited in its ability ing. Ultrasound Obstet Gynecol 2006;27:23–27 CrossRef Medline to identify split cord malformation in our series. Finally, al- 16. Servaes S, Hernandez A, Gonzalez L, et al. Fetal MRI of clubfoot associated with myelomeningocele. Pediatr Radiol 2010;40:1874– though cervical spinal cord syrinx is uncommon, we found 79. CrossRef Medline fetal MR imaging to be a useful tool for detecting it in patients 17. McEwing RL, Pretorius DH, James HE, et al. Prenatal assignation of with OSD. lesion levels in neural tube defects by using ultrasonography: case AJNR Am J Neuroradiol 39:1947–52 Oct 2018 www.ajnr.org 1951 report and review of the literature. J Neurosurg 2005;102(2 Suppl): 26. Rufener SL, Ibrahim M, Raybaud CA, et al. Congenital spine and 248–51 CrossRef Medline spinal cord malformations: pictorial review. AJR Am J Roentgenol 18. Aaronson OS, Hernanz-Schulman M, Bruner JP, et al. Myelo- 2010;194(3 Suppl):26–37 CrossRef Medline meningocele: prenatal evaluation-comparison between trans- 27. Tomsick TA, Peak E, Wang L. Fluid-signal structures in the cervical abdominal US and MR imaging. Radiology 2003;227:839–43 spinal cord on MRI: anterior median fissure versus central canal. CrossRef Medline AJNR Am J Neuroradiol 2017;38:840–45 CrossRef Medline 19. Adzick NS. Fetal surgery for myelomeningocele: trials and 28. Higashida T, Sasano M, Sato H, et al. Myelomeningocele associated tribulations—Isabella Forshall Lecture. J Pediatr Surg 2012;47: with split cord malformation type I: three case reports. Neurol Med 273–81 CrossRef Medline Chir (Tokyo) 2010;50:426–30 Medline 20. Carreras E, Maroto A, Illescas T, et al. Prenatal ultrasound evalua- 29. Borkar SA, Mahapatra A. Split cord malformations: a two years experi- tion of segmental level of neurological lesion in fetuses with ence at AIIMS. Asian J Neurosurg 2012;7:56–60 CrossRef Medline myelomeningocele: development of a new technique. Ultrasound 30. Ansari S, Nejat F, Yazdani S, et al. Split cord malformation asso- Obstet Gynecol 2016;47:162–67 CrossRef Medline ciated with myelomeningocele. J Neurosurg 2007;107(4 Suppl): 21. Babcook CJ, Drake CM, Goldstein RB. Spinal level of fetal 281–85 CrossRef Medline myelomeningocele: does it influence ventricular size? AJR Am J 31. Iskandar B, Mclaughlin C, Oakes W. Split cord malformations in my- Roentgenol 1997;169:207–10 CrossRef Medline elomeningocele patients. Br J Neurosurg 2000;14:200–03 CrossRef 22. Tulipan N, Wellons J, Thom EA, et al; MOMS Investigators. Prenatal surgery for myelomeningocele and the need for cerebrospinal fluid Medline shunt placement. J Neurosurg Pediatr 2015;16:613–20 CrossRef Medline 32. Bixenmann B, Kline-Fath BM, Bierbrauer KS, et al. Prenatal and 23. Gunay H, Sozbilen MC, Gurbuz Y, et al. Incidence and type of foot postnatal evaluation for syringomyelia in patients with spinal dys- deformities in patients with spina bifida according to level of lesion. raphism. J Neurosurg Pediatr 2014;14:316–21 CrossRef Medline Childs Nerv Syst 2016;32:315–19 CrossRef Medline 33. Levine DN. The pathogenesis of syringomyelia associated with le- 24. Sonigo-Cohen P, Schmit P, Zerah M, et al. Prenatal diagnosis of di- sions at the foramen magnum: a critical review of existing theories astematomyelia. Childs Nerv Syst 2003;19:555–60 CrossRef Medline and proposal of a new hypothesis. J Neurol Sci 2004;220:3–21 25. Kutuk MS, Ozgun MT, Tas M, et al. Prenatal diagnosis of split CrossRef Medline cord malformation by ultrasound and fetal magnetic resonance 34. Coleman BG, Langer JE, Horii SC. The diagnostic features of imaging: case report and review of the literature. Childs Nerv Syst spina bifida: the role of ultrasound. Fetal Diagn Ther 2015;37: 2012;28:2169–72 CrossRef Medline 179–96 CrossRef Medline

1952 Nagaraj Oct 2018 www.ajnr.org ORIGINAL RESEARCH SPINE

Dilated Vein of the Filum Terminale on MRI: A Marker for Deep Lumbar and Sacral Dural and Epidural Arteriovenous Fistulas

X W. Brinjikji, X C.A. Hilditch, X J.M. Morris, X A.A. Dmytriw, X H. Cloft, X V. Mendes Pereira, X G. Lanzino, and X T. Krings

ABSTRACT BACKGROUND AND PURPOSE: Conventional MR imaging can provide important clues regarding the location of a spinal vascular malformation. We hypothesized that a dilated vein of the filum terminale, identified as a curvilinear flow void on T2WI, could be an imaging marker for a lower lumbar (L3–L5) or sacral fistula.

MATERIALS AND METHODS: We retrospectively identified all spinal dural and spinal epidural arteriovenous fistulas from 2 large tertiary referral centers from 2005 to 2018. All patients had a lumbar spinal MR imaging and a conventional spinal angiography. Images were reviewed by 2 neuroradiologists who categorized the level of the arterial feeder to the fistula and the presence or absence of a dilated vein of the filum terminale on T2WI and T1 postcontrast images. We calculated the sensitivity, specificity, positive predictive value, and negative predictive value of the presence of a dilated filum terminale vein for a deep lumbar or sacral fistula.

RESULTS: One hundred sixty-two patients were included. An enlarged filum terminale vein was identified in 39 patients. Sensitivity, specificity, positive predictive value, and negative predictive value of the presence of a dilated filum terminale vein for a deep lumbar or sacral fistula were 86%, 98.3%, 94.9%, and 95.1%, respectively.

CONCLUSIONS: The presence of a dilated vein of the filum terminale can accurately localize a spinal dural arteriovenous fistula/spinal epidural arteriovenous fistula to the lower lumbar or sacral spine in patients being evaluated for such lesions. This finding can be used to facilitate both noninvasive and conventional spinal angiography.

ABBREVIATIONS: SDAVF ϭ spinal dural arteriovenous fistula; SEDAVF ϭ spinal epidural arteriovenous fistula; VFT ϭ vein of the filum terminale

pinal dural arteriovenous fistulas (SDAVFs) and spinal epi- and substantial costs from unnecessary serologic and imaging Sdural arteriovenous fistulas (SEDAVFs) are the 2 most investigations.7 common types of spinal vascular malformations. These lesions Once the imaging findings suggestive of a spinal vascular mal- can result in considerable morbidity from congestive myelop- formation are properly identified, further investigations are re- athy secondary to chronic venous hypertension.1-6 Delay in quired to characterize the angioarchitecture and location of the the identification of spinal vascular malformations has been dominant arterial feeders to the lesion. This is important for plan- shown to result in high rates of irreversible morbidity due to ning surgical and/or endovascular therapies. In many centers, a the natural history of the disease, unnecessary surgeries including spinal CT angiogram or time-resolved large-FOV MR angiogram spinal cord biopsies and laminectomies, costly-yet-ineffective medi- is obtained to get a sense of the location of the arterial feeders so cal interventions including intravenous immunoglobulin therapy, that spinal angiography can be focused on a few levels.8-11 This is important because a complete spinal angiography can be time- consuming and result in high radiation exposure for the patient Received April 13, 2018; accepted after revision July 17. and operator. A complete catheter spinal angiography also re- From the Departments of Radiology (W.B., J.M.M., H.C., G.L.) and Neurosurgery quires high volumes of iodinated contrast material, which can (W.B., G.L.), Mayo Clinic, Rochester, Minnesota; and Joint Department of Medical Imaging (W.B., C.A.H., A.A.D., V.M.P., T.K.), Toronto Western Hospital, Toronto, result in renal toxicity in some patients. Ontario, Canada. While noninvasive spinal angiography can be highly accu- Timo Krings and Giuseppe Lanzino contributed equally as senior supervising rate in identifying the location of the fistula, it has limitations authors for this work. related to low spatial resolution (imaging typically requires a Please address correspondence to Waleed Brinjikji, MD, Mayo Clinic, 200 1st St SW, Rochester, MN 55905; e-mail: [email protected]; @wbrinjikji large FOV) and poor temporal resolution. While it was previ- http://dx.doi.org/10.3174/ajnr.A5784 ously believed that conventional MR imaging findings of spi-

AJNR Am J Neuroradiol 39:1953–56 Oct 2018 www.ajnr.org 1953 Outcomes and Statistical Analysis The primary outcomes of this study were sensitivity, specificity, negative predictive value, positive predictive value, and accuracy of the presence of an enlarged VFT on conventional lum- bar spine MR imaging in identifying lower lumbar and sacral fistulas. Lower lumbar and sacral fistulas were defined as fistulas located at the L3–S5 levels—the levels of the filum termi- nale. All statistical values are reported with their associated 95% confidence intervals. A ␬ statistic measuring inter- observer agreement for the presence of FIG 1. A 63-year-old woman with a sacral dural arteriovenous fistula at S1. A, Sagittal T2-weighted MR a VFT was calculated as well. All statis- imaging shows a dilated vein of the filum terminale (white arrow). B, Contrast-enhanced T1-weighted tical analyses were performed using MR imaging again demonstrates the dilated vein of the filum terminale (white arrow). C, Rotational JMP13.0 (SAS Institute, Cary, North digital subtraction angiogram reformatted in the coronal plane following injection into the right internal iliac artery demonstrates a fistula at S1 (arrowhead), with a dilated vein of the filum terminale. Carolina). nal dural AVF such as the location of cord edema or the epi- RESULTS center of the dilated perimedullary vessels cannot provide Patient Population and Baseline Characteristics information about the location of the fistula,2 there have re- One hundred sixty-two patients were included in this study. cently been a few case reports and smaller case series suggesting Mean patient age was 64.4 Ϯ 12.3 years; 78.4% (127/162) of pa- an association between a dilated vein of the filum terminale tients were men. There were 119 SDAVFs and 43 SEDAVFs. For- 12,13 (VFT) and a deep lumbar or sacral AVF. We thus per- ty-three fistulas (26.5%) were deep lumbar or sacral, and 119 formed a retrospective review of SDAVFs and SEDAVFs from 2 fistulas (73.4%) were upper lumbar or thoracic. Of the 43 deep tertiary referral centers to test the hypothesis that a dilated lumbar or sacral fistulas, 20 fistulas were sacral fistulas and 23 VFT, identified as a curvilinear midline flow void on T2WI or were deep lumbar fistulas. All patients had noncontrast lumbar a dilated contrast-enhancing vessel below the conus on T1- spine MRIs with T2-weighted imaging in the sagittal plane, and weighted postcontrast images, is an imaging marker for a lower 102 patients (63%) had T1 contrast enhanced lumbar spine MR lumbar (L3–L5) or sacral fistula. imaging in the sagittal plane. MATERIALS AND METHODS Patient Selection Dilated Vein of the Filum Terminale and Fistula Location Following institutional review board approval at 2 institutions, we Thirty-nine patients had a dilated VFT. Interobserver agreement retrospectively identified all patients with SDAVFs and SEDAVFs for identifying the presence of a dilated VFT was excellent (␬ ϭ who underwent lumbar spinal MR imaging and conventional spi- 0.84; 95% CI, 0.81–0.87). In all cases, a dilated VFT was present nal angiography from 2005 to 2018. All patients had an angio- on both T2 and T1 contrast enhanced imaging. graphically confirmed SDAVF or SEDAVF. Children and patients The sensitivity of a dilated VFT in identifying a lower lum- who did not consent to the use of their charts for retrospective bar or sacral fistula was 86% (37/43; 95% CI, 72.1%–94.7%). research were excluded. Patients with intramedullary, perimedul- The specificity of a dilated VFT in identifying a lower lumbar lary, and metameric-type vascular malformations were excluded. or sacral fistula was 98.3% (117/119; 95% CI, 94.1%–99.8%). The positive predictive value of a dilated VFT in identifying a Imaging Review lower lumbar or sacral fistula was 94.9% (37/39; 95% CI, Images were reviewed by 2 diagnostic and interventional neuro- radiologists. MR images were reviewed for the presence of a di- 82.3%–98.7%). The negative predictive value of a dilated VFT lated VFT, which was defined as a curvilinear flow void coursing in identifying a lower lumbar or sacral fistula was 95.1% (117/ along the filum terminale on sagittal T2-weighted images and/or a 123; 95% CI, 90.3%–97.6%). The accuracy of a dilated VFT in curvilinear enhancing VFT coursing along the filum on contrast- identifying a lower lumbar or sacral fistula was 95.1% (154/ enhanced T1-weighted images. The neuroradiologists were 162; 95% CI, 90.5%–97.8%). The area under the curve of a blinded to the level and type of fistula at the time of MR imaging dilated VFT in identifying a lower lumbar or sacral fistula was interpretation. Following interpretation of the MR images, the 2 0.921. Ninety-five percent (19/20) of sacral SEDAVFs and neuroradiologists reviewed the conventional spinal angiograms SDAVFs had a dilated VFT, and 78.3% (18/23) of lower lumbar and identified the location of the arterial feeder to the spinal vas- SEDAVFs and SDAVFs had a dilated VFT. Age, sex, and fistula cular malformation as well as the type of spinal vascular malfor- type were not associated with the presence of a dilated VFT. mation (SEDAVF versus SDAVF). Examples of dilated VFTs are provided in Figs 1–3. 1954 Brinjikji Oct 2018 www.ajnr.org DISCUSSION tivity, specificity, accuracy, negative predictive value, and positive Our large retrospective study found that the presence of a dilated predictive value were all high, and interobserver agreement was VFT on conventional T2WI is a reliable indicator for the presence excellent. Ninety-five percent of sacral spinal vascular malforma- of a lower lumbar or sacral spinal vascular malformation. Sensi- tions were characterized by the presence of a dilated VFT, and Ͼ80% of lower lumbar spinal vascular malformations had a di- lated VFT. These findings are important because the identifica- tion of a dilated VFT can help in focusing noninvasive angio- graphic imaging and conventional spinal angiography on the lumbosacral levels, thus allowing more efficient identification of arterial feeders to these lesions. The vascular supply to the filum terminale has been well-de- scribed in both radiographic and cadaveric studies.14,15 The filum terminale has a single artery arising from the termination of the anterior spinal artery, which travels along the anterior aspect of the filum into the sacral canal. The vein of the filum terminale, which is continuous with the anterior spinal vein, travels in front of the filum but behind the artery and is of uniform caliber along its course. There are no veins along the dorsal aspect of the filum. Because the VFT is the only intradural venous structure below the L2 vertebral body level, it is the only longitudinal collecting vein below this level. Thus, all dural and epidural AVFs with intradural venous drainage must drain through the VFT.14,15 The VFT can drain in 2 directions: 1) descending to the sacral venous plexus and hypogastric vein, and 2) ascending toward the anterior spinal vein along the spinal cord. It is thought that the ascending route is generally the preferred route. The propensity for the VFT to drain superiorly could explain the propensity for deep lumbar and sacral fistulas to result in a dilated VFT and the low rate of more superiorly located fistulas presenting with dilated VFTs.14,15 A small number of patients with deep lumbar and sacral fistulas did not have a dilated VFT on MR imaging. Angiographically, these lesions drained into the VFT; however, the VFT was likely not engorged enough to be easily seen on MR imaging. FIG 2. A 75-year-old man with a sacral dural arteriovenous fistula at Prior studies and case reports have shown high rates of VFT S1. A, Sagittal T2-weighted MR imaging shows a dilated vein of the enlargement in the setting of deep lumbar and sacral fistulas. One filum terminale (white arrow). B and C, Contrast-enhanced MRIs dem- onstrate marked dilation of the vein of the filum terminale (white of the first case reports demonstrating such an association was arrow). D, Conventional angiogram in the anteroposterior plane fol- published by Chen and Hsu in 2002.12 The authors reported a lowing injection into the right internal iliac artery demonstrates a sacral dural AVF, which presented with an enlarged VFT on fistula at S1 with a dilated vein of the filum terminale (black arrow). sagittal T2-weighted MR imaging. The presence of such a vein prompted the in- vestigators to pursue pelvic angiography before thoracolumbar angiography, re- sulting in prompt identification of the fistulous point.12 In a more recent series of sacral fistulas by Gioppo et al,16 the authors found that all 15 sacral fistulas had an enlarged VFT on MR imaging. Meanwhile, a recently published series of spinal vascular malformations at the L5 level or below found that 60% of fis- tulas had a dilated filum terminale vein, further reinforcing the association be- FIG 3. An 80-year-old man with a sacral dural arteriovenous fistula at S1. A, Sagittal T2-weighted tween a dilated VFT and the presence of MR imaging shows a dilated vein of the filum terminale (white arrows). B, Contrast-enhanced a deep lumbar or sacral fistula.13 Our T1-weighted MR imaging again demonstrates the dilated vein of the filum terminale (white ar- study differs from these prior studies be- rows). C, Conventional angiogram in the anteroposterior plane following injection into the right internal iliac artery demonstrates a fistula at S1 with a dilated vein of the filum terminale (black cause we are able to confirm that a di- arrows). lated VFT identified using conventional AJNR Am J Neuroradiol 39:1953–56 Oct 2018 www.ajnr.org 1955 MR imaging techniques is a sensitive and specific sign for local- REFERENCES ization of SDAVF/SEDAVFs to the lower lumbar or sacral spine in 1. Krings T, Geibprasert S. Spinal dural arteriovenous fistulas. AJNR patients being evaluated for such lesions. Am J Neuroradiol 2009;30:639–48 CrossRef Medline Numerous studies have demonstrated the utility of various 2. Krings T, Mull M, Gilsbach JM, et al. Spinal vascular malformations. Eur Radiol 2005;15:267–78 CrossRef Medline MR and CT angiographic techniques in identifying the feeding 3. Morris JM. Imaging of dural arteriovenous fistula. Radiol Clin North artery and angioarchitecture of spinal vascular malforma- Am 2012;50:823–39 CrossRef Medline tions.3,17-19 These techniques have proved instrumental in focus- 4. Morris JM, Kaufmann TJ, Campeau NG, et al. Volumetric myelo- ing spinal angiography to make conventional spinal angiography graphic magnetic resonance imaging to localize difficult-to-find more efficient and reduce operative time, radiation exposure, and spinal dural arteriovenous fistulas. J Neurosurg Spine 2011;14:398– 404 CrossRef Medline contrast dose. However, these techniques have their limitations. 5. Brinjikji W, Yin R, Nasr DM, et al. Spinal epidural arteriovenous For gadolinium bolus spinal angiography, some centers lack the fistulas. J Neurointerv Surg 2016 Jan 20. [Epub ahead of print] capability and scanner time to image the entire spine in patients CrossRef Medline with suspected spinal vascular malformations. Thus, such centers 6. Nasr DM, Brinjikji W, Clarke MJ, et al. Clinical presentation and will often perform a gadolinium bolus MRA focused on the tho- treatment outcomes of spinal epidural arteriovenous fistulas. J Neurosurg Spine 2017;26:613–20 CrossRef Medline racic and upper lumbar spine and will miss deep lumbar and 7. Brinjikji W, Nasr DM, Morris JM, et al. Clinical outcomes of patients 20 sacral fistulas on initial evaluation. This can be mitigated by with delayed diagnosis of spinal dural arteriovenous fistulas. AJNR performing 2 gadolinium-bolus MRAs; however, it is preferable Am J Neuroradiol 2016;37:380–86 CrossRef Medline to avoid this situation. In the setting of noninvasive 4D MR im- 8. Condette-Auliac S, Boulin A, Roccatagliata L, et al. MRI and MRA of aging, there is often a trade-off among spatial, temporal, and con- spinal cord arteriovenous shunts. J Magn Reson Imaging 2014;40: 20 1253–66 CrossRef Medline trast resolution. Thus, the ability to narrow the FOV to a smaller 9. Lindenholz A, TerBrugge KG, van Dijk JM, et al. The accuracy and region in the setting of a dilated VFT could allow improved non- utility of contrast-enhanced MR angiography for localization of invasive characterization of spinal vascular malformations.20 Our spinal dural arteriovenous fistulas: the Toronto experience. Eur Ra- study suggests that when a dilated VFT is identified, radiologists diol 2014;24:2885–94 CrossRef Medline should focus their noninvasive angiographic investigations in the 10. Oda S, Utsunomiya D, Hirai T, et al. Comparison of dynamic con- trast-enhanced 3T MR and 64-row multidetector CT angiography lumbosacral region, including branches of the iliac arteries. We for the localization of spinal dural arteriovenous fistulas. AJNR have found that lack of adequate angiographic investigation of the Am J Neuroradiol 2014;35:407–12 CrossRef Medline iliac branches is one of the most common reasons for missing 11. Amarouche M, Hart JL, Siddiqui A, et al. Time-resolved contrast- spinal vascular malformations on false-negative spinal angio- enhanced MR angiography of spinal vascular malformations. AJNR graphic studies.7 Am J Neuroradiol 2015;36:417–22 CrossRef Medline 12. Chen CJ, Hsu HL. Engorged and tortuous intradural filum termi- nale vein as a sign of a sacral dural arteriovenous malformation. Eur Limitations J Radiol 2002;44:152–55 CrossRef Medline Our study has limitations. First, given its retrospective nature, 13. Jablawi F, Nikoubashman O, Schubert GA, et al. Clinical and radio- there is a propensity for selection bias. We included only patients logic characteristics of deep lumbosacral dural arteriovenous fistu- with SDAVFs and SEDAVFs; thus, our results do not apply to las. AJNR Am J Neuroradiol 2018;39:392–98 CrossRef Medline 14. Djindjian M, Ribeiro A, Ortega E, et al. The normal vascularization fistulas of the filum terminale, pial AVFs, and nidus-type AVMs. of the intradural filum terminale in man. Surg Radiol Anat 1988;10: A fistula of the filum terminale will, by definition, have a dilated 201–09 CrossRef Medline VFT.21 Thus, there is the potential for an overreliance on this sign 15. Namba K. Vascular anatomy of the cauda equina and its implica- to cause clinicians to miss such a lesion. However, these lesions are tion on the vascular lesions in the caudal spinal structure. Neurol also characterized by a markedly enlarged anterior spinal artery. Med Chir (Tokyo) 2016;56:310–16 CrossRef Medline 16. Gioppo A, Farago` G, Giannitto C, et al. Sacral dural arteriovenous The patients included in this study were imaged during a 13-year fistulas: a diagnostic and therapeutic challenge—single-centre ex- period using various MR imaging scanners, MR imaging proto- perience of 13 cases and review of the literature. J Neurointerv Surg cols, and various field strengths (1.5T versus 3T). This feature can 2018;10:415–21 CrossRef Medline limit the generalizability of our results. 17. Lai PH, Pan HB, Yang CF, et al. Multi-detector row computed to- mography angiography in diagnosing spinal dural arteriovenous fistula: initial experience. Stroke 2005;36:1562–64 CrossRef Medline CONCLUSIONS 18. Lai PH, Weng MJ, Lee KW, et al. Multidetector CT angiography in Our retrospective study including Ͼ160 patients with SDAVFs diagnosing type I and type IVA spinal vascular malformations. and SEDAVFs and lumbar spine imaging found that the presence AJNR Am J Neuroradiol 2006;27:813–17 Medline of a dilated VFT is sensitive and specific for localization of 19. Mathur S, Symons SP, Huynh TJ, et al. First-pass contrast-enhanced MRA for pretherapeutic diagnosis of spinal epidural arteriovenous SEDAVF/SDAVFs to the lower lumbar or sacral spine in patients fistulas with intradural venous reflux. AJNR Am J Neuroradiol 2017; being evaluated for such lesions. Our findings suggest that the 38:195–99 CrossRef Medline presence of an enlarged VFT should prompt the neuroangiogra- 20. Mathur S, Bharatha A, Huynh TJ, et al. Comparison of time-resolved pher to initially catheterize the lower lumbar and internal iliac and first-pass contrast-enhanced MR angiography in pretherapeu- vessels when attempting to identify the location and arterial feed- tic evaluation of spinal dural arteriovenous fistulas. AJNR Am J Neuroradiol 2017;38:206–12 CrossRef Medline ers of a SEDAVF/SDAVF. Our results are also valuable in plan- 21. Giordan E, Brinjikji W, Ciceri E, et al. Arteriovenous fistulae of ning noninvasive spinal angiography by allowing imagers to nar- the filum terminale. J Neurointerv Surg 2018;10:191–97 CrossRef row their FOVs or center on the lumbosacral region. Medline

1956 Brinjikji Oct 2018 www.ajnr.org ORIGINAL RESEARCH SPINE

Spinal Instrumentation Rescue with Cement Augmentation

X A. Cianfoni, X M. Giamundo, X M. Pileggi, X K. Huscher, X M. Shapiro, X M. Isalberti, X D. Kuhlen, and X P. Scarone

ABSTRACT BACKGROUND AND PURPOSE: Altered biomechanics or bone fragility or both contribute to spine instrumentation failure. Although revision surgery is frequently required, minimally invasive alternatives may be feasible. We report the largest to-date series of percuta- neous fluoroscopically guided vertebral cement augmentation procedures to address feasibility, safety, results and a variety of spinal instrumentation failure conditions.

MATERIALS AND METHODS: A consecutive series of 31 fluoroscopically guided vertebral augmentation procedures in 29 patients were performed to address screw loosening (42 screws), cage subsidence (7 cages), and fracture within (12 cases) or adjacent to (11 cases) the instru- mented segment. Instrumentation failure was deemed clinically relevant when resulting in pain or jeopardizing spinal biomechanical stability. The main study end point was the rate of revision surgery avoidance; feasibility and safety were assessed by prospective recording of periprocedural technical and clinical complications; and clinical effect was measured at 1 month with the Patient Global Impression of Change score.

RESULTS: All except 1 procedure was technically feasible. No periprocedural complications occurred. Clinical and radiologic follow-up was available in 28 patients (median, 16 months) and 30 procedures. Revision surgery was avoided in 23/28 (82%) patients, and a global clinical benefit (Patient Global Impression of Change, 5–7) was reported in 26/30 (87%) cases at 1-month follow-up, while no substantial change (Patient Global Impression of Change, 4) was reported in 3/30 (10%), and worsening status (Patient Global Impression of Change, 3), in 1/30 (3%).

CONCLUSIONS: Our experience supports the feasibility of percutaneous vertebral augmentation in the treatment of several clinically relevant spinal instrumentation failure conditions, with excellent safety and efficacy profiles, both in avoidance of revision surgery and for pain palliation.

ABBREVIATIONS: PGIC ϭ Patient Global Impression of Change; PMMA ϭ polymethylmethacrylate

pinal instrumentation is widely used in the treatment of de- disassembly, bone resorption around the screws, or impaction Sgenerative, traumatic, and neoplastic conditions. Altered bio- fracture adjacent to an implanted cage.1 In addition, vertebral mechanics and/or bone fragility may lead to instrumentation fail- insufficiency fractures can occur within the instrumented seg- ure, bone resorption, or new fractures with consequent instability ment or at adjacent levels (junctional fractures). In many in- 1,2 and recurrent or progressive pain. The most commonly en- stances described above, revision surgery is performed,3 with the countered types of instrumentation failure are implant fracture or potential of further morbidity, increased cost, and reduced pa- tient satisfaction. Re-operation is often an unattractive option in elderly, medically complex, and fragile patients. Received May 21, 2018; accepted after revision July 23. Minimally invasive options would be desirable to address in- From the Departments of Neuroradiology (A.C., M.P., M.I.) and Neurosurgery (M.G., D.K., P.S.), Neurocenter of Southern Switzerland, Ente Ospedaliero Cantonale, Lu- strumentation failure. Vertebral cement augmentation is used in gano, Switzerland; Department of Neurosurgery (K.H.), Hoˆpital du Valais, Sion, Switzerland; Department of Radiology (M.S.), New York University Langone Medi- the treatment of painful osteoporotic and tumor-related com- cal Center, New York, New York; and Department of Neuroradiology (A.C.), pression fractures.4-7 Numerous reports also document the Inselspital, Bern, Switzerland. utility of polymethylmethacrylate (PMMA) to augment pedic- Please address correspondence to Alessandro Cianfoni, PD, MD, Department 8-12 of Neuroradiology, Neurocenter of Southern Switzerland, Ospedale Regionale ular screws at the time of insertion. di Lugano, Via Tesserete 46, 6900 Lugano, Switzerland; Department of Neuro- To date, several small series have described the use of cement radiology, Inselspital, Freiburgstrasse 18, 3010 Bern, Switzerland; e-mail: [email protected] augmentation in implant failure, including junctional fractures Indicates article with supplemental on-line table. and screw loosening, both in osteoporotic13-15 and neoplastic http://dx.doi.org/10.3174/ajnr.A5795 settings.16,17

AJNR Am J Neuroradiol 39:1957–62 Oct 2018 www.ajnr.org 1957 the study was approved by the Comitato Etico Cantonale del Ticino institutional ethics committee. Indications for primary spinal instru- mentation included a variety of traumatic, degenerative, and neoplastic conditions. Indications for cement rescue treatment were the following: clinically relevant screw loosening with bone resorption, cage subsidence, and vertebral fracture within or adjacent to the instrumented segment. Instrumentation failure was deemed clinically relevant if accompa- nied by new or recurrent pain correlated with imaging findings or deemed to be a threat to spinal biomechanical stability. Determination of implant failure, the decision to offer treatment, and the treatment approach were based on case review by the Spine Unit staff of our in- stitution, composed of neurosurgeons, neurologists, neuroradiologists, pain physicians, and physical medicine and rehabilitation physicians. All patients underwent preprocedural noncontrast CT in addition to pre-existing imaging studies. Patients with uncorrectable coagulopathy and local or systemic in- fection were excluded.

Cement-Augmentation Procedure Procedures were performed with the pa- tient under moderate sedation and local anesthesia. A single dose of prophylactic intravenous antibiotic (cefazolin, 2 g) FIG 1. Schematic representation of different access approaches to the instrumented vertebra. A, was administered 1 hour before the pro- Axial CT image of a vertebra instrumented with bilateral pedicular screws. Yellow arrows repre- sent transpedicular access, with a thin, flexible, beveled needle contacting the proximal screw cedure. All procedures were performed shaft, then bending and sliding along the screw shaft. The red arrow represents transpedicular under fluoroscopic guidance in a mono- access targeting the tip of the screw, with a slightly more oblique course than the screw path. The or bi-plane angiosuite. A variety of ap- dashed blue arrow represents extrapedicular access targeting the anterior third of the vertebral body along the midline, crossing the course of the screw at the level of posterior wall with an proaches to the target vertebral body obliquity from lateral to medial. B and C, Volume-rendered CT lateral and posteroanterior views were used (Fig 1) to overcome access of an instrumented spine segment. Red arrows in B and C show transpedicular access parallel to constraints posed by the presence of im- the screw used to augment loose screws, while dashed arrows represent extrapedicular accesses to the vertebral body used to augment vertebral body fractures, coursing lateral to medial to the plants. Vertebroplasty 15-ga bevel-tip screw, traversing the screw course from cranial to caudal (blue dashed arrows) or from caudal to trocars were inserted via a transpedicu- cranial (green dashed arrow), respectively, passing cranial or caudal to the transverse process. lar or extrapedicular approach to reach the desired target location, and high-vis- We report the largest and most encompassing series to date of cosity PMMA cement (VertaPlex HV; Stryker, Kalamazoo, Mich- percutaneous fluoroscopically guided vertebral cement augmen- igan) was injected under real-time fluoroscopic control. Postpro- tation in a variety of clinically significant instrumentation failure cedural target-level CT was performed. Patients were mobilized conditions, including screw loosening, cage subsidence, and ver- immediately following recovery from moderate sedation. Most tebral fractures within or adjacent to the instrumented segment. patients were discharged the same day.

MATERIALS AND METHODS Vertebral Body Access A retrospective analysis of 31 consecutive procedures in 29 pa- tients (male/female, 10:19; mean age, 71.6 years; range, 50–82 Transpedicular Approach. An access trajectory parallel to the years) with instrumentation failure treated by percutaneous ce- screw was used for treatment of screw loosening. An en face view ment augmentation between May 2013 and October 2016 was of the screw was obtained as a “bull’s eye” projection, the fluoro- performed. Informed consent was obtained from all patients, and scopic camera was then angled slightly toward the side of the 1958 Cianfoni Oct 2018 www.ajnr.org dle position was targeted as discussed in the transpedicular section above. For instrumented-level fractures, the ap- proach was to cross the path of the screw at the junction between the pedicle and ver- tebral body, from lateral to medial, and cranial or caudal to the screw, depend- ing on target fracture location (Fig 3).19 In case of access to the vertebral body without pedicular screws, a standard ap- proach was used, with only necessary adjustments to avoid the vertical rods. Injection of high-viscosity PMMA cement was performed under real-time high-resolution fluoroscopic control, predominantly in the lateral view, with intermittent anteroposterior or oblique control views. In case of screw loosening (Fig 2), ce- ment injection was aimed at filling the halo of bone resorption around the screw, to simulate screw oversizing, and reducing or nulling screw micromobility. When- ever possible, the adjacent trabecular net- work was also filled with cement in an ef- fort to achieve a more stable anchoring cast between the screw and the vertebral FIG 2. Cement augmentation of bilateral S1 screw loosening. A–C, Multiplanar preprocedural CT body. In case of vertebral fracture, cement shows circumferential osteolysis around the screws in S1. Frontal (D) and lateral oblique views (E) of S1 screws, with bilateral placement of needles along the screws and the needle tip in bone injection was aimed at filling, with a tra- osteolysis around the screw. F–H, Fluoroscopic and CT MIP images post-cement augmentation, becular interdigitation pattern, the ante- demonstrating optimal filling of the osteolytic area (arrows), acting as screw oversizing, and rior two-thirds of the vertebral body, from potentially reducing hypermobility. I, Follow-up CT 3 months postaugmentation shows stable results in this patient reporting clinical amelioration. superior to inferior endplates on both sides of the midline. In case of cage subsi- desired modified access, and the proximal or midportion of the dence, cement injection was aimed at augmenting the bone-metal screw was set as the target. With use of a 15-cm 15-ga beveled-tip interface and arresting bone compaction (Fig 3). vertebroplasty needle, contact with the midshaft of the screw was obtained and the flexible needle was pushed to slide with the bevel Follow-Up along the distal shaft of the screw, maintaining the needle tip Patients were followed 1 month after the procedure with a clinical position inside the bone resorption halo. Alternatively, the tip of visit and standing plain films; when deemed necessary, CT or MR the screw was set as the fluoroscopic target on the en face view and imaging was performed. Clinical and imaging follow-up was con- aligned with a transpedicular access route a few millimeters away tinued at variable intervals, depending on the clinical situation from the screw. and the referring physician’s preference. For access to instrumented level fractures (rather than loosen- ing), a trajectory crossing from lateral to medial and from cranial Study End Points to caudal relative to pedicle screw was usually chosen. In this case, Study end points were feasibility, safety, and efficacy. Feasibility the fluoroscope was angled from the bull’s eye view of the screw and safety were assessed by review of chart records and peripro- slightly laterally and cranially. Whenever the needle contacted a cedural complications; efficacy was based on the rate of revision screw, appropriately turning the bevel would allow the needle to surgery, while the clinical effect on pain was measured at 1-month slide past the screw and continue forward progress to reach the follow-up. The Patient Global Impression of Change (PGIC) desired target in the vertebral body. 7-point response was scored as follows: 1, extremely worse; 2, Extrapedicular Access. In general, we thought that small-caliber, much worse; 3, a little worse; 4, no change; 5, a little better; 6, 20 straight anteroposterior, and steep craniocaudal orientation of much better; and 7, extremely better. mid and upper thoracic pedicles favored an extrapedicular ap- proach between the pedicle and the rib head.18 The access trajec- RESULTS tory could be slightly lateral and cranial to the screw and then The On-line Table summarizes the results. parallel to the screw, exploiting the bevel design and flexibility of In 31 procedures, performed in 29 patients (2 patients under- the access needle. For loose screw indications, a similar final nee- went 2 procedures), 42 loose screws, 7 levels with cage subsidence, AJNR Am J Neuroradiol 39:1957–62 Oct 2018 www.ajnr.org 1959 major complications21 occurred; specif- ically, there were no neurologic compli- cations or clinically significant PMMA leaks. Clinical and radiologic follow-up at 1 month was available in 28 patients; extended follow-up ranged from 2 to 54 months (median, 16 months; interquar- tile range, 14.7 months). Specifically, fol- low-up was available at 1 month in 28 patients, at 3 months in 26, at 6 months in 24, and beyond 12 months in 18. One patient was lost to follow-up. During the follow-up period, 5/28 (18%) patients required revision sur- gery: The patient whose procedure was aborted due to access failure underwent revision surgery with replacement of a loose screw; another patient developed re-fracture and kyphosis despite cement augmentation of a junctional fracture requiring extension of instrumentation. Three patients underwent implant re- moval. In 1 of these 3 patients, instru- mentation failure was ultimately due to a low-grade chronic infection. In the 28 patients (30 procedures) with available follow-up, global clinical benefit (PGIC 5–7) was reported in FIG 3. Cage subsidence/fracture and multiple targets. A and B, CT images of an L1 fracture 26/30 (87%) cases at 1 month after the treated with corpectomy, cage grafting, and T11–L3 posterior stabilization in a patient with os- procedure (PGIC 7 in 10/26, PGIC 6 in teoporosis. Due to bone compaction/fracture cranial and caudal to the cage, there is cage 9/26, PGIC 5 in 7/26), while no signifi- subsidence and focal kyphosis (arrows in B). Another fracture is noted at T11 (arrowhead in B), and there is bone resorption and screw loosening at L3 (not shown), with initial screw pullout. C and cant changes in status (PGIC 4) were D, Anteroposterior and lateral fluoroscopy views after placement of multiple needles to perform reported in 3/30 (10%) and worsening cement augmentation at the cranial and caudal bone-metal cage interface in T12 and L2 (arrow- status (PGIC 3) occurred in 1/30 (3%) heads), in the T11 fracture (arrow), in L3 (arrow) to augment the screw osseous purchase, and in T10 to perform prophylactic augmentation (arrow). E, Postprocedural sagittal MIP CT image demon- in the absence of obvious procedural strates satisfactory cement filling of the target levels. F, Standing plain film at 12-month follow-up, complications. with stable results.

12 vertebral body fractures at levels within the instrumented seg- DISCUSSION ment, and 11 fractures at levels adjacent to the instrumented seg- In this series, minimally invasive fluoroscopically guided percu- ment (junctional fractures) were treated. In addition, prophylac- taneous cement augmentation was associated with successful tic cement augmentation was performed at nonfractured adjacent avoidance of revision surgery in 82% of patients with screw loos- levels in 9 cases with the intent of preventing subsequent junc- ening, cage subsidence, or vertebral fractures within or adjacent to tional fractures in patients with osteoporosis when focal kyphosis the instrumented segment. Although limited by the retrospective was noted at the junctional level. In some patients, during the study design, we believe that implant failure would have otherwise same procedure, distant vertebral or sacral fractures were also led to revision surgery, as discussed in the Spine Unit multidisci- treated with cement augmentation. Prophylactic and distant site plinary review. Technical feasibility was highly satisfactory, with augmentation was not included in analysis. In all 32 thoracic, 24 lumbar, 13 sacral, and 3 pelvic (iliac 30/31 procedures executed with the desired technical results. screws) targets, a total of 72 targets, were treated. Safety was excellent, with no periprocedural complications, and The mean interval between the last spinal instrumentation and the procedure resulted in significant pain amelioration in 87% of the procedure was 14 months (range, 1 week to 11 years; median, cases. 4.5 months). Spinal instrumentation performed for degenerative, trau- All except 1 procedure was technically feasible and successfully matic, or oncologic diseases that may cause deformity or instabil- accomplished (ie, satisfactory fluoroscopic target visibility, needle ity aims at providing stability while osseous fusion develops. Ad- placement, and cement injection). One procedure was aborted equate bone quality is of primary importance in this success. due to access failure caused by implant-related inability to Inadequate fixation and subsequent segmental microinstability visualize fluoroscopic landmarks. No periprocedural minor or may put the implant at risk of ultimate failure.22 A combination of 1960 Cianfoni Oct 2018 www.ajnr.org factors such as initial or subsequent poor bone density/quality, Clinical Results instrumentation-induced stress shielding and subsequent disuse The main clinical objective of this study was to assess the rate of osteopenia, plastic deformation at the bone-metal interface, or revision surgery avoidance. In a retrospective study, it is not pos- high static stress combined with cyclic loading also contribute to sible to rule out that in some circumstances, a conservative ap- instrumentation failure.2 These processes are often accompanied proach would have been sufficient. Pain palliation was also as- by new or recurrent pain, altered biomechanics, and deformity. sessed; however, back pain in this kind of complex patient cohort Implant failure can occur at variable time intervals after instru- is an elusive target: It is indeed rather difficult to definitively at- mentation from weeks to years. To interrupt this vicious spiral tribute new or recurrent symptoms to imaging findings of im- and palliate pain, in most cases, revision surgery with screw re- plant failure. We preferred to use the PGIC scale rather than a placement, cage replacement with the addition of allograft or Visual Analog Scale for pain assessment because, in our opinion, other bony substitutes, or instrumentation extension are usually it yields a more global and general assessment of the patient’s necessary. All these options carry perioperative risks and morbid- perception of treatment effectiveness. Among patients with avail- ity and are sometimes contraindicated by patients’ clinical condi- able follow-up, the clinical benefit of the procedure was reported tions, posing a therapeutic dilemma; in addition, revision surgery in 26/30 cases. The 1 patient in whom the procedure was aborted has an important cost burden.23 While in cases of major instru- due to access failure reported unchanged clinical status and ulti- mentation failure, such as implant fracture or disassembly, open mately underwent revision surgery. Two more patients reporting a PGIC 4 and 1 patient with PGIC 3 required revision surgery, surgery seems absolutely necessary, in other conditions, there while 1 patient reporting initial clinical benefit (PGIC 6) after may be room for a minimally invasive treatment option such as augmentation of 2 loose screws and a junctional fracture had a cement augmentation. This technique, established for the delayed recurrent collapse of the junctional level and underwent treatment of vertebral body compression fractures4-7 and used instrumentation extension. Finally, 1 patient was lost to follow-up to reinforce pedicular screws at the time of insertion,8-12 has and remained unreachable; this patient might have undergone been reported in small series and case reports as a potential revision surgery at another institution. solution in case of implant-related vertebral fractures and In our series of 29 patients with implant failure, it was neces- screw loosening.13-17 sary to treat a total of 72 targets, including loose screws, cage subsidence, and vertebral fractures within or adjacent to the in- Technical Feasibility and Safety strumented segment. Our data suggest that an important cause of Despite widespread use of vertebral augmentation, there is a pau- implant failure is poor bone quality, a systemic problem leading city of reports on this technique applied to conditions of instru- to multilevel breakdown (Fig. 3). We therefore stress the extreme mentation failure. This might be, in part, related to technical hur- importance of aggressive osteoporosis management in this pa- dles posed by the presence of implants, skepticism regarding tient population.24 procedure efficacy, and lack of awareness. The presence of dense Infrequently, bone resorption around implants can be caused metallic structures sometimes obscures the visibility of known by an infectious process. In case of clinical-radiologic suspicion, fluoroscopic landmarks or creates artifacts when CT is used for biopsies and cultures are recommended before proceeding to ce- guidance. Commonly used access routes to the vertebral body, ment augmentation. Despite these measures, 1 patient in this se- namely the transpedicular approach, are often occupied by im- ries, following revision surgery with implant removal, was diag- plants, and the operator is therefore forced to seek other trajecto- nosed with a low-grade infection, despite biopsy performed 2 ries within narrow anatomic windows. Finally, after the opera- weeks before cement augmentation resulting in cultures negative tion, spinal anatomy can be altered by laminectomies and bony for infection. This infection was likely present before cement aug- fusion masses. mentation. Suboptimal sensitivity of spine biopsies for low-grade Both CT and fluoroscopic guidance were described in prior infections remains a challenge in such cases.25,26 reports.13-17 We invariably used fluoroscopic guidance based on careful planning and mandatory preoperative CT. Transpedicular Limitations access was favored in the lumbar and lower thoracic spine, while Major limitations include the retrospective study design and an an extrapedicular approach was preferably adopted in the mid intrinsically subjective definition of clinically relevant instrumen- and upper thoracic spine. A small-caliber 15-ga beveled-tip ver- tation failure. Also recognized is the subjective nature of patient tebroplasty needle was used, allowing steerability and flexibility to self-assessment. Although this is by far the largest series of its kind precisely reach targets. to date, some conditions such as cage subsidence are infrequent and Precise needle positioning, use of high-viscosity cement, and considerably more experience is necessary. Given the frequency of use of high-quality fluoroscopic imaging equipment might have spinal instrumentation surgery, it is likely that percutaneous salvage contributed to the excellent safety profile of the procedures in our options are currently underused; therefore, greater awareness and series. Although not used in this series, intraoperative CT, O-arm prospective investigations are necessary. Multidimensional Surgical Imaging System (Medtronic, Minne- apolis, Minnesota), or conebeam CT guidance could be imple- CONCLUSIONS mented to aid needle insertion.13,17 Moreover, when necessary, This series supports the feasibility of safe, efficacious, minimally injection of cement can be directed by the use of a coaxial curved invasive percutaneous vertebral cement augmentation in the cannula.15 treatment of clinically relevant instrumentation failure, with an AJNR Am J Neuroradiol 39:1957–62 Oct 2018 www.ajnr.org 1961 excellent safety profile and efficacious clinical results in terms of tion of pedicle screw for osteoporotic spinal surgery: a novel tech- pain palliation and avoidance of revision surgery. A larger pro- nique. Spine (Phila Pa 1976) 2008;33:E317–24 CrossRef Medline spective cohort will be necessary to determine optimal candidates 12. Fransen P. Increasing pedicle screw anchoring in the osteoporotic spine by cement injection through the implant: technical note and for this treatment and to provide more generalizable outcome report of three cases. J Neurosurg Spine 2007;7:366–69 CrossRef data. Medline 13. Amoretti N, Bertrand AS, Gallo G, et al. Percutaneous consolidation Disclosures: Dominique Kuhlen—UNRELATED: Board Membership: DePuy Synthes of loosened spine arthrodesis under CT and fluoroscopy guidance Advisory Board.* Pietro Scarone—RELATED: Consulting Fee or Honorarium: Depuy by radiologists: a new useful technique. Eur Radiol 2015 25:1135–39 Synthes, Comments: I am a consultant spine surgeon for this company, from which CrossRef Medline my institution receives fees*. *Money paid to the institution. 14. Yun DJ, Hwang BW, Oh HS, et al. Salvage percutaneous vertebral augmentation using polymethyl methacrylate in patients with REFERENCES failed interbody fusion. World Neurosurg 2016;95:618.e13–618.e20 1. DeWald CJ, Stanley T. Instrumentation-related complications of CrossRef Medline multilevel fusions for adult spinal deformity patients over age 65: 15. Clerk-Lamalice O, Irani Z, Growney M, et al. Parapedicular vertebral surgical considerations and treatment options in patients with augmentation with polymethylmetacrylate for pedicle screw loos- poor bone quality. Spine (Phila Pa 1976) 2006;31(19 Suppl):S144–51 ening. J Neurointerv Surg 2018 Feb 14. [Epub ahead of print] CrossRef CrossRef Medline Medline 2. Pearson HB, Dobbs CJ, Grantham E, et al. Intraoperative biome- 16. Puri AS, Erdem E. Salvage percutaneous vertebral augmentation in chanics of lumbar pedicle screw loosening following successful ar- failed spinal interbody fusions associated with multiple myeloma. throdesis. J Orthop Res. 2017;35:2673–81 CrossRef Medline Spine J 2010;10:e5–10 CrossRef 3. Kang SH, Kim KT, Park SW, et al. A case of pedicle screw loosening 17. Xu R, O’Connor K, Krol G, et al. Cement salvage of instrumenta- treated by modified transpedicular screw augmentation with tion-associated vertebral fractures. AJNR Am J Neuroradiol 2014;35: polymethylmethacrylate. J Korean Neurosurg Soc 2011;49:75–80 2197–201 CrossRef Medline CrossRef Medline 18. Cianfoni A, Boulter DJ, Rumboldt Z, et al. Guidelines to imaging 4. Clark W, Bird P, Gonski P, et al. Safety and efficacy of vertebroplasty landmarks for interventional spine procedures: fluoroscopy and for acute painful osteoporotic fractures (VAPOUR): a multicentre, CT anatomy. Neurographics 2011;1;39–48 CrossRef randomised, double-blind, placebo-controlled trial. Lancet 2016; 19. Cianfoni A, Massari F, Ewing S, et al. Combining percutaneous 388:1408–16 CrossRef Medline pedicular and extrapedicular access for tumor ablation in a tho- 5. Klazen CA, Lohle PN, de Vries J, et al. Vertebroplasty versus conser- racic vertebral body. Interv Neuroradiol 2014;20:603–08 CrossRef vative treatment in acute osteoporotic vertebral compression frac- Medline tures (Vertos II): an open-label randomised trial. Lancet 2010;376: 20. Hurst H, Bolto J. Assessing the clinical significance of change scores 1085–92 CrossRef Medline recorded on subjective outcome measures. J Manipulative Physiol 6. Fourney DR, Schomer DF, Nader R, et al. Percutaneous vertebro- Ther 2004;27:26–35 CrossRef Medline plasty and kyphoplasty for painful vertebral body fractures in can- 21. Radvany MG, Murphy KJ, Millward SF, et al; Technology Assessment cer patients. J Neurosurg 2003;98(1 Suppl):21–30 CrossRef Medline Committee of the Society of Interventional Radiology. Research re- 7. Berenson J, Pflugmacher R, Jarzem P, et al; Cancer Patient Fracture porting standards for percutaneous vertebral augmentation. J Vasc Evaluation (CAFE) Investigators. Balloon kyphoplasty versus non- Interv Radiol 2009;20:1279–86 CrossRef Medline surgical fracture management for treatment of painful vertebral 22. Young PM, Berquist TH, Bancroft LW, et al. Complications of spinal body compression fractures in patients with cancer: a multicentre, randomised controlled trial. Lancet Oncol 2011;12:225–35 CrossRef instrumentation. Radiographics 2007;27:775–89 CrossRef Medline Medline 23. Theologis AA, Miller L, Callahan M, et al. Economic impact of revi- 8. Amendola L, Gasbarrini A, Fosco M, et al. Fenestrated pedicle screws sion surgery for proximal junctional failure after adult spinal de- for cement-augmented purchase in patients with bone softening: a formity surgery: a cost analysis of 57 operations in a 10-year expe- review of 21 cases. J Orthop Traumatol 2011;12:193–99 CrossRef rience at a major deformity center. Spine (Phila Pa 1976) 2016;41: Medline E964–72 CrossRef Medline 9. Sawakami K, Yamazaki A, Ishikawa S, et al. Polymethylmethacrylate 24. Ohtori S, Inoue G, Orita S, et al. Comparison of teriparatide and augmentation of pedicle screws increases the initial fixation in os- bisphosphonate treatment to reduce pedicle screw loosening after teoporotic spine patients. J Spinal Disord Tech 2012;25:E28–35 lumbar spinal fusion surgery in postmenopausal women with os- CrossRef Medline teoporosis from a bone quality perspective. Spine (Phila Pa 1976) 10. Burval DJ, McLain RF, Milks R, et al. Primary pedicle screw augmen- 2013;38:E487–92 CrossRef Medline tation in osteoporotic lumbar vertebrae: biomechanical analysis of 25. Prodi E, Grassi R, Iacobellis F, et al. Imaging in spondylodiskitis. pedicle fixation strength. Spine (Phila Pa 1976) 2007;32:1077–83 Magn Reson Imaging Clin N Am 2016;24:581–600 CrossRef Medline CrossRef Medline 26. Zimmerli W. Clinical practice: vertebral osteomyelitis. N Engl J Med 11. Chang MC, Liu CL, Chen TH. Polymethylmethacrylate augmenta- 2010;362:1022–29 CrossRef Medline

1962 Cianfoni Oct 2018 www.ajnr.org LETTERS

Comment on “Common Origin of Brachiocephalic and Left Common Carotid Arteries: Proposal of New Terminology”

e read with interest the letter “Common Origin of Brachio- tery combined with a variation in the development of the right Wcephalic and Left Common Carotid Arteries: Proposal of subclavian artery, in which the right subclavian artery, instead of New Terminology.”1 The authors describe the variations in the originating from the brachiocephalic trunk, takes its origin from a origins of arch vessels, including the common origin of the bra- bulbous dilation from the proximal descending thoracic aorta. chiocephalic trunk and left common carotid artery, and propose a This is due to the involution of the embryonic right fourth aortic new terminology. arch and persistence of the embryonic proximal right dorsal aorta The most common variation in aortic arch branching seen is and seventh intersegmental artery.5 The bulbous dilation of the the common origin of brachiocephalic trunk and left common embryonic proximal right dorsal aortic component is called the carotid artery commonly called the “bovine arch.” The bovine Kommerell diverticulum. arch (in cattle) bears no resemblance to this branching pattern Finally, the authors propose a new terminology for the com- described in humans.2 The aortic arch in cattle gives rise to a large mon origin: “brachio-bicephalic trunk” for its brevity and ana- trunk, which gives off a bicarotid trunk and both subclavian ar- tomic correctness. The term “bicephalic” in Latin means having 2 teries. The embryologic basis of variations in aortic arch branch- heads.6 Hence, we think that the term “brachio-bicarotid trunk” ing in humans can be explained by the aberrations in the 6 em- may be more appropriate for describing this pattern of aortic arch bryonic aortic arches. The true bovine aortic arch in humans, branching.7 which presents a common trunk further branching into the bilat- eral carotid and subclavian arteries, can be explained by the per- REFERENCES sistence of the embryonic fifth aortic arch along with involution of 1. Tawfik AM, Sobh DM, Batouty NM. Common origin of brachioce- the embryonic fourth aortic arch distal to the left subclavian ar- phalic and left common carotid arteries: proposal of new terminol- 3,4 tery. The common origin of the brachiocephalic trunk and the ogy. AJNR Am J Neuroradiol 2018;39:E86–87 CrossRef Medline left common carotid artery (erroneously called the “false bovine 2. Budras KD, Habel RE. Thoracic cavity. In: Budras KD, Habel RE. aortic arch”) results from the involution of the embryonic fourth Mu¨lling CK, et al. Bovine Anatomy: An Illustrated Text. Hanover: aortic arch between the left common carotid artery and the left Schlutersche; 2003:62–65 3. Layton KF, Kallmes DF, Cloft HJ, et al. Bovine aortic arch variant in subclavian artery with a persistent embryonic fifth aortic arch. humans: clarification of a common misnomer. AJNR Am J Neurora- The second most common variation of aortic arch branching diol 2006;27:1541–42 Medline in our Indian population is a separate origin of the left verte- 4. Naimo PS, Vazquez-Alvarez MC, d’Udekem Y, et al. Double-lumen bral artery from the aortic arch, which can also be described as aortic arch: persistence of the fifth aortic arch. Ann Thorac Surg 2016; a 4-vessel origin from the aortic arch. 101:e155–56 CrossRef Medline 5. Kommerell B. Verlagerung des Osophagus durch eine abnorm ver- The reverse pattern of a common origin of the brachiocephalic laufende Arteria subclavia dextra (arteria lusoria). Fortschr Geb Ront- trunk and left common carotid artery may be seen in patients with genstr Nuklearmed Erganyungsband 1936;54:590–95 a right aortic arch in whom there may be a common origin of left 6. American Heritage Publishing Company. American Heritage Diction- brachiocephalic trunk and right common carotid artery. This is a ary of the English Language. 5th ed. Boston: Houghton Mifflin; 2016 commonly seen association in patients with tetralogy of Fallot 7. Szpinda M. Morphometric study of the brachiobicarotid trunk in human fetuses. Ann Anat 2007;189:569–74 Medline with a right aortic arch. The bicarotid trunk is basically a common origin of the brachiocephalic trunk and left common carotid ar- X R. Rajagopal X S. Sharma Department of Cardiovascular Radiology and Endovascular Interventions All India Institute of Medical Sciences http://dx.doi.org/10.3174/ajnr.A5761 New Delhi, India

AJNR Am J Neuroradiol 39:E105 Oct 2018 www.ajnr.org E105 REPLY: as an aberrant right subclavian artery. Moreover, the current under- standing is that an aberrant right subclavian artery is not necessarily e would like to thank Drs Rajagopal and Sharma for their associated with a Kommerell diverticulum. Winterest in our letter “Common Origin of Brachiocephalic Finally, the prefix “bi-” has, unsurprisingly, more than 1 and Left Common Carotid Arteries: Proposal of New Terminol- meaning.6 In our suggested terminology, the intended meaning is ogy.”1 We are glad they agree that the term “bovine” is a misno- “both sides”; hence, the term brachio-bicephalic trunk refers to a mer and should no longer be used to describe a very common trunk for a brachial artery and 2 arteries for both sides of the head. variation in the normal human anatomy. However, there are a few Moreover, the term is short and is similar to the original brachio- points in their comment that are worth further discussion. cephalic term. First, in their attempt to explain the embryology of the brachio- bicephalic trunk, they included a theory based on the persistence of the fifth embryonic arch and the involution of the fourth arch be- REFERENCES tween the left common carotid and left subclavian arteries. This the- 1. Tawfik AM, Sobh DM, Batouty NM. Common origin of brachioce- ory is not widely accepted on the basis of the rarity of the persistent phalic and left common carotid arteries: proposal of new terminol- fifth arch in contrast to the common occurrence of the brachio-bice- ogy. AJNR Am J Neuroradiol 2018;39:E86–87 CrossRef Medline phalic trunk.2 The embryologic development of the brachio-bice- 2. Gupta SK, Bamforth SD, Anderson RH. How frequent is the fifth arch artery? Cardiol Young 2015;25:628–46 CrossRef Medline phalic trunk was not in the scope of our original letter; however, there 3. Nelson ML, Sparks CD. Unusual aortic arch variation: distal origin of are other simpler and more reasonable theories based on slower common carotid arteries. Clin Anat 2001;14:62–65 CrossRef Medline growth of the ventral aortic roots between the third and fourth arches 4. Moorehead PA, Kim AH, Miller CP, et al. Prevalence of bovine aortic or arrested bifurcation of the aortic sac allowing fusion between the arch configuration in adult patients with and without thoracic aortic brachiocephalic and left common carotid arteries.3,4 pathology. Ann Vasc Surg 2016;30:132–37 CrossRef Medline 5. Popieluszko P, Henry BM, Sanna B, et al. A systematic review and Second, they described the “bi-carotid” trunk as a brachio- meta-analysis of variations in branching patterns of the adult aortic bicephalic trunk in which the right subclavian artery has an aber- arch. J Vasc Surg 2018;68:298–306.e10 CrossRef Medline rant origin from a bulbous dilation from the proximal descending 6. The American Heritage Dictionary of the English Language. 5th ed. thoracic aorta (the Kommerell diverticulum). We would like to Boston: Houghton Mifflin Harcourt; 2018. https://ahdictionary.com/. further explain that the bi-carotid trunk denotes a common trunk for Accessed July 14, 2018 both carotid arteries and separate origins of the subclavian arteries. X A.M. Tawfik Thus, the right subclavian artery may arise in its original location as X D.M. Sobh 5 X N.M. Batouty the first aortic arch branch or arise as the last branch from the aorta Department of Diagnostic and Interventional Radiology Faculty of Medicine, Mansoura University http://dx.doi.org/10.3174/ajnr.A5778 Mansoura, Egypt

E106 Letters Oct 2018 www.ajnr.org LETTERS

Quality-Control Assessment to Improve the Accuracy of Dynamic Contrast-Enhanced MR Imaging Perfusion

e read with much interest the article by Dr Morales and 2-compartment exchange or 2CX model4 can better quantify the Wcolleagues published in the May 2018 issue of American permeability measurement. Journal of Neuroradiology1 in which they showed that T1- It would be very interesting to perform a second segmentation weighted dynamic contrast-enhanced (DCE) MR imaging by a second reader, if possible an inexperienced one, to evaluate might help differentiate atypical hemangiomas from metastatic ver- the inter- and intraobserver variability and to reinforce the find- tebral lesions. ings of the study and support its use in clinical practice. The authors used a dynamic 2-step approach with a visual Besides this feedback, our colleagues’ very interesting work assessment of T1 curves and a modelization of the DCE MRI using could considerably help radiologists before and during diagnosis the extended Tofts 2-compartment pharmacokinetic model, pro- and lead to changes in patient management, reducing biopsies, viding quantitative perfusion parameters such as the plasma vol- additional imaging, and patient anxiety in persons with atypical ume (Vp) and the volume transfer constant (Ktrans). This ap- hemangiomas. proach could have been strengthened by the integration of a quality-control assessment of the DCE based on the individual Disclosures: Julien Savatovsky—UNRELATED: Consultancy: IRIS Institut de Recher- arterial input function (AIF) curves. When an individual AIF is che International Servier; Grants/Grants Pending: ANRS Agence Nationale pour la Recherche Scientifique*; Payment for Lectures Including Service on Speakers near the mean AIF of the population, it might be integrated into Bureaus: Bayer Healthcare, Philips Healthcare, Medtronic, Biogen, Sanofi; Travel/ the simplified visual assessment of the curves. Alternatively, when Accommodations/Meeting Expenses Unrelated to Activities Listed: Philips Health- an individual’s AIF is far from the mean AIF of the population, 2 care, GE Healthcare, Bayer Healthcare. *Money paid to the institution. approaches might be used to provide more accurate quantitative data: the first one using only the pharmacokinetic model approach to analyze the quantitative parameters; the second correcting data with REFERENCES 1. Morales KA, Arevalo-Perez J, Peck KK, et al. Differentiating atypical a B-spline-based model-independent deconvolution to obtain a hemangiomas and metastatic vertebral lesions: the role of T1- 2 renormalized tissular kinetic. This latter approach gathers more ac- weighted dynamic contrast-enhanced MRI. AJNR Am J Neuroradiol curate quantitative parameters. 2018;39:968–73 CrossRef Medline The authors used the Ktrans to demonstrate preservation of 2. Jerosch-Herold M, Swingen C, Seethamraju RT. Myocardial blood permeability in cases of atypical hemangiomas with an abnor- flow quantification with MRI by model-independent deconvolu- tion. Med Phys 2002;29:886–97 CrossRef Medline mally elevated Vp. This approach should be used cautiously be- 3. Sourbron SP, Buckley DL. Classic models for dynamic contrast-en- trans cause the K method of calculation includes both perfusion hanced MRI. NMR Biomed 2013;26:1004–27 CrossRef Medline and permeability-related phenomena, leading to a possible mis- 4. Brix G, Zwick S, Kiessling F, et al. Pharmacokinetic analysis of tissue interpretation of the parameters, even when data are accurately microcirculation using nested models: multimodel inference and pa- fitted.3 The use of the extravascular and extracellular volume frac- rameter identifiability. Med Phys 2009;36:2923–33 CrossRef Medline tion (Ve) provided by the extended Tofts model, or, even better, X A. Lecler the use of a more comprehensive and complex model such as the X J.C. Sadik X J. Savatovsky Department of Radiology Fondation Ophtalmologique Adolphe de Rothschild http://dx.doi.org/10.3174/ajnr.A5787 Paris, France

AJNR Am J Neuroradiol 39:E107 Oct 2018 www.ajnr.org E107 REPLY: that the commentators use their suggestions on their own spine DCE data, and I look forward to their publications in this area along with a appreciate the comments in the letter to the Editor and the discussion of histopathology of hemangiomas. Iinterest in our publication. The parameters used in our project best fit our dynamic contrast-enhanced (DCE) model with the FDA- X S. Karimi approved software. I find the comments interesting and truly hope Department of Radiology Memorial Sloan-Kettering Cancer Center http://dx.doi.org/10.3174/ajnr.A5826 New York, New York

E108 Letters Oct 2018 www.ajnr.org LETTERS

Vertebroplasty: Expectation or Evidence-Based Interventional Radiology?

he review of the latest data concerning vertebroplasty and an integral part of the medical curriculum. Editors must restrain Tkyphoplasty for osteoporotic vertebral fractures seemed a researchers with preliminary observations from “sciensational- desperate attempt to cleanse the unconscionable.1 ism” (sensationalism in science).8 Terms such as “therapeutic rev- The first vertebroplasty case report appeared in an obscure olution” should be unacceptable.9 Professional societies must low–Impact Factor (0.8) journal, from a team who subsequently scrutinize their recommendations and grade them according to published dozens of articles but never a trial.2 Since 1987, al- the evidence. though thousands of articles have been published, it is exceptional to find an adequately designed trial (eg, blinded, compared Disclosures: Susan Bewley—UNRELATED: Chair of HealthWatch, a charity that stands for with a sham procedure, adequate number of patients, large better understanding by the public and the media of the importance of application of evidence from robust clinical trials (https://www.healthwatch-uk.org/). patient cohorts, long clinical follow-up using relevant outcomes such as patients’ quality of life). In 2016, the Vertebroplasty for REFERENCES Acute Painful Osteoporotic Fractures (VAPOUR) trial showed 1. Chandra RV, Maingard J, Asadi H, et al. Vertebroplasty and kyphop- improved pain relief but with an absurdly short 14-day follow- lasty for osteoporotic vertebral fractures: what are the latest data? 3 up. The most recent trial on vertebroplasty showed robust evi- AJNR Am J Neuroradiol 2018;39:798–806 CrossRef Medline dence that it did not result in benefit for patients with acute os- 2. Galibert P, Deramond H, Rosat P, et al. Preliminary note on the treat- teoporotic vertebral compression fractures.4 ment of vertebral angioma by percutaneous acrylic vertebroplasty Sadly, health care systems are resistant to learning from error. [in French]. Neurochirurgie 1987;33:166–68 Medline 3. Clark W, Bird P, Gonski P, et al. Safety and efficacy of vertebroplasty The 2009 Angioplasty and Stenting for Renal Artery Lesions for acute painful osteoporotic fractures (VAPOUR): a multicentre, (ASTRAL) trial showed that dilation and stent placement in renal randomised, double-blind, placebo controlled trial. Lancet 2016;388: arteries provided no benefit over drug treatments but only in- 1408–16 CrossRef Medline creased serious harm. However, the procedure had spread like 4. Firanescu CE, de Vries J, Lodder P, et al. Vertebroplasty versus sham wildfire since the 1980s with 45,000 procedures annually in the procedure for painful acute osteoporotic vertebral compression 5 fractures (VERTOS IV): randomised sham controlled clinical trial. United States. Surgeons and radiologists are not alone because BMJ 2018;361:k1551 CrossRef Medline the issue also concerns medicines. Encainide and flecainide have 5. Wheatley K, Ives N, Gray R, et al; ASTRAL Investigators. Revascular- long been the standard of care after myocardial infarction to sup- ization versus medical therapy for renal-artery stenosis. N Engl J Med press ventricular premature complexes, until the Cardiac Ar- 2009;361:1953–62 CrossRef Medline 6. Greene HL, Roden DM, Katz RJ, et al. The Cardiac Arrhythmia Sup- rhythmia Suppression Trial (CAST) investigators showed it in- ខខខ 6 pression Trial: first CAST then CAST-II. J Am Coll Cardiol 1992; creased mortality. Presently, most approvals of anticancer 19:894–98 CrossRef Medline medicines are based on flimsy or untested surrogate end points, 7. Kim C, Prasad V. Strength of validation for surrogate end points used and most drugs offer marginal benefits that may be lost in the real in the US Food and Drug Administration’s approval of oncology world of heterogeneous patients in whom only harms appear.7 drugs. Mayo Clin Proc 2016 May 10. [Epub ahead of print] CrossRef The solution is a challenge. In 1660, Pascal warned, “People al- Medline 8. Braillon A. Sciensationalism. Am J Med 2011;124:e13 CrossRef most invariably arrive at their beliefs not on the basis of proof but on Medline the basis of what they find attractive” (The Art of Persuasion). 9. Braillon A. Direct-acting antiviral drugs and hepatitis C virus: a ther- Pragmatically, Chandra et al1 should have pledged that all new apeutic revolution? Cancer 2015;121:4268–69 CrossRef Medline patients be included in randomized controlled trials and, if that X A. Braillon were not possible, in registries for monitoring adverse effects. University Hospital At the system level, training in history and humility should be Amiens, France X S. Bewley Women’s Health Academic Centre King’s College London http://dx.doi.org/10.3174/ajnr.A5762 London, UK

AJNR Am J Neuroradiol 39:E109 Oct 2018 www.ajnr.org E109 REPLY: 2. Clark W, Bird P, Gonski P, et al. Safety and efficacy of vertebroplasty for acute painful osteoporotic fractures (VAPOUR): a multicentre, e thank Braillon and Bewley for their interest in our article, randomised, double-blind, placebo-controlled trial. Lancet 2016; 388:1408–16 CrossRef Medline W“Vertebroplasty and Kyphoplasty for Osteoporotic Verte- 3. Edidin AA, Ong KL, Lau E, et al. Morbidity and mortality after ver- bral Fractures: What Are the Latest Data?” The stated aim of this tebral fractures: comparison of vertebral augmentation and non- article was to provide an update to clinicians on the evolution of operative management in the Medicare population. Spine 2015;40: evidence for reduction in pain and disability from vertebroplasty 1228–41 CrossRef Medline and kyphoplasty for osteoporotic vertebral fractures and, in particu- 4. Kanis JA, Oden A, Johnell O, et al. Excess mortality after hospi- talisation for vertebral fracture. Osteoporos Int 2004;15:108–12 lar, highlight the limitations of the various prospective randomized CrossRef Medline controlled trials (RCTs). The most recent trial, A Randomised Sham- 5. Lange A, Kasperk C, Alvares L, et al. Survival and cost comparison of Controlled Trial of Vertebroplasty for Painful Acute Osteoporotic kyphoplasty and percutaneous vertebroplasty using German Vertebral Fractures (VERTOS IV)1 had not been published at the claims data. Spine 2014;39:318–26 CrossRef Medline time this review was conducted. Notably, there are important differ- 6. Lau E, Ong K, Kurtz S, et al. Mortality following the diagnosis of a vertebral compression fracture in the Medicare population. J Bone ences in the enrolled patient cohorts between A Controlled Trial of Joint Surg Am 2008;90:1479–86 CrossRef Medline Vertebroplasty for Acute Painful Osteoporotic Fractures (VAPOUR) 7. Lin JH, Chien LN, Tsai WL, et al. Early vertebroplasty associated and VERTOS IV, which again highlight the challenge in the interpre- with a lower risk of mortality and respiratory failure in aged pa- tation of outcomes from these procedures.2 tients with painful vertebral compression fractures: a population- Apart from the RCTs, there are also strong signals of benefit based cohort study in Taiwan. Spine J 2017;17:1310–18 CrossRef Medline from large national or insurance-based-claims datasets from Ger- 8. Ong KL, Beall DP, Frohbergh M, et al. Were VCF patients at 3-9 many, Sweden, France, Taiwan, and the United States. In one higher risk of mortality following the 2009 publication of the of the largest analyses of more than 2 million patients during 10 vertebroplasty “sham” trials? Osteoporos Int 2018;29:375–83 years from the US Medicare dataset, there was a strong signal of CrossRef Medline reduced mortality after vertebral augmentation compared with 9. Zampini JM, White AP, McGuire KJ. Comparison of 5766 vertebral 8 compression fractures treated with or without kyphoplasty. Clin medical management. This signal of survival benefit has been Orthop Relat Res 2010;468:1773–80 CrossRef Medline 5 7 replicated in further analysis of German and Taiwanese health 10. Barr JD, Jensen ME, Hirsch JA, et al; Society of Interventional insurance datasets. In addition, various national and medical societ- Radiology, American Association of Neurological Surgeons, Con- ies have varied in their interpretation of the evidence, depending on gress of Neurological Surgeons, American College of Radiology, when they examined the literature.10-12 Most notably, the National American Society of Neuroradiology, American Society of Spine Radiology, Canadian Interventional Radiology Association, Soci- Institute for Health and Care Excellence, which provides evidence- ety of Neurointerventional Surgery. Position statement on percu- based guidance and advice to the National Health Service in the taneous vertebral augmentation: a consensus statement devel- United Kingdom, recommends vertebroplasty and kyphoplasty as oped by the Society of Interventional Radiology (SIR), American treatment options for patients with severe pain after a recent osteo- Association of Neurological Surgeons (AANS) and the Congress porotic vertebral compression fracture and concluded that it was of Neurological Surgeons (CNS), American College of Radiology 13 (ACR), American Society of Neuroradiology (ASNR), American reasonable to assume that the procedures reduce mortality. Society of Spine Radiology (ASSR), Canadian Interventional Akin to many other areas in medicine, clinicians must inte- Radiology Association (CIRA), and the Society of NeuroInter- grate their clinical expertise with patient values and interpretation ventional Surgery (SNIS). J Vasc Interv Radiol 2014;25:171–81 of the research evidence to provide optimized and meaningful CrossRef Medline care. For years, the results from the various RCTs have shown that 11. Chandra RV, Meyers PM, Hirsch JA, et al; Society of NeuroInterven- tional Surgery. Vertebral augmentation: report of the Standards vertebroplasty and kyphoplasty are best considered for patients and Guidelines Committee of the Society of NeuroInterventional with severe pain and disability and only after rigorous clinical and Surgery. J Neurointerv Surg 2014;6:7–15 CrossRef Medline advanced imaging selection. Moreover, earlier treatment (potentially 12. Tsoumakidou G, Too CW, Koch G, et al. CIRSE guidelines on per- Ͻ3 weeks from fracture onset) may provide the best chance of ben- cutaneous vertebral augmentation. Cardiovasc Intervent Radiol efit. Important questions remain unanswered; for example, what are 2017;40:331–42 CrossRef Medline 13. National Institute for Health and Care Excellence. Technology ap- the implications of the progressive height loss evident in untreated- praisal guidance TA279. Percutaneous vertebroplasty and percutane- versus-cemented levels in VERTOS IV and VAPOUR? Does this pre- ous balloon kyphoplasty for treating osteoporotic vertebral compres- vention of height loss help explain the mortality benefit observed in sion fractures. April 24, 2013; https://www.nice.org.uk/guidance/ almost all claims-based studies? We concur with Braillon and Bewley ta279. Accessed July 14, 2018 that new patients should be included in further RCTs to clarify the X R.V. Chandra Interventional Neuroradiology Unit, Monash Imaging role of these procedures or included in large registries in which data Monash Health can be pooled and additional meaningful conclusions reached. Melbourne, Victoria, Australia Faculty of Medicine, Nursing and Health Sciences Monash University REFERENCES Melbourne, Victoria, Australia 1. Firanescu CE, de Vries J, Lodder P, et al. Vertebroplasty versus sham X J. Maingard procedure for painful acute osteoporotic vertebral compression Interventional Neuroradiology Service, Department of Radiology fractures (VERTOS IV): randomised sham controlled clinical trial. Austin Hospital Melbourne, Victoria, Australia BMJ 2018;361:k1551 CrossRef Medline School of Medicine Faculty of Health, Deakin University http://dx.doi.org/10.3174/ajnr.A5781 Waurn Ponds, Victoria, Australia

E110 Letters Oct 2018 www.ajnr.org X H. Asadi Interventional Neuroradiology Unit, Monash Imaging Monash Health Melbourne, Victoria, Australia Interventional Neuroradiology Service, Department of Radiology Austin Hospital Melbourne, Victoria, Australia School of Medicine Faculty of Health, Deakin University Waurn Ponds, Victoria, Australia X L.-A. Slater Interventional Neuroradiology Unit, Monash Imaging Monash Health Melbourne, Victoria, Australia Faculty of Medicine, Nursing and Health Sciences, Monash University Melbourne, Victoria, Australia X J.A. Hirsch Neuroendovascular Program Massachusetts General Hospital, Harvard Medical School Boston, Massachusetts

AJNR Am J Neuroradiol 39:E110–E111 Oct 2018 www.ajnr.org E111