Note: This copy is for your personal, non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at www.rsna.org/rsnarights. Evaluation of Heavily Calcifi ed Vessels with Coronary CT

n : Comparison of

Iterative and Filtered Back Projection Image Reconstruction1 ORIGINAL RESEARCH

Matthias Renker , BS Purpose: To prospectively compare traditional fi ltered back pro- John W. Nance , Jr , MD jection (FBP) and iterative image reconstruction for the U. Joseph Schoepf , MD evaluation of heavily calcifi ed with coronary com- Terrence X. O’Brien , MD , MS puted (CT) angiography. Peter L. Zwerner , MD Mathias Meyer , BS Materials and The study had institutional review board approval and was J. Matthias Kerl , MD Methods: HIPAA compliant. Written informed consent was obtained Ralf W. Bauer , MD from all patients. Fifty-fi ve consecutive patients (35 men, Christian Fink , MD 20 women; mean age, 58 years 6 12 [standard deviation]) Thomas J. Vogl , MD with Agatston scores of at least 400 underwent coronary CT Thomas Henzler , MD angiography and . Image data were reconstructed with both FBP and iterative reconstruction techniques with corresponding cardiac algorithms. Image noise and subjective image quality were compared. To ob- jectively assess the effect of FBP and iterative reconstruc- tion on blooming artifacts, volumes of circumscribed cal- cifi cations were measured with dedicated volume analysis software. FBP and iterative reconstruction series were in- dependently evaluated for coronary greater than 50%, and their diagnostic accuracy was compared, with cardiac catheterization as the reference standard. Statistical analyses included paired t tests, Kruskal-Wallis analysis of variance, and a modifi ed McNemar test.

Results: Image noise measured signifi cantly lower ( P = .011–.035) with iterative reconstruction instead of FBP. Image qual- ity was rated signifi cantly higher ( P = .031 and .042) with iterative reconstruction series than with FBP. Calcifi ca- 1 From the and Vascular Center, Medical University tion volumes measured signifi cantly lower (P = .019 and of South Carolina, Ashley River Tower, 25 Courtenay Dr, .026) with iterative reconstruction (44.3 mm3 6 64.7 and Charleston, SC 29425-2260 (M.R., J.W.N., U.J.S., T.X.O., 46.2 mm3 6 68.8) than with FBP (54.5 mm3 6 69.5 and P.L.Z., M.M., J.M.K., R.W.B., C.F., T.H.); Department of 56.3 mm3 6 72.5). Iterative reconstruction signifi cantly , Johann Wolfgang Goethe University, Frankfurt, improved some measures of per-segment diagnostic accu- Germany (M.R., J.M.K., R.W.B., T.J.V.); Ralph H. Johnson Veterans Affairs Medical Center, Charleston, SC (T.X.O.); racy of coronary CT angiography for the detection of sig- and Institute of Clinical Radiology and , nifi cant stenosis compared with FBP (accuracy: 95.9% vs University Medical Center Mannheim, Heidelberg University, 91.8%, P = .0001; specifi city: 95.8% vs 91.2%, P = .0001; Mannheim, Germany (M.M., C.F., T.H.). Received December positive predictive value: 76.9% vs 61.1%, P = .0001). 30, 2010; revision requested February 28, 2011; revision received March 18; accepted April 2; fi nal version accepted Conclusion: Iterative reconstruction reduces image noise and bloom- April 20. Supported in part by the Research and Develop- ing artifacts from calcifi cations, leading to improved diag- ment Program of the Department of Veterans Affairs. Address correspondence to U.J.S. (e-mail: schoepf nostic accuracy of coronary CT angiography in patients with @musc.edu ). heavily calcifi ed .

The contents do not represent the views of the Department q RSNA, 2011 of Veterans Affairs or the United States Government.

q RSNA, 2011

390 radiology.rsna.org n Radiology: Volume 260: Number 2—August 2011 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

large body of literature has shown of iterative reconstruction techniques in which is indicative of substantial calcifi ed that coronary computed tomogra- CT applications throughout the body have plaque burden. Exclusion criteria were A phy (CT) angiography can be used since been reported (9–13 ). However, the a history of contrast material reaction to sensitively rule out coronary artery potential of this approach to improve and impaired renal function (creatinine stenosis in patients with chest pain (1 ). the diagnostic accuracy of coronary CT higher than 1.5 mg/dL and/or glomerular Nevertheless, those studies also con- angiography, especially in problematic fi ltration rate lower than 60 mL/min). sistently demonstrate that, despite ad- scenarios (eg, the presence of heavy cal- vances in CT technology, the specifi city cifi cations), has not been explored. Scanning Technique of this test frequently remains limited Accordingly, the aim of our prospec- Calcium scoring and coronary CT angiog- by heavy coronary artery calcifi cations, tive study was to compare traditional raphy were performed with a second- which are the most common reason for FBP with iterative image reconstruction generation dual-source CT scanner false-positive fi ndings. Heavy calcifi ca- for the evaluation of heavily calcifi ed ar- (Somatom Defi nition Flash; Siemens tions cause blooming artifacts, which can teries with coronary CT angiography. Healthcare). Prior to contrast material– lead to overestimation of lesions ( 2 ), enhanced coronary CT angiography, often prompting unnecessary coronary calcium scoring was performed in all Materials and Methods catheterization or myocardial perfusion patients by using a prospective electro- studies (3,4 ). U.J.S. is a consultant for Bayer Health- cardiographically (ECG)-triggered high- The current standard CT image re- care (Berlin, Germany), Medrad (In- pitch spiral acquisition technique (14 ). construction technique is fi ltered back dianola, Pa), and Siemens Healthcare Transverse images were reconstructed projection (FBP). However, FBP has lim- (Forchheim, Germany), whose products with a section thickness of 3 mm and itations vis-à-vis three-dimensional cone- were used in this investigation. The au- 50% overlap. The coronary CT angiog- beam geometry, data completeness, and thors who are not consultants for those raphy technique was chosen individually low photon environments ( 5 ). To some companies had control of inclusion of for each patient depending on heart rate extent, blooming artifacts arising from any data and information that might and/or rhythm and body mass index, heavy calcifi cations are attributable to present a confl ict of interest. with the goal of minimizing radiation these technical shortcomings (6 ). exposure. Scan techniques included tra- Iterative reconstruction techniques Patients ditional retrospective ECG gating with have been proposed for over 3 decades This study was part of a larger investi- default use of ECG-dependent tube cur- to improve CT image quality by reduc- gation evaluating the accuracy of coro- rent modulation, prospective ECG trig- ing quantum noise and artifacts (7 ) but nary CT angiography for detection of gering, and prospective ECG-triggered were used mainly in the context of emis- coronary artery stenosis in symptom- high-pitch spiral acquisitions. Contrast sion tomography (8 ). Only recently has atic patients referred for cardiac cath- enhancement was achieved by injecting computer power evolved enough to en- eterization. The study was approved by 60–90 mL of iodinated contrast mate- able iterative image reconstruction within the institutional review board of the rial (370 mg I/mL iopromide, Ultravist; a clinically acceptable time frame for gen- Medical University of South Carolina eral CT applications. Benefi cial effects and was conducted in accordance with Health Insurance Portability and Ac- Published online before print Advances in Knowledge countability Act regulations. Written in- 10.1148/radiol.11103574 Content code: formed consent was obtained from all n Heavy coronary artery calcifi ca- Radiology 2011; 260:390–399 patients. We prospectively included cor- tions show signifi cantly lower vol- onary CT angiograms of 55 consecutive Abbreviations: umes at coronary CT angiography patients (35 men, 20 women; mean age, ECG = when iterative image reconstruc- FBP = fi ltered back projection 58 years 6 12 [standard deviation]) tion is used compared with tradi- PPV = positive predictive value with an Agatston score of at least 400, tional fi ltered back projection ROI = region of interest (FBP), indicating decreased Author contributions: blooming artifacts from calcifi ed Implication for Patient Care Guarantors of integrity of entire study, M.R., U.J.S., M.M., coronary atherosclerotic plaques. n An increase in the specifi city and T.H.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript n Compared with FBP, iterative PPV of coronary CT angiography drafting or manuscript revision for important intellectual image reconstruction signifi cantly with iterative reconstruction content, all authors; approval of fi nal version of submitted improves accuracy, specifi city, techniques could reduce the manuscript, all authors; literature research, M.R., J.W.N., and positive predictive value number of unnecessary follow-up U.J.S., T.X.O., M.M., J.M.K., R.W.B., C.F., T.H.; clinical stud- (PPV) of coronary CT angiogra- studies performed as a result of ies, M.R., U.J.S., T.X.O., P.L.Z., J.M.K., R.W.B., C.F., T.J.V.; phy for the evaluation of coro- false-positive fi ndings at coronary statistical analysis, M.R., U.J.S., M.M., R.W.B., T.J.V., T.H.; and manuscript editing, all authors nary artery stenosis in patients CT angiography in patients with with heavy vessel calcifi cations. heavily calcifi ed plaques. Potential confl icts of interest are listed at the end of this article.

Radiology: Volume 260: Number 2—August 2011 n radiology.rsna.org 391 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

Bayer Healthcare), injected at 6 mL/sec iteration is performed. Therefore, the and diffuse (20 ) coronary artery calci- through an 18-gauge intravenous ante- iterative loops enable noise reduction fi cations within FBP and iterative recon- cubital by using a dual-syringe while preserving edge information and struction data sets. Calcifi cations were injector (Stellant D; Medrad). Acquisi- low-contrast structures. Iterative and semiautomatically segmented by defi n- tion parameters were as follows: detec- FBP reconstructions were performed ing ROIs around calcifi cations, which tor collimation, 2 3 64 3 0.6 mm; gan- at a section thickness of 0.75 mm and were then automatically propagated to try rotation time, 280 msec; and tube a position increment of 0.4 mm. Cor- the neighboring sections and manually current–time product, 320 mAs per responding vascular and high-spatial- corrected, if necessary. Voxels within rotation. The tube potential used was resolution kernels were applied for FBP these ROIs with attenuation values in 120 kV in patients with a body mass (B26f and B46f, respectively) and itera- the range of contrast material were au- index of 25 kg/m2 or greater, 100 kV tive reconstructions (I26f and I46f, re- tomatically excluded from the segmenta- in patients with a body mass index of spectively). The B46f and I46f kernels, tion, with the goal of only including cal- less than 25 kg/m2 but greater than or which are sharp edge-preserving convolu- cium. To achieve this, we measured the equal to 20 kg/m2 , and 80 kV in pa- tion algorithms, were included because maximum contrast attenuation within tients with a body mass index of less of the previously described benefi cial the ascending and added 20% to than 20 kg/m2 . Images were acquired in effects of these higher-spatial-frequency this measurement to defi ne the mini- the craniocaudal direction from above algorithms on the evaluation of high den- mum attenuation level of voxels to be the origin of the coronary arteries to sity structures, such as heavy coronary included in the volumetric analysis. This below the dome of the diaphragm. Ef- artery calcifi cations and (16–19 ). level was kept constant between the dif- fective radiation dose was derived by ferent reconstruction techniques in in- multiplying the dose-length product by Image Noise, Attenuation, and Subjective dividual patients. a chest-specifi c conversion coeffi cient Image Quality (k = 0.014 mSv/Gy/cm) (15 ). All data sets were transferred to a stand- Coronary CT Angiography versus alone image processing workstation Image Reconstruction (Syngo MMWP VE 36A; Siemens). In All data sets and reconstructions were Images were reconstructed from the each data set, one observer (M.R.) jointly evaluated by two radiologists coronary CT angiography raw data with measured image noise, which was de- (U.J.S., T.H.) for the presence of stenotic both FBP and iterative image reconstruc- fi ned as the standard deviation of the (. 50%) using the tion (Iterative Reconstruction in Image measured attenuation (in Hounsfi eld American Heart Association 15-segment Space [IRIS]; Siemens Healthcare) (Fig 1 ). units) within circular regions of inter- model ( 21 ). Readers were blinded to This iterative reconstruction approach est (ROIs) in the ascending aorta, inter- the reconstruction technique. For le- is based on an initial master FBP recon- ventricular septum, and left ventricular sion detection, readers were provided struction with a very sharp convolution cavity. The size of the ROIs was adapted with a combination of the vascular and algorithm, or kernel, still containing all to account for anatomic differences of high-spatial-resolution algorithm recon- frequencies and, therefore, all informa- our patients; however, between the dif- struction series based on FBP (B26f plus tion of the initial raw data. Subsequent ferent reconstruction approaches, the B46f) and iterative (I26f plus I46f) recon- iterative processing loops are applied ROI size was kept constant within each struction. FBP and iterative reconstruc- to the image volume with the goal of patient. Subjective image quality was tion image series were presented in ran- reducing image noise while preserving independently rated by two radiologists dom order at least 3 weeks apart in the spatial resolution. During each itera- (U.J.S. and T.H., with 10 and 5 years same patient to minimize reader recall. tion, general image properties (eg, edge experience in coronary CT angiography, The readers were permitted to adjust information and contrast-to-noise ratio) respectively). FBP and iterative recon- window and level settings individually are analyzed based on a noise model of structed images were reviewed in ran- for each study. Cardiac catheterization the system, which is directly derived dom order. Images were rated on a fi ve- served as the reference standard for from the raw data. The Gaussian noise point Likert scale according to the se- stenosis detection and was performed model is applied in a regularization step. verity of image noise, quality of contour with the conventional Judkins technique Image noise is estimated from image delineation, and general image impres- on the same day as the coronary CT volume data by locally computing the sion (1 = poor, 2 = fair, 3 = moderate, 4 = angiography. At least two views of the minimum noise variance. The strength good, 5 = excellent). right coronary artery and four views of the regularization controls the effect of the left coronary artery were inter- of the edge-preserving low-pass fi lter in Volumetric Analysis of Coronary Artery preted for greater than 50% stenosis each update of the iteration procedure Calcifi cations by two cardiologists (T.X.O. and P.L.Z., ( 5 ). The result of the regularization step One observer (M.R.) used a threshold- both with more than 15 years experi- is compared with the initial data to gen- based volumetry software tool (Volume ence) in consensus by using the same erate an update image, which is added Analysis, version VE31A; Siemens) to 15-segment American Heart Associa- to the previous data set before the next determine the volumes of fragmented tion model.

392 radiology.rsna.org n Radiology: Volume 260: Number 2—August 2011 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

Table 1 Statistical Analysis Patient Demographics and Scan Parameters Statistical analyses were performed by using dedicated statistical software Parameter Datum (SPSS 12.0; SPSS, Chicago, Ill). The Age (y) 58.2 6 12.0 Shapiro-Wilk W test was used to iden- Male-to-female ratio 35:20 tify normally distributed data. Signifi - Height (cm) 170.4 6 10.3 cance was investigated with x 2 statistics Weight (kg) 92.2 6 20.7 for categorical variables. Continuous vari- Body mass index (kg/m2 ) 31.6 6 5.8 ables are presented as means 6 stan- Heart rate (beats per minute) * 63.1 6 8.0 (53–72) dard deviations and were compared by Agatston score * 710 6 289 (466–2934) using one-on-one comparisons with ei- No. of patients at each tube potential ther an independent t test for normally 80 kVp 1 distributed data or a Mann-Whitney U 100 kVp 10 test for nonnormally distributed data. 120 kVp 44 Ordinal variables (ie, image quality) are Tube current–time product (mAs) 331.4 6 22.9 presented as medians with interquartile CT dose index volume (mGy) 35.0 6 21.2 ranges and were compared by using Dose-length product (mGy { cm) 609.1 6 394.3 the Kruskal-Wallis analysis of variance. Scan length (cm) 12.9 6 1.8 P values less than .05 were considered Effective dose (mSv) to indicate a signifi cant difference. In- Retrospective ECG gated (n = 35) 10.5 6 4.2 Prospective ECG triggered (n = 11) 6.6 6 3.1 terobserver agreement for subjective High pitch spiral (n = 9) 2.3 6 1.4 image quality was quantifi ed by using k statistics. The diagnostic accuracy Note.—Unless otherwise specifi ed, data are means 6 standard deviations. (ie, accuracy, sensitivity, specifi city, nega- * Data in parentheses are the range. tive predictive value, and positive pre- dictive value [PPV]) of each recon- struction technique for detection of Figure 1 large ( . 50%) stenoses was calculated with cardiac catheterization as the ref- erence standard. Differences in diag- nostic accuracy between the two recon- struction techniques were compared by creating matched sample tables and by using a modifi ed McNemar test to cal- culate P values.

Results All 55 coronary CT angiograms were successfully completed and considered to be of diagnostic image quality. Pa- tient demographics and coronary CT angiogram characteristics are provided in Table 1 . The average body mass index in our patient cohort was 31.6 kg/m2 , indicative of a high prevalence of obe- sity, and the average Agatston calci- um score was 710, refl ecting advanced Figure 1: Schematic of the iterative reconstruction process (Iterative Reconstruc- . tion in Image Space [IRIS]; Siemens Healthcare). A single FBP image reconstruction initially takes place in the raw data domain to generate a master reconstruction. All Image Noise, Attenuation, and Subjective further reconstruction steps take place in image space. Further steps include a Image Quality regularization loop based on prior image information with the goal of suppressing In both image reconstructions using the artifacts and reducing image noise. − = subtraction of detected image noise, vascular I26f and B26f as well as the 1 = modeling of image noise and validation of corrections, n = image no. high spatial resolution B46f and I46f

within iterative loop, n- 1 = image of prior iterative loop, NIter = no. of iterations. convolution algorithms, mean image

Radiology: Volume 260: Number 2—August 2011 n radiology.rsna.org 393 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

Table 2 Mean Image Noise and Attenuation for the Four Different Reconstructions Vascular Algorithm High-Spatial-Resolution Algorithm Iterative Reconstruction Iterative Reconstruction Anatomic Region FBP with B26f Kernel with I26f KernelP Value FBP with B46f Kernel with I46f KernelP Value

Mean Image Noise Ascending aorta 33.4 6 11.7 24.9 6 10.8 .013 48.3 6 15.2 35.9 6 10.0 .025 Interventricular septum 30.5 6 13.7 22.9 6 12.5 .023 52.8 6 16.6 38.0 6 13.1 .011 Left ventricular cavity 38.6 6 16.9 31.9 6 16.3 .035 64.1 6 28.2 47.8 6 21.3 .017 Mean Attenuation (HU) Ascending aorta 427.0 6 131.8 430.7 6 131.0 .085 431.7 6 174.5 435.2 6 176.9 .087 Interventricular septum 124.2 6 42.2 119.1 6 30.5 .069 128.7 6 28.4 126.8 6 35.2 .088 Left ventricular cavity 367.9 6 133.0 372.1 6 125.3 .075 373.1 6 198.8 384.4 6 216.4 .073

Note.—Unless otherwise specifi ed, data are means 6 standard deviations.

noise measured signifi cantly lower us- Figure 2 ing iterative reconstruction than FBP in all ROIs, while there was no signifi cant difference in mean attenuation within the same anatomic regions (Table 2 ) ( Fig 2 ). Image quality of coronary CT an- giograms reconstructed with iterative reconstruction was rated signifi cantly higher than that of those reconstructed with FBP by both observers. With both iterative reconstructions (I26f and I46f kernels), the median image quality score was 5; whereas with both FBP recon- structions (B26f and B46f kernels), the median image quality score was 4. Pair- wise Wilcoxon rank sum tests showed statistically higher ratings for iterative reconstruction than for FBP (I26f vs B26f, P = .042; I46f vs B46f, P = .031). Interobserver agreement was excellent for iterative and FBP reconstructions (I26f, k = 0.89; I46f, k = 0.92; B26f, k = 0.84; B46f, k = 0.88). Figure 2: Contrast-enhanced retrospectively ECG-gated coronary CT angiographic images in a 63-year-old Volumetric Analysis of Coronary Artery man (body mass index, 35.2 kg/m2 ) referred for coronary catheterization because of chest pain on exertion. Calcifi cations Left: Right coronary artery as curved multiplanar reformation along vessel centerline (purple). Right: Trans- A total of 142 circumscribed coronary verse sections through vessel (purple), orthogonal to centerline. Top: FBP reconstruction shows image noise artery calcifi cations were volumetrically in a circular aortic root ROI of 35.8 HU (red). Bottom: iterative reconstruction shows image noise in a circular analyzed among all four reconstructions. aortic root ROI of 28.6 HU (red). Suppression of blooming artifacts in iterative reconstructions improves ves- Coronary artery calcifi cations showed sel delineation adjacent to the calcifi ed lesion (arrows) and facilitates estimation of the true degree of signifi cantly lower volumes on iterative luminal narrowing. reconstruction images compared with FBP images (I26f vs B26f: 46.2 mm 3 6 Coronary CT Angiography versus ization as well as with coronary CT an- 68.8 vs 56.3 mm3 6 72.5, P = .026; I46f vs Coronary Catheterization giography. Cardiac catheterization showed B46f: 44.3 mm3 6 64.7 vs 54.5 mm 3 6 A total of 825 coronary artery segments stenoses larger than 50% in 31 (56%) 69.5, P = .019) ( Fig 3 ). were analyzed with cardiac catheter- patients, with a total of 104 lesions. Six

394 radiology.rsna.org n Radiology: Volume 260: Number 2—August 2011 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

patients had one-vessel disease, 18 pa- Figure 3 tients had two-vessel disease, and seven patients had three-vessel disease. By using iterative reconstruction instead of FBP, there was a signifi cant improve- ment in overall accuracy, specifi city, and PPV on a per-segment level and in spec- ifi city and PPV on a per-patient level for the detection of signifi cant stenosis with coronary CT angiography ( Table 3 ). In 33 of 825 segments and three of 55 pa- tients, iterative reconstruction enabled reclassifi cation from false-positive to true-negative results (Figs 4, 5 ). In one patient, an isolated short but important stenosis in the distal posterior descend- ing artery was missed in both iterative reconstruction and FBP series.

Discussion Our fi ndings show that the use of itera- tive reconstruction instead of traditional FBP signifi cantly improves the specifi c- ity, PPV, and overall accuracy of coro- nary CT angiography for stenosis detec- tion in patients with heavily calcifi ed coronary arteries. These improvements are likely related to decreases in image noise, improved diagnostic image qual- ity, and reduced blooming artifacts. Advances in CT technology, with im- provements in temporal and spatial res- Figure 3: Contrast-enhanced prospectively ECG-triggered coronary CT angiographic images in a 76-year- olution, have continuously increased the old woman with chest pain and abnormal nuclear myocardial perfusion scan. Transverse (a) FBP and robustness and accuracy of coronary CT (b) iterative reconstruction images at the level of the aortic root show extensive calcifi ed plaque burden (arrows) angiography for the noninvasive assess- in the proximal left anterior descending coronary artery. Threshold-dependent volumetry of calcifi cations ment of coronary artery disease (22,23 ). (purple) resulted in a measured volume of 61.34 mm3 on (c) FBP reconstructions and of 45.32 mm3 on A multitude of studies have demonstrated (d) iterative reconstructions. the ability of this test to be used to re- liably exclude coronary artery stenosis positive fi ndings do not detract from this obscuring the coronary artery lumen as compared with coronary catheteriza- test’s sensitivity; however, they decrease ( 33–35 ). Blooming artifacts can be re- tion ( 23–28 ) or as assessed through pa- specifi city, often leading to unnecessary duced by increasing spatial resolution tient outcome (29–31 ). However, these subsequent layered testing to exclude through thinner collimated section widths studies have also identifi ed several per- stenotic coronary artery disease, which and reconstruction thickness, as well as sistent limitations that remain consider- incurs additional cost, patient radiation use of higher-resolution sharper recon- able detractors from the diagnostic per- exposure, and potential complications struction algorithms. However, these ap- formance of coronary CT angiography. from coronary catheterization, thus lim- proaches come at the expense of higher Among these are patient obesity, with iting the clinical usefulness of coronary image noise or higher radiation dose re- associated high image noise, and the CT angiography (3,33 ). quirements for suppressing image noise presence of heavy coronary artery calci- Blooming artifacts from calcifi ed when traditional FBP is used. Iterative fi cations (25,26,32 ). Blooming artifacts plaques are partially attributed to the reconstruction techniques, to a certain exaggerate the size of densely calcifi ed limited spatial and point-spread reso- extent, allow decoupling of spatial res- plaques and limit the accurate evalua- lution of CT image reconstruction al- olution and image noise and offer the tion of the adjacent coronary artery lu- gorithms, which create a spillover ef- potential to selectively improve high- men, typically leading to overestimation fect from high-attenuation structures contrast resolution without affecting im- of lesion severity. The resulting false- into adjacent lower-attenuation voxels, age noise in low-contrast areas ( 5,36 ).

Radiology: Volume 260: Number 2—August 2011 n radiology.rsna.org 395 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

Table 3 Diagnostic Accuracy of FBP and Iterative Reconstruction for the Detection of Coronary Artery Stenoses Larger than 50% versus Reference Standard Cardiac Catheterization Per Segment Per Patient Parameter FBP Iterative Reconstruction P Value FBP Iterative Reconstruction P Value

Accuracy (%) * 91.8 (71.7, 91.1) 95.9 (78.2, 94.9) .0001 83.6 (89.7, 93.5) 89.1 (94.3, 97.0) NS Sensitivity (%) * 95.2 (83.8, 99.4) 96.2 (83.8, 99.4) NS 96.7 (89.2, 97.9) 96.7 (90.5, 98.5) NS Specifi city (%) 91.2 (46.7, 82.0) 95.8 (59.5, 90.8) .0001 66.7 (89.0, 93.1) 79.2 (94.1, 97.1) .0189 Negative predictive value (%) * 99.2 (69.2, 99.7) 99.4 (73.1, 99.7) NS 94.1 (98.1, 99.7) 95.0 (98.4, 99.8) NS PPV (%) * 61.1 (62.2, 89.9) 76.9 (69.0, 94.6) .0001 78.9 (53.1, 68.6) 85.7 (68.6, 83.7) .0403 No. of true-positive fi ndings 99 100 NS 30 30 NS No. of false-positive fi ndings 63 30 .0001 8 5 NS No. of true-negative fi ndings 658 691 .0001 16 19 NS No. of false-negative fi ndings 5 4 NS 1 1 NS

Note.—NS = not signifi cant. * Data in parentheses are 95% confi dence intervals.

Accordingly, the properties of iterative Figure 4 reconstruction seem better suited than those of traditional FBP to address the requirements of coronary CT angiogra- phy, where both high spatial and con- trast resolution are of importance for the evaluation of small target vessels containing both high-attenuation (cal- cifi cations, stents) and low-attenuation (noncalcifi ed plaque) structures. As a postacquisition image reconstruction approach, the effects of iterative re- construction should be largely inde- pendent of the image acquisition tech- nique (eg, prospective ECG triggering vs retrospective ECG gating), although we did not subanalyze our results for this aspect. Figure 4: Contrast-enhanced prospectively The limitations of coronary CT an- ECG-triggered coronary CT angiographic im- giography will not be completely over- ages in a 55-year-old man with chest pain. Curved multiplanar reformations of the left anterior come with iterative reconstruction tech- descending coronary artery show heavy niques. However, the observed increase calcifications (arrows) in the midportion of in specifi city and PPV with use of this the vessel. Blooming artifacts limit evaluation technique indicates the potential to re- of adjacent vessel lumen on (a) FBP reconstruc- duce the need for unnecessary further tions but are reduced on (b) iterative recon- testing in patients undergoing coronary structions, which enabled correct classifi cation CT angiography. We specifi cally selected of the lesion as eccentric and only minimally for individuals with Agatston scores of stenotic, as confi rmed with (c) subsequent at least 400, whereas the 2010 appropri- coronary catheterization. ate use criteria for cardiac computed to- mography (37 ) consider the usefulness of coronary CT angiography in this scenario as uncertain. The more widespread im- plementation of iterative reconstruction techniques may hold potential for wid- ening the scope of patients in whom

396 radiology.rsna.org n Radiology: Volume 260: Number 2—August 2011 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

Figure 5

Figure 5: Contrast-enhanced prospectively ECG-triggered coronary CT angiographic images in a 73-year-old man with chest pain. Curved multiplanar reformations of the left anterior descending coronary artery show heavy calcifi cations (arrows) in the proximal vessel. Blooming artifacts limit evaluation of adjacent vessel lumen on (a) FBP reconstructions, mimicking a substantial stenosis, but are reduced on (b) iterative reconstructions, which enabled correct classifi cation of the lesion as not signifi cantly stenotic, as confi rmed with (c) subsequent coronary catheterization. coronary CT angiography can be effec- because of the distinct differences in a test. However, this does not detract tively and benefi cially used. image characteristics between the two from the observation that a long-standing In this investigation, we focused approaches. Moreover, in our study, limitation of coronary CT angiogra- solely on possible improvements in diag- all images were derived from the same phy (ie, heavy calcifi cations) might be nostic accuracy garnered by the use of raw data for each patient, leaving the ameliorated by abandoning traditional iterative reconstruction over FBP in pa- possibility of some biases and infl ated FBP as the standard method for image tients in whom heavily calcifi ed coronary statistical signifi cance owing to multi- reconstruction. arteries pose a diagnostic challenge. reader effects that were not accounted In summary, the results of our study Recent reports indicate the potential for. The effect of interreader variation, suggest that accuracy, specifi city, and of radiation dose reduction through de- which may have infl uenced the results PPV of stenosis detection with coro- creases in image noise with use of itera- on image quality, is likely small since the nary CT angiography in heavily calcifi ed tive reconstruction techniques (10,11 ). overall agreement between readers was vessels can be incrementally improved However, this was not the focus of our excellent (. 0.6), and the variance of with use of iterative reconstruction in- current study. k statistics was lower than 10%. Based stead of FBP. Accordingly, iterative re- Due to the size of our cohort, only on the design of this study, intrareader construction should be preferentially a limited number of patients were cor- variation was closely related to the choice used in patients with advanced athero- rectly reclassifi ed as free of stenosis on of different kernels used and could be sclerosis to reduce the number of un- the basis of iterative reconstruction, interpreted as a systematic infl uence necessary follow-up studies. and the improvement in per-patient ac- across both reconstruction techniques. Disclosures of Potential Confl icts of Interest: curacy did not reach signifi cance. How- Another limitation is the absence of M.R. Financial activities related to the present ever, the signifi cant increases in speci- an outside reference standard, such as article: none to disclose. Financial activities not fi city and PPV both on a per-segment intravascular ultrasonography, to val- related to the present article: receives travel/ac- commodations/meeting expenses from Siemens. and per-patient level hold promise to idate our volumetric measurements Other relationships: none to disclose. J.W.N. No improve the relatively low performance of coronary artery calcifi cations. Since potential confl icts of interest to disclose. U.J.S. ( 38 ) of this test for the evaluation of we selected for patients with advanced Financial activities related to the present article: none to disclose. Financial activities not related heavily calcifi ed coronary arteries. In atherosclerosis whose symptoms had to the present article: is a consultant for and particular, when extrapolated to the indicated the need for coronary catheter- is on the speakers bureau of Bayer Healthcare, overall number of patients undergoing ization, the disease prevalence was high, Bracco, GE Healthcare, Medrad, and Siemens coronary CT angiography, the benefi ts exceeding 50%. Thus, the extrapolation Healthcare; institution has grants from Bayer Healthcare, Bracco, GE Healthcare, Medrad, and of reduced follow-up testing could be of the general performance parameters Siemens Healthcare. Other relationships: none notable. A general inherent limitation of coronary CT angiography observed in to disclose. T.X.O. No potential confl icts of in- of all studies that compare iterative re- our study to a population with a lower terest to disclose. P.L.Z. No potential confl icts of construction with traditional FBP is the pretest likelihood of disease is limited interest to disclose. M.M. No potential confl icts of interest to disclose. J.M.K. Financial activities fact that observers cannot be effectively by the well-known effects of disease related to the present article: none to disclose. blinded to the reconstruction technique prevalence on the predictive value of Financial activities not related to the present

Radiology: Volume 260: Number 2—August 2011 n radiology.rsna.org 397 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

article: is a consultant for and on the speakers coronary angiography with standard fi ltered coronary artery disease: report of the Ad bureau of Siemens. Other relationships: none to back-projection and invasive diagnostic Hoc Committee for Grading of Coronary disclose. R.W.B. Financial activities related to coronary angiography. Heart 2010 ; 96 ( 12 ): Artery Disease, Council on Cardiovascular the present article: none to disclose. Financial 922 – 926 . , American Heart Association. Cir- activities not related to the present article: is culation 1975; 51 ( 4 Suppl): 5 – 40 . on the speakers bureau of and receives travel/ 11 . Leipsic J , Labounty TM , Heilbron B , et al. accommodations/meeting expenses from Sie- Estimated radiation dose reduction using 22 . Wang Y, Zhang Z , Kong L , et al. Dual- mens Healthcare. Other relationships: none to adaptive statistical iterative reconstruction in source CT coronary angiography in patients disclose. C.F. No potential confl icts of interest to coronary CT angiography: the ERASIR study. with atrial fi brillation: comparison with disclose. T.J.V. No potential confl icts of interest AJR Am J Roentgenol 2010 ; 195 ( 3 ): 655 – 660 . single-source CT. Eur J Radiol 2008 ; 68 ( 3 ): to disclose. T.H. No potential confl icts of interest 434 – 441. to disclose. 12. Prakash P , Kalra MK , Ackman JB , et al. Diffuse disease: CT of the chest with 23 . Rist C , Johnson TR , Müller-Starck J , et al . adaptive statistical iterative reconstruction Noninvasive coronary angiography using References technique. Radiology 2010 ;256 ( 1 ): 261 – 269 . dual-source computed tomography in pa- 1 . Vanhoenacker PK , Heijenbrok-Kal MH , Van tients with atrial fi brillation. Invest Radiol 13. Prakash P , Kalra MK , Kambadakone AK , et al. Heste R , et al . Diagnostic performance of 2009 ;44 (3 ): 159 – 167 . Reducing abdominal CT radiation dose with multidetector CT angiography for assess- adaptive statistical iterative reconstruction 24 . Hamon M , Biondi-Zoccai GG , Malagutti P , ment of coronary artery disease: meta- technique. Invest Radiol 2010 ; 45 ( 4 ): 202 – 210 . et al. Diagnostic performance of multislice analysis. Radiology 2007 ;244 ( 2 ): 419 – 428 . spiral computed tomography of coronary ar- 14. Achenbach S , Marwan M , Ropers D, et al . 2 . Meijs MF , Meijboom WB , Prokop M , et al . teries as compared with conventional inva- Coronary computed tomography angiogra- Is there a role for CT coronary angiogra- sive coronary angiography: a meta-analysis. phy with a consistent dose below 1 mSv us- phy in patients with symptomatic ? J Am Coll Cardiol 2006 ; 48 ( 9 ): 1896 – 1910 . ing prospectively electrocardiogram-triggered effect of coronary calcium score on identifi - high-pitch spiral acquisition. Eur Heart J 2010; 25 . Stein PD , Beemath A , Kayali F , Skaf E , cation of stenosis. Int J Cardiovasc Imaging 31 ( 3): 340 – 346 . Sanchez J, Olson RE. Multidetector computed 2009 ;25 (8 ): 847 –854 . tomography for the diagnosis of coronary 15 . Bongartz G , Golding SJ , Jurik AG , et al . 3 . Raff GL , Gallagher MJ , O’Neill WW , Gold- artery disease: a systematic review. Am J Med 2004 CT quality criteria . Luxembourg: Eu- stein JA . Diagnostic accuracy of noninvasive 2006 ; 119 (3 ): 203 – 216 . ropean Commission, 2004 . coronary angiography using 64-slice spiral 26 . Scheffel H , Alkadhi H , Plass A , et al. Accuracy computed tomography. J Am Coll Cardiol 16. Achenbach S , Boehmer K , Pfl ederer T , et al. of dual-source CT coronary angiography: 2005 ;46 (3 ): 552 –557 . Infl uence of slice thickness and reconstruc- fi rst experience in a high pre-test probability tion kernel on the computed tomographic 4 . Zhang LJ , Wu SY , Wang J , et al . Diagnostic population without heart rate control. Eur attenuation of coronary atherosclerotic accuracy of dual-source CT coronary angiog- Radiol 2006; 16 (12 ): 2739– 2747. plaque . J Cardiovasc Comput Tomogr 2010 ; raphy: the effect of average heart rate, heart 4 ( 2 ): 110 – 115 . 27 . Martuscelli E , Romagnoli A , D’Eliseo A , rate variability, and calcium score in a clinical et al. Accuracy of thin-slice computed tomog- perspective. Acta Radiol 2010 ; 51 ( 7 ): 727 – 740 . 17 . Cademartiri F , La Grutta L , Runza G , et al. raphy in the detection of coronary stenoses. Infl uence of convolution fi ltering on coronary 5 . Thibault JB, Sauer KD, Bouman CA , Hsieh J. Eur Heart J 2004; 25 (12 ): 1043 – 1048 . plaque attenuation values: observations in A three-dimensional statistical approach to an ex vivo model of multislice computed to- 28 . Hoffmann MH , Shi H , Schmitz BL , et al. improved image quality for multislice helical CT. mography coronary angiography. Eur Radiol Noninvasive coronary angiography with mul- Med Phys 2007 ; 34 (11 ): 4526 – 4544 . 2007 ; 17 ( 7 ): 1842 – 1849 . tislice computed tomography. JAMA 2005 ; 6 . Zhang S , Levin DC , Halpern EJ , Fischman D , 293 ( 20 ): 2471 – 2478. 18. Cademartiri F, Runza G, Mollet NR , et al . Savage M , Walinsky P. Accuracy of MDCT Infl uence of increasing convolution kernel 29 . Min JK , Shaw LJ, Devereux RB , et al . Prog- in assessing the degree of stenosis caused by fi ltering on plaque imaging with multislice nostic value of multidetector coronary com- calcifi ed coronary artery plaques. AJR Am J CT using an ex-vivo model of coronary puted tomographic angiography for predic- Roentgenol 2008 ; 191 ( 6 ): 1676 – 1683 . angiography . Radiol Med (Torino) 2005 ; tion of all-cause mortality. J Am Coll Cardiol 7 . Brooks RA , Di Chiro G . Theory of image 110 ( 3 ): 234 – 240 . 2007 ;50 (12 ): 1161 – 1170 . reconstruction in computed tomography. 19. Seifarth H , Raupach R , Schaller S , et al . 30 . Gaemperli O , Valenta I , Schepis T , et al. Radiology 1975 ; 117( 3 Pt 1 ): 561 – 572 . Assessment of coronary artery stents using Coronary 64-slice CT angiography predicts 8 . Lange K , Carson R . EM reconstruction algo- 16-slice MDCT angiography: evaluation of a outcome in patients with known or sus- rithms for emission and transmission tomog- dedicated reconstruction kernel and a noise pected coronary artery disease. Eur Radiol raphy . J Comput Assist Tomogr 1984 ; 8 ( 2 ): reduction filter. Eur Radiol 2005 ; 15 ( 4 ): 2008; 18 ( 6 ): 1162 – 1173 . 306 –316 . 721 – 726 . 31 . Ostrom MP , Gopal A , Ahmadi N, et al. Mor- 9 . Flicek KT , Hara AK , Silva AC , Wu Q , Peter 20. Thilo C , Gebregziabher M , Mayer FB , tality incidence and the severity of coronary MB , Johnson CD . Reducing the radiation Zwerner PL , Costello P , Schoepf UJ . Cor- atherosclerosis assessed by computed to- dose for CT colonography using adaptive sta- relation of regional distribution and mor- mography angiography. J Am Coll Cardiol tistical iterative reconstruction: a pilot study. phological pattern of calcifi cation at CT 2008; 52 ( 16 ): 1335– 1343 . AJR Am J Roentgenol 2010 ; 195 ( 1 ): 126 – 131 . coronary artery calcium scoring with non- 32 . Catalán P , Leta R , Hidalgo A , et al. Ruling calcifi ed plaque formation and stenosis. 10 . Gosling O, Loader R, Venables P , et al. A com- out coronary artery disease with noninva- Eur Radiol 2010 ;20 (4 ): 855 – 861 . parison of radiation doses between state-of- sive coronary multidetector CT angiography the-art multislice CT coronary angiography 21 . Austen WG , Edwards JE , Frye RL , et al. A before noncoronary cardiovascular surgery. with iterative reconstruction, multislice CT reporting system on patients evaluated for Radiology 2011 ; 258 ( 2 ): 426 – 434 .

398 radiology.rsna.org n Radiology: Volume 260: Number 2—August 2011 CARDIAC IMAGING: Coronary CT Angiography: Iterative versus Filtered Back Projection Reconstruction Renker et al

33 . Leber AW , Knez A , von Ziegler F, et al . Quan- computed tomography angiography. J Am Tomography, the American College of Radi- tifi cation of obstructive and nonobstructive Coll Cardiol 2006; 47 ( 8 , Suppl ): C40 –C47 . ology, the American Heart Association, the coronary lesions by 64-slice computed to- 36 . Hara AK , Paden RG , Silva AC , Kujak JL , American Society of , the mography: a comparative study with quan- Lawder HJ , Pavlicek W . Iterative recon- American Society of Nuclear , the titative coronary angiography and intravas- struction technique for reducing body radia- North American Society for Cardiovascular cular ultrasound. J Am Coll Cardiol 2005 ; tion dose at CT: feasibility study. AJR Am J Imaging, the Society for Cardiovascular An- 46( 1): 147 –154 . Roentgenol 2009 ; 193 ( 3 ): 764 – 771 . giography and Interventions, and the Soci- 34 . Brodoefel H , Burgstahler C , Tsifl ikas I , et al. ety for Cardiovascular Magnetic Resonance. 37 . Taylor AJ , Cerqueira M , Hodgson JM, et al . J Am Coll Cardiol 2010 ;56 (22 ): 1864 – 1894 . Dual-source CT: effect of heart rate, heart ACCF/SCCT/ACR/AHA/ASE/ASNC/NASCI/ rate variability, and calcifi cation on image SCAI/SCMR 2010 appropriate use criteria 38 . Brodoefel H , Tsifl ikas I , Burgstahler C , et al. quality and diagnostic accuracy. Radiology for cardiac computed tomography: a report Cardiac dual-source computed tomography: 2008 ;247 (2 ): 346– 355. of the American College of Cardiology Foun- effect of body mass index on image qual- 35 . Cordeiro MA , Lima JA . Atherosclerotic dation Appropriate Use Criteria Task Force, ity and diagnostic accuracy . Invest Radiol plaque characterization by multidetector row the Society of Cardiovascular Computed 2008; 43 ( 10 ): 712 – 718 .

Radiology: Volume 260: Number 2—August 2011 n radiology.rsna.org 399