Performance Evaluation of Gamma Cameras with Multi-Pinhole Collimators for Brain SPECT Imaging

Performance evaluation of gamma cameras with multi-pinhole collimators for brain SPECT imaging

Performance Evaluation von Gamma Kameras mit Multi-Pinhole Kollimatoren für SPECT Bildgebung des Gehirns

KRISTIAN TECKLENBURG BORNON OCTOBER 16 1991

MASTER THESISIN PHYSICS UNIVERSITYOF HAMBURG

2019

1. REVIEWER PROF.DR. RER. NAT.ERIKA GARUTTI

INSTITUTEOF EXPERIMENTAL PHYSICS,UNIVERSITYOF HAMBURG

2. REVIEWER DR. RER. NAT.RALPH BUCHERT

DEPARTMENT FOR DIAGNOSTICAND INTERVENTIONAL RADIOLOGYAND NUCLEAR MEDICINE,

UNIVERSITY MEDICAL CENTER HAMBURG-EPPENDORF For my friends ❆❜❛❝

In this master thesis a standard for performance evaluation of gamma cameras with multi- pinhole (MPH) collimators for clinical brain SPECT imaging is proposed. The evaluation standard has been developed on a triple-head SPECT system (Mediso AnyScan® TRIO) using a novel 20-pinhole collimator speciﬁcally designed for dopamine transporter (DAT) SPECT. MPH collimators, very successful in small animal SPECT, promise to overcome the trade-off between spatial resolution and count sensitivity in clinical SPECT of "small" organs, that is otherwise associated with photon collimation using parallel-hole or fan-beam collimators.

Count sensitivity and spatial resolution were measured with a 99mTc point source placed axially on the center of rotation axis and radially at the axial sensitivity peak in increments of 1 cm covering the field-of-view (FOV). Acquisition and reconstruction parameters have been optimized for clinical patient scanning using an anthropomorphic striatum phantom with activity filling representative of a typical patient scan. Image artifacts due to incomplete sampling have been reduced by investigating different helical scanning modes using a self-made multi-disk phantom. Recovery of putamen-to-background contrast was assessed by the contrast-recovery- coefficient (CRC). Measurements were repeated with conventional 2-head SPECT systems with fan-beam and low-energy-high-resolution-high-sensitivity (LEHRHS) collimators. A patient referred to DAT SPECT because of suspicion of Parkinson’s disease was scanned with both LEHRHS and MPH collimators after a single tracer injection.

The MPH sensitivity proﬁle was almost symmetrical around its peak but decreased signiﬁcantly towards the edges of the FOV. However, it outperformed typical sensitivity of 2-head parallel- hole collimators in a range of 129 3mm axially and > 200mm transaxially with a peak sen- ± sitivity of 608 5cps/MBq compared to 223 2cps/MBq for the 2-head fan-beam collimator. ± ± Spatial resolution of MPH SPECT was determined to range from 3 mm to 4 mm throughout the whole FOV. MPH SPECT improved striatal contrast recovery by 20% compared to fan-beam ≥ SPECT. The patient scan demonstrated very good image quality of MPH SPECT with almost PET-like delineation of putamen and caudate nucleus, clearly superior to LEHRHS SPECT.

MPH SPECT provides considerable improvement of the resolution-sensitivity tradeoff in DAT SPECT compared to SPECT with fan-beam or LEHRHS collimators. Improved sensitivity opens the possibility to reduce scan time and/or tracer dose. ❑✉③❢❛✉♥❣

In dieser Masterarbeit wird ein Standard zur Performance Evaluation von Gammakameras mit multi-pinhole (MPH) Kollimatoren für die klinische SPECT Bildgebung des Gehirns vorgeschlagen. Das Verfahren wurde an einem 3-Kopf SPECT System entwickelt (Medi- so AnyScan® TRIO), welches mit einem neuartigen 20-Pinhole Kollimator, speziell für die Dopamin-Transporter SPECT, ausgestattet ist. Multi-Pinhole Kollimatoren werden bereits sehr erfolgreich bei der Kleintier SPECT angewendet. Darüber hinaus versprechen diese auch bei der klinischen SPECT "kleiner Organe" den Kompromiss zwischen Sensitivität und räum- licher Auﬂösung überwinden zu können, der konventionell bei Parallelloch- und Fanbeam- Kollimatoren vorherrscht.

Die Sensitivität und räumliche Auﬂösung wurden anhand einer 99mTc Punktquelle gemessen. Hierzu wurde das Gesichtsfeld des Detektors axial entlang der Rotationsachse und am axia- len Sensitivitätsmaximum auch radial in Abständen von 1 cm abgerastert. Akquisitions- und Rekonstruktionsparameter wurden zuvor für die klinische Anwendung mit Hilfe eines anthro- pomorphischen Striatum-Phantoms optimiert, welches mit einer klinisch-typischen radioakti- ven Aktivitätskonzentration befüllt wurde. Bildartefakte aufgrund unvollständiger Projektionen wurden durch die Untersuchung verschiedener Helix-Aufnahmemodi mit einem selbst gebau- ten Multi-Disk Phantom reduziert. Die Wiederherstellung des Striatum-Hintergrund-Kontrastes wurde anhand des "contrast recovery coefﬁcient" (CRC) untersucht. Die Messungen wurden mit Fan-Beam und LEHRHS Kollimatoren an einer konventionellen 2-Kopf SPECT Kamera wiederholt. Ein Patient mit Verdacht auf Morbus Parkinson wurde sowohl mit dem LEHRHS als auch mit dem MPH Kollimatoren nach einer einzigen Kontrastmittel-Injektion gemessen.

Das MPH Sensitivitätsproﬁl zeigt einen beinahe symmetrischen Verlauf um das Maximum, fällt jedoch stark zu den Rändern des Gesichtsfeldes hin ab. Nichtsdestotrotz wird die typische Sensitivität konventioneller 2-Kopf Systeme in einem Bereich von 129 3mm mm axial und ± > 200mm transaxial übertroffen mit einem Sensitivitätsmaximum von 608 5cps/MBq ver- ± glichen mit 223 2cps/MBq beim Fanbeam-Kollimator. Die räumliche Auﬂösung des MPH ± SPECT liegt im Bereich von 3 mm bis 4 mm über das gesamte Gesichtsfeld. Die Kontrast- Wiederherstellung ist beim MPH SPECT 20% höher als beim Fanbeam-SPECT. Auch die ≥ Patientenaufnahme demonstriert eine sehr gute Bildqualität der MPH SPECT mit nahezu PET- ähnlicher Darstellung von Putamen und Nucleus Caudatus, deutlich überlegen zur LEHRHS SPECT.

MPH SPECT ermöglicht eine deutliche Verbesserung des Kompromisses zwischen Auﬂösung und Sensitivität bei DAT SPECT verglichen mit SPECT unter Verwendung von Fanbeam oder LEHRHS Kollimatoren. Die erhöhte Sensitivität eröffnet die Möglichkeit die Aufnahmedauer und/oder Kontrastmitteldosis zu reduzieren. ❈♦♥❡♥

▲✐ ♦❢ ❋✐❣✉❡ VI

▲✐ ♦❢ ❚❛❜❧❡ VIII

❆❜❜❡✈✐❛✐♦♥ IX

✶ ■♥♦❞✉❝✐♦♥ 1

✷ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞ 3 2.1 Emission Tomography ...... 3 2.2 Imaging Principles in Nuclear Medicine ...... 4 2.3 Photon Emission ...... 4 2.3.1 Technetium-99m ...... 5 2.3.2 Iodine-123 ...... 5 2.4 Photon Detection ...... 5 2.4.1 Single Photon Emission Computed Tomography ...... 5 2.4.2 Collimator types in SPECT ...... 8 2.4.3 Image reconstruction ...... 9 2.4.4 Counting Statistics ...... 10 2.4.5 Attenuation Correction in SPECT ...... 11 2.4.6 Performance Evaluation ...... 12

✸ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞ 13 3.1 Triple-head SPECT system ...... 13 3.2 Multi-pinhole collimator ...... 14 3.3 Optimization of acquisition and reconstruction protocol ...... 17 3.4 Count sensitivty and spatial resolution ...... 20 3.5 Visual image quality and contrast recovery as function of the total counts . . . . 23 3.6 Patient scan ...... 23

✹ ❘❡✉❧ 24 4.1 Optimization of acquisition and reconstruction protocol ...... 24 4.2 Count sensitivty and spatial resolution ...... 37 4.3 Visual image quality and contrast recovery as function of the total counts . . . . 44

■❱ ❈♦♥❡♥

4.4 Patient scan ...... 47 4.5 Tests of gamma camera detectors with multi-pinhole collimators (proposal to extend NEMA NU 1 - section 3.3) ...... 48 4.5.1 System Axial and Radial Count Sensitivity ...... 48 Test Conditions ...... 48 Test Equipment ...... 48 Measurement Procedure ...... 49 Calculations and Analysis ...... 49 Reporting ...... 52 4.5.2 System Axial and Radial Spatial Resolution ...... 53 Test Conditions ...... 53 Test Equipment ...... 53 Measurement Procedure ...... 54 Calculations and Analysis ...... 54 Reporting ...... 55 4.6 Analysis of axial and radial sensitivity proﬁles ...... 56

✺ ❉✐❝✉✐♦♥ 59

✻ ❈♦♥❝❧✉✐♦♥ 66

✼ ❆❝❦♥♦✇❧❡❞❣♠❡♥ 67

✽ ❇✐❜❧✐♦❣❛♣❤② 68

✾ ❆♣♣❡♥❞✐① 74 9.1 MatLab code - Point-source analysis (FWHM) ...... 74 9.2 Additional MatLab code ...... 79 9.2.1 Anonymize .dcm data in case of patient scan ...... 79 9.2.2 Slice summing tool ...... 80 9.2.3 Multi-disk analysis - intensity projection on COR-axis ...... 80

❙❛❡♠❡♥ ✭●❡♠❛♥✮ 81

❱ ▲✐ ♦❢ ❋✐❣✉❡

2.1 Overview of decay schemes for 99mTc and 123I...... 5 2.2 Photon-matter interaction ...... 6 2.3 Scintillation crystal - electronic band structure ...... 7 2.4 Geometry of common parallel-hole collimators ...... 8 2.5 Geometry of fan-beam and multi-pinhole collimators ...... 9

3.1 MPH collimator with striatum aperture ...... 14 3.2 Projection pattern on the gamma detector using striatum MPH collimator . . . . 15 3.3 Striatum positioning in FOV/CFOV ...... 16 3.4 Multi-disk (Defrise) phantom - photograph ...... 17 3.5 Anthropomorphic striatum phantom - photograph ...... 18 3.6 Anthropomorphic striatum phantom - CT image and ROI ...... 19

4.1 Angular sampling optimiziation using a star phantom ...... 25 4.2 Static projection pattern using striatum MPH collimator ...... 25 4.3 Axial sampling: circular vs. helical acquisition ...... 26 4.4 Explosion view of the multi-disc phantom ...... 27 4.5 Construction view of the multi-disk phantom B ...... 28 4.6 Construction view of the multi-disk phantom A ...... 29 4.7 Multi-plexing with the multi-disk phantom ...... 30 4.8 Multi-disk phantom at different helical scan modes ...... 30 4.9 Multi-disk phantom: intensity proﬁles ...... 31 4.10 Multi-disk phantom: ROI analysis ...... 32 4.11 Multi-disk phantom: Effect of AC on TeraTom™ reconstruction ...... 33 4.12 TeraTomo™ reconstruction optimization: MCQ / number of subsets ...... 34 4.13 TeraTomo™ reconstruction optimization: number of iterations ...... 35 4.14 Quantitative analysis of striatum MPH SPECT images (Fig. 4.12 and Fig. 4.13) 36 4.15 Striatum phantom: Effects of CT-based attenuation correction ...... 37 4.16 Overview of system count sensitivity proﬁles ...... 38 4.17 Peak sensitivity and resolution depending on td ...... 39 4.18 Axial sensitivity depending on td - Striatum MPH ...... 40 4.19 Axial sensitivity depending on td - Alzheimer MPH ...... 40 4.20 Overview of reconstructed spatial resolution ...... 41 4.21 SPECT of the hot-rod phantom ...... 42

❱■ ▲✐ ♦❢ ❋✐❣✉❡

4.22 Detector configuration for 2-head and 3-head operation ...... 43 4.23 Sensitivity and resolution dependence on acquisition radius ...... 43 4.24 Quantification of image quality as function of total counts ...... 44 4.25 SPECT image quality as function of total counts ...... 45 4.26 Histogram of selected low- and high-count phantom images ...... 46 4.27 Patient scan: MPH versus LEHRHS ...... 47 4.28 Sample holder for point source measurements ...... 49 4.29 KPIs for description of sensitivity curve ...... 51 4.30 Striatum MPH (td = 0mm) - Fit of axial and radial sensitivity curve ...... 53 4.31 Point source analysis to determine FWHM ...... 55 4.32 Analysis of axial sensitivity profiles: Striatum MPH (td = 0,40,80mm) . . . . 56 4.33 Analysis of axial sensitivity profiles: Alzheimer MPH (td = 0,40mm) . . . . . 57 4.34 Analysis of radial sensitivity profiles: Striatum/Alzheimer MPH (td = 0mm) . 58

❱■■ ▲✐ ♦❢ ❚❛❜❧❡

4.1 Typical parallel-hole collimator specifications ...... 50 4.2 Striatum MPH (td = 0mm) - KPI analysis of sensitivity curve ...... 52 4.3 KPIs of axial sensitivity profiles: Striatum MPH (td = 0,40,80mm) ...... 56 4.4 KPIs of axial sensitivity profiles: Alzheimer MPH (td = 0,40mm) ...... 57 4.5 KPIs of radial sensitivity profiles: Striatum/Alzheimer MPH (td = 0mm) . . . 58

❱■■■ ❆❜❜❡✈✐❛✐♦♥

CoV coefﬁcient of variance td table displacement

3D three dimensional

123I iodine-123

99mTc technetium-99m

BG background

CFOV central ﬁeld of view

CI conﬁdence interval

COR center-of-rotation cps counts per second

CRC contrast-recovery-coefﬁcient

CT computed tomography

CT-ACSC CT-based attenuation and scatter correction

CUPS clinical uncertain parkinsonian syndrome

DAT dopamine transporter

FOV ﬁeld of view

FWHM full-width-at-half-maximum

HEGP high energy general purpose

HEHR high energy high resolution

HEHS high energy high sensitivity

HWHM half-width-at-half-maximum

■❳ ❆❜❜❡✈✐❛✐♦♥ it iterations

LEGP low energy general purpose

LEHR low energy high resolution

LEHRHS low-energy-high-resolution-high-sensitivity

LEHS low energy high sensitivity

LoR line of response

MCQ Monte Carlo Quality

MPH multi-pinhole

NAC no attenuation correction

NaI sodium-iodine

NEMA National Electrical Manufacturer Association

PD Parkinson’s disease

PET positron emission tomography

PM photo-multiplier

PS parkinsonian syndrome

ROI region-of-interest

SD standard deviation

SNR signal-to-noise-ratio

SPECT single photon emission computed tomography ss subsets

Tl thallium

UFOV useful ﬁeld of view

VOI volume-of-interest

❳ ✶ ■♥♦❞✉❝✐♦♥

As average life expectancy increases in developed countries, incidences of neurodegenerative diseases become more frequent since age is a major risk factor for almost all neurodegenerative diseases[1]. Parkinson’s disease (PD) is the second most common neurodegenerative disorder after Alzheimer’s disease[2]. PD used to be characterized by its symptoms alone (bradykinesia, resting tremor, rigidity, and postural instability)[3]. Functional imaging in diagnostic nuclear medicine, namely by means of single photon emission computed tomography (SPECT), is able to diagnose PD based on characteristic physiological degeneration in the brain (nigrostriatal degeneration) already at an early stage before movement disorders arise. Functional imaging also provides a differentiation measure between PD and secondary parkinsonian syndrome (PS) that shares the same symptoms as PD but can be treated more effectively[4].

SPECT is able to create a 3-dimensional activity concentration map of the radioactively marked ligand 123I-FP-CIT that binds to the dopamine transporter (DAT) in a subcortical region of the brain called the striatum and can therefore detect a loss of dopaminergic neurons characteristic for PD. This transition from symptom-based to biomarker-based diagnosis using DAT SPECT got adopted by clinical practice guidelines[5, 6]. It also became standard procedure for differentiation between PD and secondary PS in cases of so called "clinical uncertain parkinsonian syndrome" (CUPS)[6–8], because secondary PS does not come with a loss of dopaminergic neurons and, therefore, represent with normal ﬁndings in DAT SPECT.

While the diagnostic accuracy based on DAT SPECT is already high ( 90 % sensitivity and ≥ specificity for experienced readers[9]) and the pharmacokinetic properties of 123I-FP-CIT is good as shown in animal studies with mice[10] and tree shrews [11], it remains challenging in very early disease stages and for patients with CUPS, in particular for less experienced readers. The major limitation of DAT SPECT is limited image quality provided by conventional SPECT technology with respect to spatial resolution, statistical noise and quantitative accuracy and pre- cision. This limitation is a consequence of the tradeoff between spatial resolution and statistical noise associated with using parallel-hole collimators for photon collimation to define the line of response of photons detected by the scintillation detector for image reconstruction[12]. For brain imaging this tradeoff got improved considerably by the introduction of fan-beam collimators, that improve both spatial resolution and count sensitivity by around 20 %[13] making fan-beam collimators the first choice for brain SPECT imaging to date.[12]

✶ ❈❤❛♣❡ ✶✳ ■♥♦❞✉❝✐♦♥

A much higher potential for concurrent improvement of count sensitivity and spatial resolution is associated with multi-pinhole (MPH) collimators.[12, 14, 15] The idea of MPH technology is based on the camera obscura already investigated as a representation of the human eye by Leonardo da Vinci[16]. An image collimated by a pinhole can result in a zoomed projection on the detector depending on the ratio of object-aperture to aperture-detector distance. The image of the detector can then be sampled with intrinsic resolution (usually much higher than extrinsic system resolution) and is only limited by the size of the detector. The line of response (LoR) is given by the detection location of the detector and the pinhole. For multi-pinhole apertures with partially overlapping projections the LoR can only be approximated using iterative reconstruction algorithms. The very convenient size ratio for small objects made this technology very popular in pre-clinical small animal SPECT with mice.[10, 17–26]

The transfer of MPH technology to humans for clinical SPECT imaging of "small" organs including heart[27, 28], thyroid[29], and brain[30–34] is not straightforward, because the proportion between the object to be imaged (e.g., human brain) and the size of the scintillation detector is less favorable in patients. Therefore, optimization of MPH collimator design for very speciﬁc applications remains a current research topic.[23, 35–37]

Performance of conventional collimator SPECT is evaluated by the industry standard NEMA NU-1 [38, 39]. Even though this standard has been updated in 2018 it still cannot be applied to MPH SPECT because determination of the two key performance indicator, i.e. count sensitivity and spatial resolution, rely on measurement procedures that use line-, ﬂat- and volume-source phantoms and standardized analysis procedures that can neither handle more then one projection per detector nor overlapping of projections of any kind.

The aim of this thesis is to propose a new standard in order to perform clinical performance evaluation of MPH collimators. This thesis may be divided in 5 "phases" from development to verification of the procedure in clinical application on a 20 pin-hole collimator specifically designed for DAT SPECT imaging. During the first phase the acquisition and reconstruction parameters are investigated and optimized for patient scans using an anthropomorphic striatum phantom as well as conventional and self-made geometric phantoms. In the second phase the performance of the mentioned collimator is evaluated by means of count sensitivity and reconstructed spatial resolution based on point source measurements. Also, key performance indicators that are suitable to measure the performance and improvements of image quality compared to conventional SPECT are identified. In the third phase these findings are summarized in a standard protocol following the NEMA standard. In the fourth phase this standard is used to evaluate the performance of a second set of MPH collimators designed for brain perfusion SPECT. Finally, in phase 5, the identified performance improvements of MPH SPECT are tested in the clinical application by scanning a patient referred to DAT SPECT because of suspicion of PD twice, both with the conventional parallel-hole collimator and with the evaluated MPH collimator using the optimized acquisition and reconstruction parameters.

✷ ✷ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞

The focus of this thesis is on the experimental part of performance evaluation of a novel component for an imaging system in nuclear medicine. This chapter will only cover the theoretical basics as far as it is necessary for understanding the experimental results. The parts in this chapter dealing with the detector technology are adapted from "SPECT detectors: the Anger Camera and beyond" by Peterson and Furenlid[40]. The introduction on image reconstruction and statistics is adapted from "Mathematics and Physics of Emerging Biomedical Imaging" by Budinger, Wehrli et al.[41] and "Physics in Nuclear Medicine" by Cherry, Sorenson and Phelps[42]. A more detailed mathematical description of computational tomography imaging can be looked up in "Principles of Computerized Tomography Imaging" by Kak and Slaney[43].

✷✳✶ ❊♠✐✐♦♥ ❚♦♠♦❣❛♣❤②

In general, tomography is a non-destructive imaging technique that creates a virtual three- dimensional (3D) model of a physical object from a series of two-dimensional (2D) cross- sections. In conventional transmission tomography, like x-ray computed tomography (CT), the photon source is ﬁxed outside the patient and the absorption spectrum along many axis is cap- tured. The photons interact with the electron shells along their way and different attenuation coefﬁcients of tissue allow the computer-based reconstruction of 2D cross-sections, that are aligned to create a 3D density map of the patient revealing their anatomical structure. The pixel in each cross-section represents a volume element and is called voxel with voxel height representing slice thickness.

However, in emission tomography the photon source is brought inside the patient, usually by intravenous injection or inhaling, in the form of radio-pharmaceuticals (tracers) carrying radioactive isotopes that decay over time and allow the physicians to create a time-resolved 3D map of metabolic activities in the patient. The concentration of the radioactive isotopes in target organs change due to their decay but also biochemical kinetics within the body. That implies that data acquisition for each cross-section image must be relatively fast compared to the change in concentration. This is by no means a disadvantage since acquiring and comparing images at different times allow the assessment of the functional state of biochemical kinetics inside the body.

Due to the very high detection efﬁciency of highly energetic photons, the amount of tracer needed is only in the order of µg compared to several g needed for CT imaging. The two main

✸ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞ advantages are a minimized exposure to radiation and virtual unlimited selection of possible tracers as carrier molecules since the toxicity always is a function of dosage. A good example is the radiopharmaceutical "FP-CIT" or "Ioﬂupane", a radioactively marked cocaine analogue that is used in the diagnosis of Parkinson’s disease. FP-CIT binds to the presynaptic dopamine transporter in a subcortical region called the striatum. The measured activity concentration and distribution corresponds to the density of dopamine transporters that are reduced in Parkinson’s disease.

There are two types of emission tomography in nuclear medicine: Single photon emission computed tomography (SPECT), which will be discussed in the following section, and positron emission tomography (PET), which will not be covered in this thesis.

✷✳✷ ■♠❛❣✐♥❣ ✐♥❝✐♣❧❡ ✐♥ ◆✉❝❧❡❛ ▼❡❞✐❝✐♥❡

✷✳✸ ❤♦♦♥ ❊♠✐✐♦♥

γ-photons in nuclear medicine are products of nuclear decay of radioactive isotopes used as markers in radio-pharmaceuticals. The two isotopes used in this thesis are technetium-99m (99mTc) and iodine-123 (123I). For any given radioactive isotope, the number of decays per unit of time is proportional to the total number of atoms N available:

dN ∝ N (2.1) − dt with dN being the total number of decays happening in the short time interval dt. The indi- − vidual decay process is random, however the average decay rate depends on the isotope speciﬁc decay constant λ.

dN − = λdt (2.2) N

The solution of this simple differential equation allows to calculate the total amount of remain- ing atoms at any given time N(t) from the total number of atoms N0 at t = 0.

λt N(t)= N0 e− (2.3)

The total number of decays (not taking into account attenuation and scatter effects) detected by a gamma camera Ntotal in the time interval [t0,t1] depends on the available activity A0 at t = t0 and the system count sensitivity sen and can be calculated by integrating over the acquisition time:

t 1 λt Ntotal = sen A0 e− dt (2.4) Zt0

✹ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞

✷✳✸✳✶ ❚❡❝❤♥❡✐✉♠✲✾✾♠

The most popular isotope in nuclear medicine is 99mTc, a meta-stable isomer of 99Tc. It has a half-life of 6.007 h and decays mainly by emission of a 141 keV gamma photon with branching ratio of 89 % to the ground state (decay scheme in Fig. 2.1a). Due to its short half-life 99mTc is suitable for clinical application. It is produced on-site with a 99mTc generator containing a decaying molybdenum-99 (99Mo) sample with a half-life of 66 h.

✷✳✸✳✷ ■♦❞✐♥❡✲✶✷✸

123I is a radioactive isotope of iodine. It has a half-life of 13.224 h and decays by electron capture to an excited state of tellurium-123 which immediately decays under emission of a 159 keV gamma photon with branching ratio of 83 % to the ground state (decay scheme in Fig. 2.1b). It is produced off-site in cyclotrons during the interaction of highly energetic protons on xenon atoms. It’s cost is much higher than locally produced 99mTc.

(a) 99mTc (b) 123I

Figure 2.1 – Overview of decay schemes for 99mTc (a) and 123I (b). Pictures taken from Nucle- onicaWiki database (https://www.nucleonica.com/wiki/index.php?title=Decay_Schemes, accessed 22.08.2019)

✷✳✹ ❤♦♦♥ ❉❡❡❝✐♦♥

✷✳✹✳✶ ❙✐♥❣❧❡ ❤♦♦♥ ❊♠✐✐♦♥ ❈♦♠♣✉❡❞ ❚♦♠♦❣❛♣❤②

SPECT is a functional imaging technique to measure the distribution of specialized radioactive tracers in the patient’s body by measuring the emission spectrum of characteristic γ - photons that are created by the nuclear decay of tracer molecules. The tracer is brought into the patient by intravenous injection or inhaling and participates in biophysical activities of the body. Ra- dioactive decay results in the emission of single gamma photons that may leave the patient and can be detected. The rate of emission corresponds to the concentration of radioactive tracers which depending on the speciﬁcs of the tracer relate to certain functional activities inside the body.

✺ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞

100 Figure 2.2 – Most likely interaction of photons with matter depending on photon energy and atomic num- 75 Pair ber of the material. Picture Z production taken from chapter "Interac- Photoelectric absorption tion of Radiation with Mat- 50 ter" from the book "Physics in Nuclear Medicine" by Compton [45] Atomic number, Atomic scattering Cherry et al.

0 0.01 0.1 1 10 100 Photon energy (MeV)

The following paragraphs in this section are partially adopted from the author manuscript "SPECT detectors: the Anger Camera and beyond" by T. Peterson et al. [40].

Photon detectors in SPECT consist of a collimator to deﬁne the LoR and a scintillation crystal to interact with incoming γ-photons. Crystals made from thallium doped sodium-iodine have already been developed 70 years ago[44] but due to their physical properties remain the most popular detector material for 140 keV γ-photons that are emitted during decay of 99mTc. The type of photon interaction depends on the photon energy Eγ and atomic number Z of the ab- sorbing material as illustrated in Figure 2.2. The two most likely interactions of γ-photons in a sodium-iodine-crystal (NaI-crystal) with typical energies of 30 keV to 250 keV is photoelectric absorption and Compton scattering.

In photoelectric absorption, the entire energy Eγ of a γ - photon is absorbed by a core electron with binding energy EB. This absorption can be resonant in an empty state or lead to ionization, i.e. the non-resonant absorption to the continuum. Due to conservation of energy the excited electron propagates with kinetic energy Ekin = Eγ EB through the material, leav- − ing an empty core hole behind. Possible recombination processes of this unstable state are a relaxation process of a higher-state electron under emission of characteristic secondary x-ray photons, emission of Auger electrons in case of re-absorption electrons or excitation of atomic lattice vibrations called phonons. The Auger effect is characterized by immediate re-absorbtion of the secondary x-ray that leads to the emission of one or more electrons coming from the more loosely bound outer shells.

In Compton scattering, the γ-photon interacts with an outer-shell electron and transfers only part of its energy Eγ to it such that the photon energy after interaction Eγ′ < Eγ . The excited electron and the Compton-scattered photon continue to propagate through the material.

Photon energies over 1.1 MeV are very rare in SPECT so the conversion of Eγ into an electron-

✻ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞

Figure 2.3 – Photoelectric absorption in the scintillation crystal with electronic band structures. Picture taken from the author manuscript "SPECT detectors: the Anger Camera and beyond" by T. Peterson et al. [40]

positron pair is very unlikely. In the typical energy range of SPECT from 30 keV to 250 keV the γ-photons experience photoelectric absorption either directly or after a maximum of two Compton scatters. The deposited energy may be reduced due to Compton-escape or the escape of secondary x-ray.

The atoms of a scintillation crystal are in a periodic structure forming an electronic band structure shown in 2.3. When a primary electron from a photoelectric absorption with a high kinetic energy travels through the crystal, the time-varying electric ﬁelds excite valence-band electrons across the bandgap and produce loosely bound electron-hole pairs known as excitons. These pseudoparticles can move across the crystal until they get trapped in luminescent centers, which generally have a smaller bandgap, where they recombine under emission of secondary scintillation photons. The luminescent centers in the typical NaI-crystal are induced by thallium (Tl) doping, which results in characteristic secondary scintillation photons with 415 nm wavelength. Since the concentration of thallium atoms is very low in the NaI-crystal (0.1 mole%)[40], these photons do not experience signiﬁcant re-absorption. Read-out of the scintillation crystal is done by an array of photo-multipliers (PMs) that convert and amplify the signal coming from secondary photons created in the scintillation process. Fast read-out electronics and a computer are connected to the PMs and determine the time, position and energy of each event detected.

The dominance of NaI(Tl) crystals as scintillator material is due to convenient properties and cost effectiveness[40]. The detection efficiency of 140 keV photons in a 1/2" crystal is 94 %. The conversion efficiency is sufficient with approximately 40 optical photons created for each keV

✼ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞ deposited in the crystal that translates to a good energy and spatial resolution. Additionally, it is possible to economically grow very large crystals of up to 80 cm in diameter for single crystal detectors and due to the short lifetime of excited states in NaI(Tl) count rates of multiple 105 counts per second (cps) can be detected.

✷✳✹✳✷ ❈♦❧❧✐♠❛♦ ②♣❡ ✐♥ ❙❊❈❚

There are a variety of lead and tungsten collimator types optimized for different isotopes and applications in SPECT. As can be seen in Figure 2.4 the wall thickness of the collimator depends on the photon energy. To maximize the detector area and by that maximizing system sensitivity it is advisable to use the lowest possible wall thickness for each isotope that still provides a sufﬁcient shielding from photons penetrating the collimator walls. Moreover, collimators with parallel-holes can be optimized for high sensitivity with the expense of spatial resolution or high resolution with the expense of sensitivity by using a thinner or thicker collimator, respectively. A thicker collimator narrows the incidence angle of photons reaching the detector, such that the LoR is determined more precisely.

(a) LEHS (b) LEHR (c) HEHS (d) HEHR

Figure 2.4 – Overview of parallel-hole collimator types optimized for different photon energies, and either high resolution or high sensitivity imaging: Low energy high sensitivity (a)(LEHS), low energy high resolution (b)(LEHR), high energy high sensitivity (c)(HEHS), high energy high resolution (d)(HEHR).

An interesting way to overcome the trade-off between count sensitivity and spatial resolution is the use of special collimators when imaging small organs. The fan-beam collimator (Fig- ure 2.5a) is widely used in clinical routine and increases the angle of photon acceptance by a converging collimator design without sacrificing spatial resolution. Figure 2.5b shows a multi- pinhole design, that is very successful in small animal SPECT and has the potential for concurrent improvement of count sensitivity and spatial resolution by projecting magnified images of the relatively small field of view (FOV) (compared to fan-beam) directly onto the scintillation crystal. The possible image quality and related contrast-to-noise ratio of a MPH collimator only depends on the number of views as well as intrinsic parameters of the camera system like the magnification ratio (limited by the maximum size of the scintillation crystal), the detection efficiency of γ-photons and the intrinsic resolution of the gamma detector, which is determined by

✽ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞ the crystal quality and its read-out electronics.[46] SPECT with either of these special collimator designs require special iterative reconstruction algorithms. To further maximize sensitivity in the case of MPH collimators, some designs allow an overlap of projections, which is known as multi-plexing, to increase the total number of pinholes and by that increasing the number of γ-photons reaching the detector. To improve angular sampling in the case of overlapping projections, helical acquisition modes have been applied in pre-clinical application[47, 48] and simulation studies[21]. The increased complexity of iterative reconstruction algorithms in case of helical SPECT has also been subject of recent research.[49, 50]

In another recent study a concurrent operation of one fan-beam and one multi-pinhole collimator has been investigated.[51] In this setup the larger FOV of the fan-beam collimator would be used to image the periphery around a rather small central ﬁeld of view (CFOV), which is covered by a MPH collimator with superior spatial resolution and count sensitivity.

(a) fan-beam (b) MPH

Figure 2.5 – Overview of special collimator types optimized for imaging small organs: Conven- tional fan-beam collimator (a) and the multi-pinhole collimator (b). The red dot is representing the individual focal point(s).

✷✳✹✳✸ ■♠❛❣❡ ❡❝♦♥✉❝✐♦♥

The aim of emission tomography in nuclear medicine is to measure the 3-dimensional (3D) activity distribution A(x,y,z) in the patient. For an ideal parallel-hole collimator with perfect spatial resolution and no septal penetration, measurements of this activity can only be done along a straight line ~g perpendicular to the face of the detector. Therefore the measured photon ﬂux Φ(x,y) at the detector represents the line integral of the activity A along ~g:

Φ(x,y)= A(x,y,z) d~r , (2.5) Z~g

Since the γ - photons need to travel through the patient along ~h to reach the detector, the 3D distribution of absorption coefﬁcients µ(x,y,z) along this path must be taken into account.

Φ(x,y)= A(x,y,z) exp µ (x,y,z)d~s d~r (2.6) Z~g −Z~h

✾ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞

The counts measured on the detector M(x,y) depend on the detector conversion efﬁciency c that describes the efﬁciency to transform an incoming γ-photon to a signal:

M(x,y)= c Φ(x,y) (2.7)

When a sufﬁcient number of line integrals is measured, it’s possible to recover the activity distribution A(x,y,z) analytically using the inverse Radon transformation. In reality SPECT detectors are read-out in two dimensions and the photons are effected by three-dimensional attenuation and scatter effects both in the patient body as well as in the detector material. The measured projection data is described by the 3D attenuated Radon transform that takes these effects into account and may be looked up in the book "Mathematics and Physics of Emerging Biomedi- cal Imaging".[41] A popular reconstruction algorithm based on inverse Radon transformation is the ﬁltered back-projection algorithm, which was used to reconstruct parallel-hole collimator SPECT images in this thesis.

In cases where the activity distribution cannot be found analytically, iterative reconstruction algorithms are used. During an iterative reconstruction the entire image acquisition process is simulated (Monte-Carlo simulation) for an initial assumption of the activity distribution A(x,y,z). The simulated acquisition data is compared with the measured acquisition data and the assumption of A(x,y,z) is adjusted based on the difference between simulation and measurement. Limiting the maximum alteration of the assumption in each iteration cycle by a regularization factor improves convergence of the reconstruction. A standard algorithm is the ordered subset expectation maximization (OS-EM) algorithm defined by Hudson and Larkin[52]. Instead of performing calculations with the entire projection data at once the data is grouped into an ordered sequence of subsets to significantly accelerate the reconstruction process. The effective number of iterations ite f f is defined as ite f f = it ss, where it is the number of iterations · per subset and ss is the number of subsets. Different iterative algorithms have been used in this thesis for reconstruction of fan-beam SPECT and MPH SPECT images.

✷✳✹✳✹ ❈♦✉♥✐♥❣ ❙❛✐✐❝

This section is partially adopted from the chapter "Nuclear Counting Statistics" in the book "Physics in Nuclear Medicine" by Cherry et al.[53]. Nuclear decay of a radioactive isotope is a random and independent process while a certain average decay rate is expected. Therefore radiation counting measurements follow the Poisson distribution, which describes the probability to measure a given number of events in a ﬁxed time frame. The Poisson distribution is a limit of the binomial distribution for small event probabilities lim and large number of trials lim . In p 0 N ∞ decay counting measurement the number of trials N is given→ by the number of atoms of a certain→ isotope and the event probability p is given by the isotope speciﬁc decay constant λ introduced in equation 2.2 (p = λ). The mean number of decays m per unit time is given by m = N p. For

✶✵ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞ the Poisson distribution the variance σ 2 is equal to the mean: σ 2 = m. The standard deviation σ is deﬁned as σ = √m. Poisson distribution is not symmetric for small values of m, such that peak probability is not equal to the mean.

However, in decay counting measurements in clinical context, the mean number of decays per unit time is large and the discrete, asymmetric Poisson distribution can be approximated well by a continuous, symmetric Gaussian distribution with peak probability m and variance σ 2 = m. In Gaussian distribution 68.3 % of measurement results fall in the interval m σ. When a ± single counting measurement with result M is conducted, it can be assumed that M m and ≈ therefore the true value can be found in the interval M √M in 68.3 % of the times. The ± relative uncertainty is proportional to 1/√M.

This Gaussian model is a good approximation for radiation counting measurement when the only error are random variations in source decay rate. When additional sources of random error are present the results are given by a Gaussian distribution with variance σ 2 m +(δN) with ≈ δN describing additional sources of random error like variations in sample preparation.

When results are obtained from a series of counting measurements the propagation of error has to be taken into account. The variance for sums or differences of independent measurements

M1,M2,M3,... is given by:

2 2 2 σ(M1 M2 M3 ...)= σ(M1) + σ(M2) + σ(M3) + ... (2.8) ± ± ± q ✷✳✹✳✺ ❆❡♥✉❛✐♦♥ ❈♦❡❝✐♦♥ ✐♥ ❙❊❈❚

The easiest way to consider attenuation effects in SPECT is done by assuming a uniform distribution of attenuation coefﬁcients in the patient body, such that µ(x,y,z)= constant. For the head and abdomen region this methods works relatively well such that an accurate activity distribution can be estimated.[41] This type of correction is also known as Chang’s attenuation correction.[54]

To improve attenuation compensation especially in body areas with a more heterogeneous distribution of attenuation coefficients (like in the chest region) a transmission CT image is used to create a precise 3 dimensional map of attenuation coefficients of the patient. This map is incorporated into iterative reconstruction algorithms to achieve a more precise compensation of attenuation effects with fewer artifacts and distortions compared to the first method.[41]

✶✶ ❈❤❛♣❡ ✷✳ ❚❤❡♦❡✐❝❛❧ ❜❛❝❦❣♦✉♥❞

✷✳✹✳✻ ❡❢♦♠❛♥❝❡ ❊✈❛❧✉❛✐♦♥

Performance measurements of gamma cameras used for conventional parallel-hole collimator SPECT are conducted according to National Electrical Manufacturer Association (NEMA) standard NU 1[38, 39]. The NEMA NU 1 standard has been developed by the NEMA NU 1 Task Force, comprising experts from industry like Siemens Healthcare, Philips Healthcare and others. The NEMA NU 1 reviewers are leading scientists from numerous hospitals in the United States, Canada and Spain.

In its current form the standard NEMA NU 1-2018[39] cannot be fully applied to gamma cameras with multi-pinhole collimators. Only measurements of intrinsic parameters, i.e. parameters that are independent of the installed collimator, can be applied. On the other hand, most system parameters (extrinsic parameters) depend on the collimator geometry that define count sensitivity and spatial resolution. Measurement protocols defined by NEMA NU 1 rely on line-, flat- and volume-source phantoms. These phantoms can have overlapping projections and therefore multi-plexing artifacts when imaged with a MPH collimator. Additionally, the analysis of measurements defined by NEMA NU 1 is incompatible with more than one projection per detector.

For MPH SPECT a new protocol for measuring count sensitivity and spatial resolution as well as parameters speciﬁc to MPH SPECT must be developed and evaluated. This was the aim of this thesis.

✶✷ ✸ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

In this chapter the camera system and the multi-pinhole collimator will be described. Before the performance of the collimator can be evaluated the acquisition and reconstruction protocols need to be optimized for clinical application using an anthropomorphic striatum phantom as well as conventional and self-made geometric phantoms. Subsequently, a measurement procedure for count sensitivity and spatial resolution as an extension of the NEMA standard NU 1[39] is developed and applied using optimized parameters. For comparison, sensitivity and resolution data is acquired with well established collimator and camera systems using the same measurement procedures. Finally, a patient referred to DAT SPECT because of suspicion of PD is measured twice with a conventional parallel-hole and the new multi-pinhole collimator and the images are compared.

✸✳✶ ❚✐♣❧❡✲❤❡❛❞ ❙❊❈❚ ②❡♠

The AnyScan® Trio (Mediso Medical Imaging Systems, Budapest, Hungary) is a general purpose triple-head SPECT system. Each head is equipped with a 3/8” NaI(Tl) detector of 585 mm (transaxial) x 470 mm (axial), 16 mm lightguide, and 94 photomultipliers arranged in a hexago- nal grid. The useful ﬁeld of view (UFOV) of the detector is speciﬁed to be 540 mm (transaxial) x 415 mm (axial) with an intrinsic resolution of 2.5 mm.

In addition to conventional parallel-hole collimators (low energy general purpose (LEGP), low energy high resolution (LEHR), low-energy-high-resolution-high-sensitivity (LEHRHS), and high energy general purpose (HEGP)) a variety of MPH collimators are available for this system. A LEHRHS collimator is a compromise between the LEHR and LEHS collimator design presented in Fig. 2.4. The evaluation in this thesis is focused on a set of MPH collimators speciﬁcally designed for dopamine transporter (DAT) SPECT, an imaging procedure used in the diagnosis of Parkinson’s disease. This collimator is optimized for imaging the striatum in the brain and will be referred to as "Striatum MPH" in this thesis. For comparison, a selection of measurements is also performed using a different set of MPH collimators designed for brain perfusion SPECT. This imaging procedure is used in the diagnosis of Alzheimer’s disease, therefore all data concerning this set of collimators will be referred to as "Alzheimer MPH".

✶✸ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

✸✳✷ ▼✉❧✐✲♣✐♥❤♦❧❡ ❝♦❧❧✐♠❛♦

The striatum MPH collimator design is speciﬁed and patented in the European patent speciﬁca- tion # EP3309588B1 and US patent # US 2018/0103918. The collimator has a FOV of 220 mm (transaxial) x 180 mm (axial). Centered in both axial and transaxial direction the FOV contains a CFOV of 120 mm (transaxial) x 120 mm (axial) with improved image quality in which the volume of interest (VOI) should be positioned.

The entire FOV is shifted 7 cm axially towards the patient (inferior direction) in order to simplify positioning of the brain within the FOV also in case of patients with short neck (Fig. 3.1a). The collimator (Fig. 3.1b) features a solid tungsten aperture of 18 mm thickness (to minimize penetration of high energy photons from I-123 decay) with 20 pinholes arranged in 5 columns (oriented in axial direction) and 4 rows (transaxial) (Fig. 3.1c). For comparison, the Alzheimer MPH only features 15 pinholes arranged in 5 axial columns and 3 transaxial rows. Each pinhole is shaped like a double truncated pyramid with the rectangular (non-square) base facing the patient. The surfaces of the pyramids are adapted to prevent scattered photons to reach the detector from lateral directions. Distance between the center-of-rotation-axis (COR- axis) and face of the detector is 285 mm, while the distance between the COR-axis and the focal plane of the pinholes is 140 mm (acquisition radius). The focal plane itself is recessed 5 mm from the face of the aperture.

(b) collimator

shift tungsten FOV aperture 18 mm CFOV

(a) patient position (c) aperture

Figure 3.1 – The pinholes and the lead blind of the MPH collimator are oriented such that the ﬁeld of view (FOV) is shifted 7 cm towards the patient in axial direction (a) which simpliﬁes positioning the VOI in the CFOV that features the best image quality, particularly for patients with short neck. Part (b) shows a photograph of the rear view of the MPH collimator. Part (c) is a technical drawing showing the arrangement of the 20 pinholes on the tungsten aperture. Technical drawing provided by system manufacturer (Mediso Medical Imaging Systems).

✶✹ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

The pinholes can be categorized in 1st, 2nd and 3rd order pinholes depending on their location and projection properties (Fig. 3.1c and Fig. 3.2). The center row in axial direction comprise the 1st order pinholes, which focus on the CFOV without any projection overlap (multi-plexing) on the detector (Fig. 3.2). When performing a full 360° SPECT acquisition, the CFOV can be illustrated as a small cylinder (Fig. 3.3a) inside a larger cylinder representing the FOV (Fig. 3.3b). In order to minimize reconstruction artifacts, the striatum should be positioned in the CFOV. The two adjacent rows in axial direction comprise the 2nd order pinholes that also focus on the CFOV but at different angles and allow for an overlap with the projections from 3rd order pinholes from the two most outer axial rows that cover the peripheral regions of the brain in the FOV. The 3rd order pinhole projections located at the corners of the detector (highlighted in red) are omitted in reconstruction in the case of standard signal read-out with hexagonally arranged photo-multipliers, because image quality and intrinsic resolution is decreased at the corners. The diameter of 1st and 2nd order pinholes is 4 mm, while the 3rd order pinholes have a diameter of 5 mm.

Figure 3.2 – Technical drawing of the projection pattern on the gamma detector using striatum MPH collimator. As there are 20 pinholes, there also are 20 projections on the detector. Figure adapted from Fig.6 of the European patent speciﬁcation # EP3309588B1 with added highlighting of the projections omitted (red, 54) at the detector corners (50) and labeling the order of pinholes (1st order (52), 2nd order (56), 3rd order (58)). Multi-plexing occurs in region (59).

✶✺ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

FOV

CFOV

(a) 1st order of pinholes covering CFOV

FOV

CFOV

(b) 3rd order of pinholes covering periphery of FOV

Figure 3.3 – Schematic representation of the positioning of the striatum in the CFOV in the transaxial cross-section during a DAT SPECT. Part (a) shows the 1st order of pinholes covering CFOV without multi-plexing. Part (b) shows the 3rd order of pinholes covering also the periphery of the FOV. Multi-plexing does occur but is not visible from the perspective of drawing.

The MPH aperture is mounted on a lead blind of 32 mm thickness in inferior direction to shield γ-photons coming from other body regions, like the bladder. The lead shielding in superior direction is only 12 mm thick to reduce weight. The pyramid-shaped shielding also determines the distance between the aperture and the scintillation crystal which deﬁnes the magniﬁcation ratio of the projections. (Figure 3.1a and 3.1b). Projection views are acquired in a 256 x 256 matrix with 2.13 mm x 2.13 mm pixel size.

For reconstruction of MPH projection data, the system software of the SPECT camera provides an iterative 3-dimensional Monte Carlo-based algorithm, called TeraTomo™ [55]. The reconstructed cross-sections have a 128 x 128 matrix with 1.72 mm x 1.72 mm pixel size and 1.72 mm slice thickness.

✶✻ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

✸✳✸ ❖♣✐♠✐③❛✐♦♥ ♦❢ ❛❝✉✐✐✐♦♥ ❛♥❞ ❡❝♦♥✉❝✐♦♥ ♣♦♦❝♦❧

The impact of angular sampling was tested with a custom-made "star" phantom with two different sections (CT image provided in results chapter, Fig. 4.1). Since the phantom has not been on site the raw data was provided by Attila Forgács, who acquired the data at Scanomed Nu- clear Medicine Center Debrecen (Debrecen, Hungary) on an identical SPECT system. The total number of views nv acquired with the 3 detector heads was varied from 6 to 96, consequently the number of gantry positions per head np was varied from 2 to 32 (np = nv/3).

A static measurement of a large homogeneously ﬁlled cylinder phantom was conducted using only a single Striatum MPH equipped detector head in order to conﬁrm the designed projection interference pattern (Fig. 3.2).

The SPECT system provides helical acquisition mode to avoid reconstruction artifacts due to insufﬁcient sampling with the MPH collimators in circular acquisition mode[56]. The total table displacement td of the helical movement for one full gantry rotation is achieved by moving the table by td/np at each angular gantry position in a "step and shoot" manner for the full SPECT acquisition. A custom-made multi-disk (MD) phantom was developed, build and measured to test the impact of td in helical acquisition modes for the td-range of 0 mm to 138 mm (Fig. 3.4)[21, 57]. A more detailed description of the MD phantom design and application is presented in the results chapter 4.1.

Figure 3.4 – Multi-disk (Defrise) phantom made from Plexiglas®. The phantom features 8 disks of 3 mm thickness to be ﬁlled with a solution of distilled water and 99mTc. The phantom can be taken apart as illustrated for cleaning and possible future mod- iﬁcation purposes.

✶✼ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

The "optimal" number of views and the "optimal" table displacement were selected based on visual inspection of high count images (over 10 million counts for high statistical quality) of the star and the multi-disk phantom reconstructed with the TeraTomo™ reconstruction algorithm (settings: no pre-ﬁlter, default regularization predeﬁned by the manufacturer, matrix size: 128 x 128, pixel size: 1.72 mm x 1.72 mm, effective number of iterations: 90, number of subsets: 1 to 3, CT-based attenuation and scatter correction, Monte Carlo quality: low, correction for radioactive decay during the scan).

Figure 3.5 – Anthropomorphic striatum phantom. The brain is made from PVC plastic and features 4 anatomi- cally shaped chambers for the striatum, i.e. putamen left/right and caudate left/right, plus a large chamber for the brain background. All chambers can be individually ﬁlled with distilled water and the desired radioactive marker. The skull is made from plaster, which is molded in epoxy resin to mimic realistic attenuation of γ-photons in bone and soft tissue.

Optimization of the reconstruction parameters for DAT SPECT in clinical routine was based on measurements of an anthropomorphic striatum phantom in Fig. 3.5. The phantom was asym- metrically filled with different radioactivity solutions to achieve striatal contrast-to-background activity ratio of 4:1 and 3:1 for left and right striatum, respectively. Caudate and putamen in one hemisphere were filled with the same radioactivity solution. Radioactivity concentration was chosen to represent a typical patient scan in clinical routine (left caudate and putamen: 37.6 kBq/ml 99mTc at the time of acquisition, right caudate and putamen: 28.2 kBq/ml, background: 10 kBq/ml). The acquisition was performed with 30 gantry positions (nv = 90) each of 60 s duration resulting in a total acquisition duration of 30 min which also is typical for clinical routine. The following parameters of the Tera-Tomo™ reconstruction were tested: Monte Carlo Quality (MCQ): low, medium, high; effective number of iterations: 15, 33, 48, 66, 90, 150, 300, 498; number of subsets: 1, 2, 3, 5. All other parameters were fixed at the values recommended for DAT SPECT by the manufacturer: no pre-filter, default regularization, matrix size: 128 x 128, pixel size: 1.72 mm x 1.72 mm. Post reconstruction attenuation correction was performed according to Chang[54].

The "optimal" reconstruction parameters were selected based on visual inspection of the reconstructed images of the anthropomorphic striatum phantom and on the contrast recovery coef- ﬁcient CRC of the putamen, which in case of PD is most strongly effected by reduced striatal

✶✽ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

DAT availability[5, 6]. The CRC was computed as

measured putamen-to-background ratio 1 CRC = − . (3.1) true putamen-to-background ratio 1 −

A caudate putamen

50 mm B R L

reference region right putamen left putamen

Figure 3.6 – A: Orthogonal CT slices (axial, sagital, coronal) through the anthropomorphic striatum phantom used to optimize image reconstruction settings for DAT SPECT in clinical routine. B: Large predeﬁned regions of interest for hottest voxels analysis of right (red) and left (yellow) putamen and for the reference region (blue) used as background for computation of the putaminal contrast recovery coefﬁcient in the reconstructed SPECT images.

The activity concentration in the unilateral putamen was obtained by hottest voxels analysis in two large putamen regions-of-interest (ROIs) that are assumed to fully cover the putamen even under a small variation in phantom positioning. Putamen ROIs are predefined in a high resolution CT image (slice thickness = 0.6 mm) of the striatum phantom (Fig. 3.6 A) [32, 33, 58]. The number of hottest voxels to be averaged in the large putamen ROI was fixed to a total volume of 4.8 ml for unilateral putamen corresponding to the volume of the putamen cavity as determined in the CT of the striatum phantom. The background was determined in a large reference region comprising the whole brain (excluding the striatum and spouts for phantom filling) with a safety margin to the edge of the brain in order to avoid partial volume effects (Fig. 3.6 B).

CRC analysis was performed fully automatically using a custom-made MatLab script that ﬁrst registered the SPECT image to the high resolution CT and then applied the predeﬁned ROIs. Processing of reconstructed SPECT data is done using the freely available MatLab toolkit

✶✾ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

"SPM12" released in October 2014 designed for the analysis of brain imaging data sequences (https://www.fil.ion.ucl.ac.uk/spm/, accessed on 14.08.2019). SPM12 is a very powerful tool that can display, realign, reslice and co-register tomographic images (and much more). File format conversion of the reconstructed SPECT data from ".dcm" (dicom) format to ".nii" (nifti) format is realized by implementation of the freely available "xiangruili/dicom2nii"-converter in version 2018.08.08 released by Xiangrui Li on the MathWorks® file exchange platform (https://de.mathworks.com/matlabcentral/fileexchange/42997-xiangruili-dicm2nii, accessed on 14.08.2019)

The MatLab script for ROI analysis performs the following steps: First the 3-dimensional ROI mask is loaded to define the VOI, i.e. the sum of ROIs over all slices. Then the script prompts the user to select one or multiple reconstructed dicom images to be analyzed. For each dicom image a new directory is created and the file is copied over before being converted to the nifti file format. In order to improve co-registration results the image is cropped to remove the "cold" air around the head that does not contain any hot voxels. The cropped image is then co-registered to find the best possible overlap with a high resolution CT image of the head that has been extracted from Fig. 3.6 (A) by manually removing the brain from the skull, which is not visible in the functional SPECT image. After co-registration of the SPECT image, Chang attenuation correction is applied. The broad beam attenuation coefficient used for Chang correction is µ(x,y,z)= 0.12 1/cm for the brain volume. Finally, the attenuated SPECT image is analyzed slice-by-slice using the ROI mask to determine the average voxel values of the hottest voxels making up a volume of 4.8 ml for unilateral putamen. The results of all analyzed images in one batch are saved in a .txt file. To allow a visual comparison of images acquired with a different number of total counts using the same color map, the SPECT images are scaled by dividing each voxel value with the average voxel value of the background VOI.

✸✳✹ ❈♦✉♥ ❡♥✐✐✈② ❛♥❞ ♣❛✐❛❧ ❡♦❧✉✐♦♥

The development of a procedure for evaluation of count sensitivity and spatial resolution for MPH SPECT was an iterative process and partially effected and also was effected by other phantom measurements. A generalized description as a proposal to extend the NEMA NU 1 standard[39] including a more detailed description of the analysis of experimental data is therefore presented as part of the results in chapter 4. In this section the focus is on the experimental methods undertaken to compare the speciﬁc MPH collimators to the established parallel-hole and fan-beam collimators, while verifying some known effects on sensitivity and resolution associated with varying acquisition parameters.

The 3-dimensional system count sensitivity proﬁle of the triple-head SPECT system with the MPH collimators was measured with a 99mTc "point source" with an activity of 5 MBq to 15 MBq in a pipetted volume of 5 µl 0.1 µl sealed in an Eppendorf Tube®. The droplet of ±

✷✵ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞ radioactive solution was placed on the COR-axis and measured across the entire axial FOV in increments of 1 cm. The radial proﬁle was acquired at the axial sensitivity peak along an axis intersecting with the COR-axis. A full SPECT acquisition was performed for each localization of the point source using the acquisition parameters optimized for clinical DAT SPECT as described in section 3.3. The total number of counts acquired during the SPECT acquisition was obtained by summing the counts over all views. Uniformity correction was turned on. Dead time correction was negligible since count rate was well below 5,000 counts per second (cps) per detector head throughout all measurements. According to documentation of the acceptance test of the system, the maximum count rate for each detector head is around 700,000 cps.

The system count sensitivity sen at the localization of the point source was obtained by ﬁrst calculating the available activity A0 at the beginning of the measurement according to formula 2.3 and by inserting and rearranging formula 2.4 and solving the integral with t0 = 0 and t1 = tacq, where tacq is the total acquisition time, usually 300 s for every position of the point source in each measurement.

N sen = total (3.2) A t1 e λtdt 0 t0 − R N λ sen total (3.3) = λt A0 (1 e acq ) − − As a consequence every measurement of a full sensitivity proﬁle across the entire FOV takes multiple hours to complete with the advantage of a very good signal-to-noise-ratio (SNR) that enabled converging iterative reconstruction of the point source even in areas of low sensitivity.

In order to measure reconstructed spatial resolution and its variability over the FOV from a clinical perspective, each point source SPECT acquisition was reconstructed by the Tera-Tomo™ algorithm with the parameter settings optimized for clinical DAT SPECT. Photon attenuation can be neglected for point sources due to the low density sample holder and small volume. Analysis of reconstructed point source SPECT data has been automated to work in batch mode using MatLab. Horizontal and vertical spatial resolution were obtained by fitting a Gaussian function to one-dimensional activity profiles obtained by first summing all transaxial image matrices and then summing over all rows or over all columns of the sum matrix. The mean of horizontal and vertical full-width-at-half-maximum (FWHM) of the fitted Gaussian was used to characterize reconstructed spatial resolution at the location of the point source.

Since the level of complexity in the image effects the convergence of an iterative reconstruction method a "hot-rod" phantom with diameterrod = distancerod rod was also measured placed − on the COR-axis and centered at the sensitivity proﬁle using the same acquisition parameters. Reconstruction of the hot-rod phantom measurement is performed both with clinical reconstruction parameters (same as point source) and with 498 iterations that are expected to be more representative regarding the level of convergence compared to point source measurements.

✷✶ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

Influenced by the findings regarding helical acquisition, count sensitivity and spatial resolution profile of the triple-head SPECT system with MPH collimators were measured for different values of total table displacement td (td = 0, 40 and 80 mm) for differently pitched helical acquisition modes. In addition, peak sensitivity was measured for td = 0, 20, 40, 60, 80, 100, 120, 136 mm.

For comparison, point source measurements for assessment of system sensitivity and reconstructed spatial resolution were also performed with the same triple-head camera equipped with LEHRHS collimators. The acquisitions with LEHRHS collimators were performed in double-head mode, because minimum radius of rotation in triple-head mode is 21 cm, which is known to result in strongly reduced spatial resolution and contrast recovery[59]. This relation has also been veriﬁed in measurements with varying acquisition radius of a point source placed in the axial and radial center of the FOV using three different conventional collimators: LEHR, LEHRHS and Fanbeam.

Acquisition parameters for the LEHRHS and LEHR collimator were as follows: total number of views nv = 120 (60 per head) resulting in 3° angular sampling, scan arc 180°, matrix size 128 x 128, pixel size = 2.43 mm. Reconstruction parameters had previously been optimized for DAT SPECT with LEHRHS collimators in double-head mode and were as follows: filtered backprojection with Butterworth filter of order 6 and 0.55 cycles per pixel cutoff frequency, no post-filtering, matrix size 128 x 128 and pixel size 2.43 mm.

Finally, point source measurements for assessment of system sensitivity and spatial resolution were also performed with a double-head SPECT/CT system equipped with fan-beam collimators of 41 cm focal length (Siemens Symbia TruePoint, Siemens Healthineers, Erlangen, Ger- many). A total number of 120 views was acquired (60 per head) for angular sampling of 3°. Projection data were sampled into a matrix of 128 x 128 pixels with 3.90 mm edge length. The scan arc was 180°. SPECT images were reconstructed into transaxial cross-section images using the ordered-subsets-expectation-maximization algorithm with resolution recovery implemented in the HybridRecon-Neurology tool of the Hermes SMART workstation v1.6. The following reconstruction parameters are used as recommended for DAT SPECT by Hermes: 128 x 128 pixels of 1.95 mm edge length, 16 iterations, 5 subsets, that is, 16 x 5 = 80 effective iterations, resolution recovery with a Gaussian model (hole diameter 1.11 mm, hole length 2.41 cm, detector resolution 3.8 mm, radius of rotation offset 3.21 cm), and post-ﬁltering with 3-dimensional Gaussian kernel of 7 mm FWHM.

The radius of rotation was 140 mm for all SPECT acquisitions independent of the SPECT system (Mediso AnyScan® TRIO with MPH (triple-head mode) and LEHR/LEHRHS (double- head mode) and the double-head system Siemens Symbia TruePoint with fan-beam collimators. Neither attenuation correction nor scatter correction was performed. In order to test for potential differences in system count sensitivity of the triple-head SPECT with MPH collimators between

✷✷ ❈❤❛♣❡ ✸✳ ▼❛❡✐❛❧ ❛♥❞ ♠❡❤♦❞

99mTc and 123I, count sensitivity was measured also with a 123I point source at several locations in the FOV.

✸✳✺ ❱✐✉❛❧ ✐♠❛❣❡ ✉❛❧✐② ❛♥❞ ❝♦♥❛ ❡❝♦✈❡② ❛ ❢✉♥❝✐♦♥ ♦❢ ❤❡ ♦❛❧ ❝♦✉♥

The anthropomorphic striatum phantom with about 4:1 (left striatum) and 3:1 (right striatum) true radioactivity concentration contrast, representative of a typical patient study (left caudate and putamen: 37.6 kBq/ml, right caudate and putamen: 28.2 kBq/ml, background: 10 kBq/ml), was scanned with scan duration ranging from 2 minutes to about 2 hours using both the triple- head system with MPH collimators and aperture APT71 (optimized acquisition and reconstruction parameters) and the double-head system with fan-beam collimators (acquisition and reconstruction parameters as described in the previous subsection).

Post reconstruction Chang attenuation correction with broad-beam attenuation coefﬁcient µ = 0.12 1/cm was performed for all images (independent of the SPECT camera) using a custom- made MATLAB script in order to avoid bias by differences with respect to the attenuation correction method. No scatter correction was performed.

In order to investigate differences regarding reconstruction artifacts for very low-count images between the iterative reconstruction algorithms TeraTomo™ (Mediso) and HybridRecon- Neurology tool (Hermes) a histogram of low- and highcount SPECT images for both the Stria- tum MPH SPECT as well as the fan-beam SPECT system was assessed.

✸✳✻ ❛✐❡♥ ❝❛♥

A male patient (54 years old) referred to DAT SPECT for suspicion of PD in the early clinical phase was scanned twice after a single intravenous injection of 182 MBq 123I-FP-CIT. The ﬁrst scan of 40 min duration started 3h 10min after tracer injection and used the triple-head camera equipped with LEHRHS collimators in double-head mode. The second scan of 30 min duration started 5h 7min after tracer injection and used the same triple-head camera equipped with MPH collimators in triple-head mode. Acquisition and reconstruction parameters with LEHRHS were as described above. Acquisition and reconstruction parameters with MPH were as optimized for DAT SPECT by the phantom measurements. Attenuation correction was performed by Chang’s method implemented in the system software.

✷✸ ✹ ❘❡✉❧

The results presented in this chapter are arranged in the order of there respective section in the chapter 3 "Materials and Methods". Some results are strongly interconnected, e.g. the position of the axial count sensitivity peak from the point source measurements determines the position of the striatum phantom, which was used to optimize reconstruction parameters also to reconstruct the point source measurements themselves. Moreover, knowledge of the expected spatial resolution got carried into the design of the multi-disk phantom presented in this section as part of the results.

✹✳✶ ❖♣✐♠✐③❛✐♦♥ ♦❢ ❛❝✉✐✐✐♦♥ ❛♥❞ ❡❝♦♥✉❝✐♦♥ ♣♦♦❝♦❧

❆♥❣✉❧❛ ❙❛♠♣❧✐♥❣ Reconstructed images of the star phantom with varying angular sampling are shown in Fig. 4.1 (B, D). The star phantom consists of a styrofoam structure placed in a plastic container in the shape of a truncated cone, therefore the diameter of the phantom in the two sections is not equal (ﬁne structure has a sligthly smaller diameter). Raw data was provided by Attila Forgács, who acquired the data at Scanomed Nuclear Medicine Center De- brecen (Debrecen, Hungary) on an identical SPECT system with striatum MPH collimators. Before this data was available it has been recommended by the manufacturer to perform MPH

SPECT with 4° angular sampling. Unfortunately, nv = 90 resulting in 4° angular sampling has not been tested with this phantom. However, visual inspection of the reconstructed images, show no signiﬁcant (clinical relevant) difference between nv = 72 and nv = 96. It has been decided to keep nv = 90 as the "optimal" total number of views.

✷✹ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Figure 4.1 – Transaxial CT slices (A, C) through the two sections of a custom-made star phantom and transaxial striatum MPH SPECT images obtained with varying total number of views nv (B, D). Color chart is scaled to the voxel of maximum intensity for each transaxial cross-section

▼✉❧✐✲♣❧❡①✐♥❣ The projection pattern on the gamma detector of a large cylinder phantom with a homogeneous activity concentration placed on the COR-axis and centered in the FOV is presented in Fig. 4.2. The measured interference pattern is in good agreement with the system design illustrated in the technical drawing in Fig. 3.2 and conﬁrms the existence of multi-plexing. The intensity gradient (top to bottom) results from the shifted FOV in inferior direction. As a result the projections of the superior pin-holes are under a steeper angle covering more radioactive volume of the cylindrical shaped phantom.

max

Figure 4.2 – Interference pattern of 20 projections from 20 pinholes on the gamma detector (RAW image) from a static acquisition of a large homogeneuous cylinder phantom using one detector head and the striatum MPH collimator. Areas of multi-plexing highligthed purple.

✷✺ ❈❤❛♣❡ ✹✳ ❘❡✉❧

❆①✐❛❧ ❙❛♠♣❧✐♥❣ ✲ ❍❡❧✐❝❛❧ ❙❝❛♥♥✐♥❣ In order to study effects of multi-plexing in MPH SPECT, i.e. overlapping projections on the detector, the main goal of phantom design was to find a geometry that maximizes overlap and therefore results in an incomplete axial sampling. In order to isolate artifacts that result from incomplete axial sampling from others resulting from low sensitivity, it was desirable to study axial sampling artifacts depending on the position in the FOV, i.e. the position in the sensitivity profile. The only geometric shape with rotational symmetry meeting these criteria are disks, lined up on the COR-axis. This type of geometry is also known as a Defrise phantom and has been described as the most challenging scenario for MPH collimators with multi-plexing, because overlap in the projection data makes it difficult for the reconstructed image to converge[21]. In this thesis the phantom will be referred to as multi-disk (MD) phantom.

Overlapping projections Overlapping projections FOV FOV

COR COR

td disk disk (a) circular acquisition (b) helical acquisition

Figure 4.3 – Axial sampling of a thin disk positioned on the COR-axis for circular (Part (a)) and helical (Part (b)) acquisition mode. Table displacement td during helical acquisition extends the axial FOV. The areas of overlapping projections are highlighted. Drawing is intentionally simpliﬁed for axial projection overlap on the detector. True overlap is transaxially between 2nd and 3rd order pin-holes as described in section 3.2.

As we know from Fig. 4.2 we do experience multi-plexing for the striatum MPH collimator in two regions due to the symmetrical design of the collimator. When a single disk phantom with a sufﬁciently large diameter is placed on the COR-axis it can be assumed that a circular acquisition orbit will result in incomplete axial sampling as well. Due to the rotational symmetry of the disk the areas of multi-plexing would stay at the same location on the detector over the full acquisition orbit, leaving a gap in the axial sampling as illustrated in Fig. 4.3a. A table displacement during acquisition as realized in a helical scanning mode distributes the overlapping projections over many more projection angles which is expected to reduce multi-plexing artifacts in the reconstructed image (Fig. 4.3b).

✷✻ ❈❤❛♣❡ ✹✳ ❘❡✉❧

5 4

Figure 4.4 – Explosion view of the multi- disk phantom showing the following com- 2 ponents: (1) top part, (2) bottom part, (3) thick disk, (4) spacer ring, (5) container tube. CAD design using software "Autodesk In- ventor Professional 2019". Technical draw- ings for each component can be found in Fig. 4.5 and Fig. 4.6.

✷✼ ❈❤❛♣❡ ✹✳ ❘❡✉❧

(3 mm) was chosen to be in the same range as the expected spatial resolution when using the MPH collimator system as recommended by Van Audenhaege et al.[12]. Precise disk thickness of 3 mm is realized by 7 mm spacer rings (Part (4), Fig. 4.5) that are recessed 2 mm in both adjacent disk separators (Part (3), Fig. 4.5).

Part 1 Part 2

A-A ( 1 : 2 ) B-B ( 1 : 2 )

7 1 5 mm

m m 110 P

130 mm 0 mm P P13

AA BB

Part 3 Part 4

m C-C ( 1 : 2 ) D-D ( 1 : 2 )

7 2 5 mm

mm 5 m P 130 m 0 mm P P13

P12 0 mm CC DD

Figure 4.5 – Construction view of the individual components including cross-sections and dimensions: Part (1) is the top part featuring the main socket. Part (2) is the bottom part featuring a secondary socket to simplify draining of the phantom or ﬁlling the phantom from the bottom up. Part (3) are the disk separators that the determine disk distance and lock in on Part (4). Part (4) are the spacer rings that determine disk thickness.

✷✽ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Multi-disk Phantom

E-E ( 1 : 2 ) E 5 mm 5 mm

E 15 mm 130 mm

Figure 4.6 – Full construction view of the assembled multi-disk phantom and cross-section E-E. The dimensions have been selected based on the available material as well as the dimensions of the head holder of the SPECT system, which is necessary to position the phantom precisely on the COR-axis.

A main concern of this design was to ensure equally and homogeneously filled disks. To ensure an even distribution of the 99mTc and distilled water solution in the phantom, the entire container was drained before measurements and the new activated solution was mixed outside the phantom prior to filling. Handling and mixing radioactive isotopes was done in a special certified laboratory using proper shielding. Air bubbles on the inside of the phantom needed to be removed, since they result in a lower then expected total activity in a particular disk and may create reconstruction artifacts when CT-based attenuation correction is applied. The following measures have been taken to ensure a bubble free phantom filling: The phantom is thoroughly washed with dish-washing detergent before filling to remove any dust leaving a smooth surface behind. Rinse agent, also commonly used for dishwashers, is added to the distilled water (1 ml rinse agent : 250 ml distilled water) to reduce the surface tension of water such that less bubbles are build up and/or are easier to move to the exit. Ethanol also reduced the surface tension but should not be left in the phantom for a long time since it reacts with the partially glued Plexiglas® components.

In a first test run, it has been verified, that the phantom is sufficiently large to generate multi- plexing on the detector as can be seen by overlapping projection in the RAW image (Fig. 4.7).

✷✾ ❈❤❛♣❡ ✹✳ ❘❡✉❧

max

Figure 4.7 – Interference pattern of projections on the detector (RAW image) from multi-disk phantom measurements with the striatum MPH collimator. Areas of multi-plexing highligthed purple.

The "optimal" table displacement was chosen based on three evaluation methods: First the reconstructed images of the multi-disk phantom with variable table displacement during helical scanning were evaluated visually (Fig. 4.8). Secondly, axial proﬁles of the reconstructed Defrise phantom with variable table displacement were created by summing all counts in each transaxial slice (Fig. 4.9). The proﬁles were also assessed visually.

td = 0 mm td = 20 mm td = 40 mm td = 60 mm

180 mm

20 mm FOV max 3 mm

td = 80 mm td = 100 mm td = 120 mm td = 138 mm

0 FOV

Figure 4.8 – Reconstructed SPECT images of the self-made MD phantom acquired with striatum MPH collimator and different helical scanning modes. The images become larger because the axial FOV length is increased due to variable table displacement (td) during helical acquisition. Re- construction performed using optimized clinical parameters and CT-based attenuation and scatter correction. Sensitivity reduction due to displacements compensated by color chart.

✸✵ ❈❤❛♣❡ ✹✳ ❘❡✉❧

td0 td20 td40 td60

1 1 1 1

0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6

rel. intensity rel. 0.4 intensity rel. 0.4 intensity rel. 0.4 intensity rel. 0.4

0.2 0.2 0.2 0.2

0 0 0 0 50 100 150 50 100 150 50 100 150 50 100 150 pixel pixel pixel pixel

td80 td100 td120 td138

1 1 1 1

0.8 0.8 0.8 0.8

0.6 0.6 0.6 0.6

rel. intensity rel. 0.4 intensity rel. 0.4 intensity rel. 0.4 intensity rel. 0.4

0.2 0.2 0.2 0.2

0 0 0 0 50 100 150 50 100 150 50 100 150 50 100 150 pixel pixel pixel pixel

Figure 4.9 – Intensity z-proﬁles of MD phantom measurements. Data obtained by intensity projection along COR-axis and scaling on individual intensity maximum

Thirdly, a quantitative assessment using a hottest-voxel analysis in the four central disks around the axial sensitivity peak of the striatum MPH system was performed (disk # 4 to # 7, Fig. 4.10a). The coefﬁcient of variance CoV, deﬁned as the ratio of the standard deviation σ to the mean µ of the hottest voxel intensity in each individual disk ROI, was determined:

σ CoV = (4.1) µ

The averaged CoV for disk 4 to 7 was calculated. As shown in Fig. 4.10b the CoV is decreasing up to td = 60 mm, from which no further improvement for higher td can be identiﬁed. Consid- ering all quantitative and qualitative analysis of the MD phantom measurements the minimum total table displacement required to avoid clear image artifacts (blurriness) in the four central disks around the axial sensitivity peak of the MPH system (disk # 4 to # 7, Fig. 4.10a) and to avoid heterogeneity of peak height of those disks in the axial proﬁle (Fig. 4.9) was td = 40mm. Considering the decreased peak count sensitivity for td 40mm as shown in Fig. 4.17, it has ≥ been decided that td = 40mm is the "optimal" table displacement in clinical application.

✸✶ ❈❤❛♣❡ ✹✳ ❘❡✉❧

ROI analysis 0.35 Disk 4 to 7 0.30 ] . u max . a 0.25

0.20

0.15

0.10

0 Coefﬁcient of Variance [ 0.05

0.00 0 20 40 60 80 100 120 140

Disk_4 Disk_5 Disk_6 Disk_7 table displacement [mm] (a) ROI analysis of disk # 4 to # 7 (b) Coefﬁcient of Variance of disk # 4 to # 7

Figure 4.10 – Part (a): Fusion view of MD phantom SPECT image (td = 80mm) with regions of interest (ROI) used for hottest-voxel analysis. Oversized ROIs account for limited resolution (partial volume effects) and reconstruction artifacts due to multi-plexing. Part (b): Coefﬁcient of variance based on hottest voxel ROI analysis.

✸✷ ❈❤❛♣❡ ✹✳ ❘❡✉❧

A secondary ﬁnding has been made when comparing reconstructed SPECT images of the MD phantom without attenuation correction (NAC) to images with CT-based attenuation and scatter correction (CT-ACSC) as shown in Fig. 4.11. For normal contrast with full dynamic range, the disks in the NAC image seem more blurry compared to the CT-ACSC image. The high contrast image with reduced dynamic range reveals that due to reconstruction artifacts some activity is lost to areas in between disks for the reconstruction without attenuation correction. This effect seems suppressed in the reconstruction with CT-based attenuation and scatter correction. Hence, the TeraTomo™ algorithm might be using the CT image during iterative reconstruction to ﬂag certain areas as "free of activity". NAC NAC

max max

0.2max

0 0

CT-ACSC CT-ACSC

Figure 4.11 – SPECT image with striatum MPH collimator of the MD phantom. Normal contrast and high contrast image of the same measurement (td = 80 mm) and same reconstruction parameters to show the difference between no attenuation correction (NAC) and with CT-based attenuation and scatter correction (CT-ACSC).

✸✸ ❈❤❛♣❡ ✹✳ ❘❡✉❧

❈❧✐♥✐❝❛❧ ❘❡❝♦♥✉❝✐♦♥ ❛❛♠❡❡ Results of the MPH SPECT measurements of the anthropomorphic striatum phantom for optimization of the Tera-Tomo™ reconstruction parameters are summarized in Fig. 4.12, Fig. 4.13 and Fig. 4.14. The reconstruction was performed on a dedicated workstation featuring state-of-the-art hardware components: Intel® Xeon® W- 2133 CPU @ 3.60 GHz, 32 GB DDR4 RAM, Nvidia GeForce RTX 2080. Based on these results, the following values were selected as "optimal" compromise between image quality and reconstruction time: low Monte Carlo quality, 90 effective iterations, 3 subsets.

Low Medium High 5:42 min 8:43 min 14:47 min

R L

A 0.700.76 0.70 0.76 0.69 0.76 4

1ss 2ss 3ss 5ss 14:23 min 7:78 min 5:42 min 3:58 min

B 0.70 0.77 0.70 0.76 0.700.76 0.68 0.74

Figure 4.12 – Central transversal slice through the striatum in the MPH SPECT of the anthropomorphic striatum phantom with true 3.76:1 (left striatum) and 2.82:1 (right striatum) contrast acquired with a total number of 4.7 million counts (30 min acquisition time) representative of a patient scan. Images were reconstructed using the TeraTomo™ algorithm with (A) variation of Monte Carlo Quality (MCQ): low, medium, high (at 66 effective iterations, 3 subsets) and (B) variation of the number of subsets (ss): 1, 2, 3, 5 (at 66 effective iterations (or 65, for 5 subsets), low MCQ). Post-reconstruction attenuation correction was performed using Chang’s method, no scatter correction was applied. The contrast recovery coefﬁcient for left and right putamen as well as the time required for reconstruction are speciﬁed with each slice

✸✹ ❈❤❛♣❡ ✹✳ ❘❡✉❧

15it 33it 48it 66it 1:28 min 2:56 min 4:18 min 5:42 min

R L

0.59 0.66 0.65 0.71 0.68 0.73 0.70 0.76 4

90it 150it 300it 498it 7:40 min 12:42 min 30:23 min 50:22 min

0.72 0.78 0.76 0.82 0.82 0.86 0.88 0.91

Figure 4.13 – Central transversal slice through the striatum in the MPH SPECT of the anthropomorphic striatum phantom with true 3.76:1 (left striatum) and 2.82:1 (right striatum) contrast acquired with a total number of 4.7 million counts (30 min acquisition time) representative of a patient scan. Images were reconstructed using the TeraTomo™ algorithm with variation of the number of effective iterations (it): 15, 33, 48, 66, 90, 150, 300, 498 (at low MCQ, 3 subsets). Chang’s attenuation correction is applied to all measurements after reconstruction, no scatter correction was applied. The contrast recovery coefﬁcient for left and right putamen as well as the time required for reconstruction are speciﬁed with each slice.

✸✺ ❈❤❛♣❡ ✹✳ ❘❡✉❧

A B Reference MPH_PL MPH_PR Reference MPH_PL MPH_PR 10.0 1.00 10.0 1.00 9.0 0.90 9.0 0.90 8.0 0.80 8.0 0.80 7.0 0.70 7.0 0.70

6.0 0.60 6.0 0.60

SNR SNR

5.0 0.50 5.0 0.50

CRC [a.u.] CRC [a.u.] CRC 4.0 0.40 4.0 0.40 3.0 0.30 3.0 0.30 2.0 0.20 2.0 0.20 low1 medium 2 high 3 1 2 3 4 5 Monte-Carlo-Quality number of subsets

C Reference MPH_PL MPH_PR 10.0 1.00 9.0 0.90 8.0 0.80 7.0 0.70

6.0 0.60

CRC SNR 5.0 0.50 4.0 0.40 3.0 0.30 2.0 0.20 0 100 200 300 400 500 number of iterations

Figure 4.14 – Quantitative analysis of the striatum MPH SPECT images shown in Fig. 4.12 and Fig. 4.13. The contrast-recovery-coefﬁcient (CRC, right y-axis) for left (PL) and right (PR) putamen (right y-axis) was computed as described in section 4.1. The signal-to-noise (SNR, left y-axis) ratio of the voxel intensity in the reference ROI (Fig. 3.6 B) was computed as the ratio of the mean to the standard deviation of the voxel intensity in the reference ROI. CRC and SNR results are shown for varying Monte Carlo quality: low / medium / high (A, 66 effective iterations, 3 subsets), varying number of subsets: 1, 2, 3, 5 (B, low Monte Carlo quality, 90 effective iterations) and varying effective number of iterations: 15, 33, 48, 66, 90, 150, 300, 498 (C, low Monte Carlo quality, 3 subsets).

In summary, "optimal" acquisition and reconstruction parameters for DAT SPECT with the triple-head SPECT system with MPH collimators and Tera-Tomo™ reconstruction were: nv = 90, td = 40mm, low Monte Carlo quality, 90 effective iterations, 3 subsets.

CT-based attenuation and scatter correction during iterative Tera-Tomo™ reconstruction resulted in improved image quality compared to post-reconstruction Chang attenuation correction without scatter correction (Fig. 4.15).

✸✻ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Figure 4.15 – Central transversal slice through the striatum in MPH SPECT of the anthropomorphic striatum phantom with true 3.76:1 (left striatum) and 2.82:1 (right striatum) contrast acquired with a total number of 4.7 million counts representative of a patient scan (A, B) compared to a high count measurement with 17.5 million counts (C, D). All images were acquired and reconstructed with the "optimal" clinical Tera-Tomo™ settings (nv = 90, td = 40mm, low Monte Carlo quality, 90 effective iterations, 3 subsets). In A and C, Chang’s method was used for post-reconstruction attenuation correction, no scatter correction. In B and D, CT-based attenuation and scatter correction was performed during Tera-Tomo™ reconstruction. Part E shows a simulation of a DAT SPECT image with the same striatum-to-background contrast, 3 mm spatial resolution and no noise (obtained by segmenting striata and background in the high resolution CT (Fig. 3.6 A), setting pixel values to 1, 3.76 and 2.82 for background, left striatum, and right striatum respectively, and then filtering the image with a Gaussian kernel of 3 mm FWHM). The contrast recovery coefficient for left and right putamen is specified with each slice.

✹✳✷ ❈♦✉♥ ❡♥✐✐✈② ❛♥❞ ♣❛✐❛❧ ❡♦❧✉✐♦♥

Proﬁles of system count sensitivity in axial direction along the COR-axis and in radial direction at the position of the axial sensitivity peak of the triple-head system with the striatum MPH and Alzheimer MPH collimator and the double-head system with fan-beam and LEHRHS collimators are shown in Fig. 4.16. Peak system sensitivity of the triple-head SPECT with striatum MPH collimators was 608 5cps/MBq and 462 4cps/MBq for the Alzheimer aperture com- ± ± pared to 223 2cps/MBq for the double-head fan-beam system and 187 2cps/MBq for the ± ± double-head LEHRHS system. Sensitivity of the MPH system decreased towards the edges of the FOV.

✸✼ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Axial sensitivity - collimator overview 700 Striatum MPH Alzheimer MPH 600 fan-beam LEHRHS

500 ] MBq

/ 400 cps

300 sensitivity [ 200

100

0 240 200 160 120 80 40 0 40 80 120 160 200 240 − − − − − − relative position on COR-axis [mm]

Radial sensitivity - collimator overview 700 Striatum MPH Alzheimer MPH fan-beam LEHRHS 600

500 ] MBq

/ 400 cps

300 sensitivity [ 200

100

0 0 20 40 60 80 100 radial distance to center of rotation [mm]

Figure 4.16 – System count sensitivity proﬁle of the triple-head SPECT system with the MPH collimators with Striatum and Alzheimer aperture compared to the double-head SPECT systems with fan-beam and LEHRHS collimators in axial direction along the center of rotation axis (top) and in radial direction at the maximum of the axial sensitivity proﬁle (bottom).

✸✽ ❈❤❛♣❡ ✹✳ ❘❡✉❧

The impact of varying table displacement for helical scanning on the axial peak sensitivity and reconstructed spatial resolution for the striatum MPH collimator is shown in Fig. 4.17. Additionally a full sensitivity proﬁle over the entire axial FOV for selected table displacements has been acquired for the striatum MPH (Fig. 4.18) and the Alzheimer collimator (Fig. 4.19).

Peak sensitivity and resolution - Striatum MPH Figure 4.17 – Peak system sensitivity 700 3.5 (left axis) and reconstructed spatial resolution at peak location (right axis) of the 600 3.0 triple-head SPECT with striatum MPH

] collimators for varying total table dis- 500 2.5 ]

MBq placement during helical acquisition. /

400 2.0 mm cps 300 1.5 FWHM [ 200 1.0 sensitivity [

100 0.5 sensitvity resolution 0 0.0 0 20 40 60 80 100 120 140 table displacement during acquisition [mm]

Peak count sensitivity of the triple-head system with the striatum MPH collimators measured with 123I was around 2 % lower compared to using 99mTc for acquisitions regardless of table displacement for helical scanning. 123I is not produced on site and was not possible to obtain in activity concentrations comparable to 99mTc. Therefore, the 123I sample volume was 10 times larger at 50 µl.

✸✾ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Axial sensitivity - Striatum MPH Figure 4.18 – Full axial count sensitivity pro- 700 ﬁle of striatum MPH collimator for selected td0 table displacements td during helical acquisi- 600 td40 tion. td80 ] 500 MBq / 400 cps 300

200 sensitivity [

100

0 160 120 80 40 0 40 80 120 160 − − − − position on COR-axis [mm]

Axial sensitivity - Alzheimer MPH Figure 4.19 – Full axial count sensitivity pro- 700 ﬁle of Alzheimer MPH collimator for selected td0 table displacements td during helical acquisi- 600 td40 tion. ] 500 MBq / 400 cps 300

200 sensitivity [

100

0 160 120 80 40 0 40 80 120 160 − − − − position on COR-axis [mm]

Reconstructed spatial resolution of MPH SPECT based on the same point source measurements and regardless of aperture with the acquisition and reconstruction parameters optimized for clinical DAT SPECT was 3 mm to 4 mm FWHM throughout the entire FOV (Fig. 4.20). Spatial resolution of fan-beam SPECT and LEHRHS SPECT was about 9 mm FWHM and about 13.5 mm FWHM, respectively (Fig. 4.20).

✹✵ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Axial resolution - collimator overview 18 Striatum MPH Alzheimer MPH 16 fan-beam LEHRHS

12 ]

mm 10

8 FWHM [ 6

0 240 160 80 0 80 160 240 − − − relative position on COR-axis [mm]

Radial resolution - collimator overview 18 Striatum MPH Alzheimer MPH 16 fan-beam LEHRHS

12 ]

mm 10

8 FWHM [ 6

0 0 10 20 30 40 50 60 70 80 90 100 radial distance to center of rotation [mm]

Figure 4.20 – Reconstructed spatial resolution in axial direction (top) and radial direction (bottom) at the position of the axial sensitivity peak. MPH SPECT images were acquired and reconstructed with parameter settings optimized for DAT SPECT in clinical routine (nv = 90, td = 40, low Monte Carlo quality, 90 effective iterations, 3 subsets), fan-beam SPECT images were reconstructed with the OS-EM1 algorithm with resolution recovery implemented in the HybridRecon-Neurology tool of the Hermes SMART workstation v1.6 with parameter settings recommended for clinical DAT SPECT by Hermes.

✹✶ ❈❤❛♣❡ ✹✳ ❘❡✉❧

❍♦✲♦❞ ♣❤❛♥♦♠ The reconstructed spatial resolution for the MPH collimators derived from the FWHM of the point source measurements are in good agreement with the measurements of the hot-rod phantom when using the same optimized acquisition and reconstruction parameters (Fig. 4.21). A reconstruction with 498 iterations is able to slightly improve spatial resolution with the cost of added noise as presented in Fig. 4.21b. Noise is partially compensated by summing up the cross-sections over the entire phantom but is still visible for phantom section with 2 mm rods.

(a) optimized reconstruction parameters (b) highest reconstruction parameters

Figure 4.21 – Reconstructed image of the hot-rod phantom: rod diameter equals rod distance (2 mm to 7 mm, 1 mm-steps in between sections). Cross-sections over the entire phantom have been summed up to reduce noise. The image has been reconstructed using optimized parameters for patient scans according to section 4.1 in (a) and with highest possible parameters in (b) to show the best possible resolution for an image of this complexity.

❉❡❡❝♦ ❝♦♥✜❣✉❛✐♦♥ Whenever this thesis compares the clinical MPH SPECT performance to conventional collimator SPECT the reader will notice that this comparison entails 3-head operation for MPH SPECT versus 2-head operation for conventional collimator SPECT. To prove that this comparison is actually in favor of the conventional collimator SPECT (from a clinical perspective), the dependence of system sensitivity and resolution on the acquisition radius r is investigated. When performing 2-head measurements with conventional collimators the minimum acquisition radius is only limited by the actual dimensions of the patient head or head holder (Fig. 4.22a), whereas for 3-head operation the minimum acquisition radius is 210 mm to prevent collision of detector heads (Fig. 4.22b). For MPH SPECT the acquisition radius is ﬁxed by the manufacturer to 140 mm in 3-head operation, which does not result in detector collision because of the collimator’s pyramid shape.

✹✷ ❈❤❛♣❡ ✹✳ ❘❡✉❧

(a) detector conﬁguration for 2-head operation (b) detector conﬁguration for 3-head operation

Figure 4.22 – Detector conﬁguration for 2-head and 3-head operation using conventional parallel- hole collimators. Minimum acquisition radius limited by head holder (a). Larger acquisition radius necessary to prevent detector-detector collision (b).

Count sensitivity results are presented in Fig. 4.23a and corresponding spatial resolution is presented in Fig. 4.23b for a selection of conventional collimators for variable acquisition radius. Due to the collimator geometry the intrinsic intensity (I) dependence I ∝ 1/r2, with r being the acquisition radius, is not represented in the sensitivity proﬁles: For parallel-hole collimators the 1/r2-dependence is compensated by the steeper incidence angle that allows penetration of radiation through more holes. For the fan-beam system the larger acquisition radius brings the source closer to the focal-point which overcompensates the reduction of intensity per unit area. Generally, spatial resolution decreases for larger acquisition radius. The reconstructed spatial resolution for the LEHRHS collimator at r = 140mm is 13.6 mm for the minimum acquisition radius necessary for 3-head operation (r = 210mm) the reconstructed spatial resolution is 15.9 mm.

1,200 24 LEHRHS LEHRHS 22 1,000 fan-beam fan-beam ] LEHR 20 LEHR ] MBq 800 / 18 mm cps 600 16 14

400 FWHM [ 12 sensitivity [ 200 10 0 8 100 150 200 250 300 350 400 100 150 200 250 300 350 400 acquisition radius [mm] acquisition radius[mm] (a) sensitivity (b) resolution

Figure 4.23 – Point source measurements with source on the COR-axis in the axial center of the FOV for different acquisition radius r. Sensitivity proﬁles for selected conventional collimators in (a) with corresponding reconstructed resolution in part (a).

✹✸ ❈❤❛♣❡ ✹✳ ❘❡✉❧

✹✳✸ ❱✐✉❛❧ ✐♠❛❣❡ ✉❛❧✐② ❛♥❞ ❝♦♥❛ ❡❝♦✈❡② ❛ ❢✉♥❝✐♦♥ ♦❢ ❤❡ ♦❛❧ ❝♦✉♥

Results of the MPH SPECT compared to fan-beam SPECT of the anthropomorphic striatum phantom for varying total number of counts are summarized in Fig. 4.24 and Fig. 4.25.

Figure 4.24 – Quantitative analysis of the SPECT images shown in Fig. 4.25. The CRC (A) for left (PL) and right (PR) putamen was computed as described in equation 3.1. The signal-to-noise (SNR, B) ratio of the voxel intensity in the background (BG) reference ROI (Fig. 3.6(B)) was computed as the ratio of the mean to the standard deviation of the voxel intensity in the reference ROI. CRC and BG-SNR results are given as function of the total number of acquired counts for striatum MPH and fan-beam measurements shown in Fig. 4.25

✹✹ ❈❤❛♣❡ ✹✳ ❘❡✉❧

0.3 m 0.6 m 1.6 m 3.4 m 2 min 4 min 10 min 22 min

R L

0.61 0.76 0.67 0.70 0.60 0.69 0.69 0.71 4

4.7 m 6.6 m 9.2 m 17.5 m 30 min 43 min 59 min 114 min

0.71 0.77 0.69 0.77 0.72 0.79 0.71 0.80

(a) MPH measurements

0.6 m 1.2 m 2.9 m 6 min 12 min 30 min

R L 0.61 0.62 0.59 0.58 0.49 0.62 4

5.9 m 8.2 m 11.4 m 63 min 87 min 123 min

0.52 0.60 0.54 0.61 0.53 0.60

(b) Fan-beam measurements

Figure 4.25 – Central transversal slice through the striatum of the anthropomorphic striatum phantom acquired with varying scan duration resulting in varying total number of counts (m = million). The images were acquired with the triple-head system with striatum MPH collimators (aperture APT71) in part (a) and the double-head system with fan-beam collimators in part (b). Image reconstruction performed as specified in section 3.5. Reconstruction attenuation correction using Chang’s method was performed with a custom-made MATLAB script in all cases. The contrast recovery coefficient for left and right putamen is specified for each slice. ✹✺ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Histograms of voxel intensity for selected low- and high-count SPECT images are presented in Fig. 4.26. Histograms of low-count images reveal different reconstruction artifacts for Ter- aTomo™ algorithm (striatum MPH) compared to Hermes OS-EM algorithm (fan-beam): The broad histogram (Fig. 4.26a) of low-count fan-beam image is expected for poor SNR. How- ever, "island forming" as in the low-count striatum MPH image in Fig. 4.25a is not expected, results in secondary peaks in the histogram (Fig. 4.26a) and may be declared as reconstruction artifacts. Since time-resolved measurements (not investigated in this thesis) require short acquisition times with usually low number of total counts, this ﬁnding has special importance for reader’s aiming to conduct time-resolved, quantitative studies on this system.

For high-count measurements the histogram of fan-beam SPECT as well as striatum MPH SPECT follows the Poisson statistics as expected and described in section 2.4.4, while the fan-beam histogram is more narrow due to superior SNR as presented in Fig. 4.24 (B).

4500 2500

4000

2000 3500

3000 1500 2500

2000 1000

number of voxel [#] voxel of number 1500 [#] voxel of number

1000 500

500

0 0 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 voxel value / ROI value [a.u.] voxel value / ROI value [a.u.] (a) striatum MPH low-count (b) striatum MPH high-count

2500 3000

2500 2000

2000 1500

1500

1000

number of voxel [#] voxel of number [#] voxel of number 1000

500 500

0 0 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 voxel value / ROI value [a.u.] voxel value / ROI value [a.u.] (c) fan-beam low-count (d) fan-beam high-count

Figure 4.26 – Histogram of selected low- and high-count phantom images. Part (a) (striatum MPH) and (c) (fan-beam) shows a histogram of all voxels in all ROIs over all slices for the reconstructed low-count images shown in Fig. 4.25 with 0.6 million total counts each. Part (b) (striatum MPH) and (d) (fan-beam) shows the histogram for the respective high-count images, i.e. 17.5 million total counts for MPH SPECT and 11.4 million total counts for the fan-beam SPECT.

✹✻ ❈❤❛♣❡ ✹✳ ❘❡✉❧

✹✳✹ ❛✐❡♥ ❝❛♥

Representative slices of the patient’s 123I-FP-CIT SPECT with MPH collimators versus LEHRHS collimators are shown in Fig. 4.27.

A 4.0

0.0 B 7.6

0.0

Figure 4.27 – Normal DAT availability in a male patient (54 years old) referred to DAT SPECT because of suspicion of PD. The patient was scanned twice after a single 123I-FP-CIT injection (total injected activity: 182 MBq). Part (A): 40 min scan in 2-head mode with LEHRHS collimators started 3 h 10 min after tracer injection. Part (B): 30 min scan in 3-head mode with striatum MPH collimators started 7 h 7 min after tracer injection. The MPH images were reconstructed with the Tera-Tomo™ algorithm with parameter settings optimized for DAT SPECT in clinical routine.

✹✼ ❈❤❛♣❡ ✹✳ ❘❡✉❧

✹✳✺ ❚❡ ♦❢ ❣❛♠♠❛ ❝❛♠❡❛ ❞❡❡❝♦ ✇✐❤ ♠✉❧✐✲♣✐♥❤♦❧❡ ❝♦❧❧✐♠❛♦ ✭♣♦♣♦❛❧ ♦ ❡①❡♥❞ ◆❊▼❆ ◆❯ ✶ ✲ ❡❝✐♦♥ ✸✳✸✮

In this section the previous ﬁndings are aggregated to form a standard protocol following the style of NEMA NU 1 standard[38, 39] in order to extend the performance evaluation of gamma cameras to be applicable to MPH collimator SPECT systems.

✹✳✺✳✶ ❙②❡♠ ❆①✐❛❧ ❛♥❞ ❘❛❞✐❛❧ ❈♦✉♥ ❙❡♥✐✐✈✐②

The system axial and radial count sensitivity profile shall be measured in increments of 1 cm with a point source of less or equal 5 µl in volume. The axial profile shall be measured along the COR-axis, the radial profile shall be measured perpendicular to the axial sensitivity peak. The total activity shall be between 10 MBq and 20 MBq. The measured values shall meet or exceed the specification of the manufacturer. Since the measurement depends on the specific multi-pinhole geometry, as well as the detector, the measurement must be reported for each multi-pinhole geometry separately.

❚❡ ❈♦♥❞✐✐♦♥

The radionuclides employed for these measurement shall be those for which the collimator were designed. The count rate shall not exceed 20,000 cps per detector head. For all radionuclides, an energy window recommended by the manufacturer for the appropriate clinical mode shall be used.

❚❡ ❊✉✐♣♠❡♥

The test equipment required for these measurements needs to be able to reproducible position the point source in space along the COR-axis and allow transaxial translations for radial measurements. A recommended sample holder (Fig. 4.28) shall consist of two low density slider stages made from polyvinyl chloride (PVC) mounted perpendicular on top of each other. The top slider carries a sample holder for an Eppendorf Tube® that contains the active point source. The bottom slider shall be mounted centrally on top of the patient table. Table height shall be adjusted such that the point source can be moved along the COR-axis. The length of the bottom slider shall exceed the axial FOV in both inferior and superior direction, enabling the acquisition of a full sensitivity proﬁle: When point source is placed at the extreme positions, measured sensitivity shall be negligibly small. The top slider length shall be maximized while accounting for the speciﬁc acquisition radius set by the manufacturer.

✹✽ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Figure 4.28 – Sample holder for point source measurements. The two-axis slider is ﬁxed to the patient table. The exact sample position can be determined using the integrated tape measure. The top slider features a transparent cylinder that contains the Eppendorf Tube® holding the point source.

▼❡❛✉❡♠❡♥ ♦❝❡❞✉❡

To ensure precise axial alignment of the point source slider along the center of rotation axis of the gantry, it is recommended to perform a full SPECT acquisition and reconstruction for at least two different point source positions inside the FOV and compare coordinates of the reconstructed point source for correct alignment. If available, supplementary CT imaging data can support slider alignment.

The FOV is sampled axially in increments of 1 cm. For each position of the point source a full SPECT acquisition is performed using recommended acquisition parameters for clinical use. Additionally, the number of background counts (cps) is measured once without a source present. For each measurement, acquisition start time and total number of counts is noted, i.e. sum of counts over all detectors and number of views. The total acquisition time should be adjusted such that at least 1,000,000 counts are collected when the point source is placed at the sensitivity maximum. All corrections recommended by the manufacturer for clinical patient scan should be turned on.

After analysis of the axial sensitivity proﬁle (see next step), a radial sensitivity proﬁle is measured perpendicular to the axial count sensitivity peak also in increments of 1 cm. Due to rotational symmetry, the radial sensitivity is measured only in one direction.

❈❛❧❝✉❧❛✐♦♥ ❛♥❞ ❆♥❛❧②✐

The total number of background counts bgtotal deﬁned as

bgtotal = bgrate tacq, where bgrate is the background count-rate [cps] and tacq is the total acquisition time [s] shall be calculated and subtracted for each scan. The total number of counts N0 is the sum of all

✹✾ ❈❤❛♣❡ ✹✳ ❘❡✉❧ counts over all slices without background. The activity A of the point source is being decay- corrected to the acquisition start time t0 of each measurement: A0 = A(t = t0). The system count sensitivity sen for each position of the point source in the FOV is calculated as:

N λ sen total (4.2) = λt , A0 (1 e acq ) − − with the appropriate decay constant λ for the used isotope.

In order to describe the axial sensitivity profile the following key performance indicators (KPIs) are calculated: peak sensitivity [cps/MBq], axial peak position [mm], half-width-at- half-maximum in inferior direction (HWHM inferior) [mm], half-width-at-half-maximum in superior direction (HWHM superior) [mm] and full profile width at 140 cps/MBq [mm]. The radial sensitivity profile is described by the following KPIs: half-width-half-maximum in radial direction (HWHM radial) [mm] and full profile width at 140 cps/MBq [mm].

The width at 140 cps/MBq is of special interest since it measures the range in the FOV where the MPH collimator matches or outperforms conventional LEHR collimators. It can be assumed that a conventional double-head SPECT system with parallel-hole collimators have at least 70 cps/MBq sensitivity per head as shown in table 4.1.

Table 4.1 – Typical system res- Collimator System resolution [mm] Count sensitivity [cps/MBq] olution and sensitivity per head for conventional parallel-hole LEHR, Siemens 7.4 91 collimators.[60]

LEHR, Philips 7.4 66

LEHR, GE Inﬁnia 7.4 72

KPI calculation can be done by fitting an asymmetric Gaussian function f (z) to the axial count sensitivity versus position profile. The axial position z [mm] in the FOV is defined such that z = 0mm is the axial center of the CFOV. The fit function f (z) is defined as:

(z b)2 a exp − 2 z 0 − 2c1 ≤ f (x)=  2 (4.3) (z b) a exp − z 0 2c2 > , − 2  where c1,c2 is the variance of f (z) in inferior and superior direction.

Some KPIs that describe the sensitivity proﬁle are directly given by the ﬁt parameters: a is the peak sensitivity [cps/MBq], b is the axial peak position [mm] in the FOV. All KPIs for description of axial sensitivity curve are illustrated in Fig. 4.29.

✺✵ ❈❤❛♣❡ ✹✳ ❘❡✉❧

HWHM inferior [mm] can be calculated by:

HWHM inferior = √2 ln2 c1 (4.4)

HWHM superior [mm] can be calculated by:

HWHM superior = √2 ln2 c2 (4.5)

600 Figure 4.29 – Visualization of KPIs a for description of axial sensitivity 500 curve.

400

300 HWHM inferior HWHM superior

200

full width at 140 cps/MBq sensitivity (cps/MBq) sensitivity 100

0 -100 -50 0b 50 100 z (mm)

For the radial sensitivity profile the position x [mm] is defined as the radial distance to the COR- axis of the system gantry. The fit function g(x) for the radial fit is also a Gaussian function, centered at x = 0mm, i.e. on the COR-axis:

x2 g(x)= a exp 2 , (4.6) −2c3 where c3 is the variance of g(x).

HWHM radial [mm] can be calculated by:

HWHM radial = √2 ln2 c3 (4.7)

The width at 140 cps/MBq is calculated based on the fit function for the axial and radial profile. The error of the width can be approximated by numerical simulation (variation of fit parameters).

✺✶ ❈❤❛♣❡ ✹✳ ❘❡✉❧

❘❡♣♦✐♥❣

The axial peak count sensitivity and position is reported along with the HWHM in inferior and superior direction. Additionally, the width of the profile at 140 cps/MBq is reported. For the radial sensitivity profile, the HWHM and the total width at 140 cps/MBq shall be reported. All key performance indicators (KPIs) shall be reported alongside their standard deviation (SD) as well as the 95% confidence interval (CI). An example for the analysis of the axial and radial sensitivity profile using the striatum MPH collimator and td = 0mm can be found in Tab. 4.2. Radial sensitivity was above 140 cps/MBq over the entire range measured [0 mm, 100 mm], therefore full width at 140 cps/MBq is stated to be > 200mm.

KPI Mean SD 95% -CI

peak sensitivity [cps/MBq] 641.0 6.6 626.3 - 655.8

peak position [mm] 1.3 1.2 -1.4 - 4.1

HWHM inferior [mm] 45.6 1.5 42.2 - 48.9 KPI Mean SD 95% -CI

HWHM superior [mm] 40.8 1.3 37.9 - 43.8 HWHM radial [mm] 80.6 0.9 78.6 - 82.6

full width at 140 cps/MBq [mm] 123.4 3.2 116.3 - 130.5 full width at 140 cps/MBq [mm] > 200

(a) Axial analysis (b) Radial analysis

Table 4.2 – Striatum MPH (td = 0mm) - KPI analysis of axial (a) and radial (b) sensitivity curve

The fitted axial count sensitivity profile is plotted together with the experimental data-points against the axial position in the FOV. The fitted radial count sensitivity profile at the position of the axial sensitivity peak is plotted against its radial distance to the center of rotation axis also including the original data-points. The deviation between fit function and experimental data is plotted for each profile in a separate graph showing only the residuals. An example of the axial and radial sensitivity profile reporting is given in Fig. 4.30a and Fig. 4.30b, respectively.

✺✷ ❈❤❛♣❡ ✹✳ ❘❡✉❧

700 700

600 600 500

500 400

300 400

200

sensitivity (cps/MBq) sensitivity (cps/MBq) sensitivity 300 100

0 200 -80 -60 -40 -20 0 20 40 60 80 0 20 40 60 80 100 z (mm) z (mm)

20 20

15 10 10

0 5

0 -10

-5

residuals (cps/MBq) residuals (cps/MBq) residuals -20 -10

-30 -15 -80 -60 -40 -20 0 20 40 60 80 0 20 40 60 80 100 z (mm) z (mm)

(a) axial proﬁle (b) radial proﬁle

Figure 4.30 – Striatum MPH (td = 0mm) - Fit of axial (a) and radial (b) sensitivity curve including a plot of respective residuals showing the difference between measurement and ﬁt.

✹✳✺✳✷ ❙②❡♠ ❆①✐❛❧ ❛♥❞ ❘❛❞✐❛❧ ❙♣❛✐❛❧ ❘❡♦❧✉✐♦♥

The system axial and radial spatial resolution shall be measured and expressed as the full-width- at-half-maximum (FWHM) of the point spread function. The measured values shall meet or exceed the speciﬁcation over the CFOV. Since the measurement depends on the speciﬁc multi- pinhole geometry, as well as the detector, the measurement must be reported for each multi- pinhole geometry separately.

❚❡ ❈♦♥❞✐✐♦♥

❚❡ ❊✉✐♣♠❡♥

Equipment explained in subsection 4.5.1.

✺✸ ❈❤❛♣❡ ✹✳ ❘❡✉❧

▼❡❛✉❡♠❡♥ ♦❝❡❞✉❡

To ensure precise axial alignment of the point source slider along the center of rotation axis of the gantry it is recommended to perform a full SPECT acquisition and reconstruction for at least two different point source positions inside the FOV and compare coordinates of the reconstructed point source for correct alignment. If available, supplementary CT imaging data can support slider alignment.

For each position of the point source a full SPECT acquisition is performed using recommended acquisition parameters for clinical use. The CFOV is sampled axially in increments of 1 cm. At the point of axial sensitivity peak, the CFOV is also sampled radially in increments of 1 cm. The sampling voxel size is recommended to be 0.2 FWHM if possible (recommendation ≤ taken from NEMA NU 1-2012, p.26 [38]) and the acquisition time shall be adjusted such that at least 1,000,000 counts are collected when the point source is placed at the sensitivity maximum.

❈❛❧❝✉❧❛✐♦♥ ❛♥❞ ❆♥❛❧②✐

The SPECT images are reconstructed using the reconstruction algorithm recommended by the manufacturer with parameters optimized for clinical use without attenuation correction. For each position of the point source the horizontal and vertical spatial resolution shall be calculated. Additionally, the mean of horizontal and vertical resolution shall be calculated. The horizontal spatial resolution is characterized by the FWHM of a Gaussian function fitted to the one-dimensional activity profiles obtained by first summing all transaxial image matrices and then summing over all rows of the sum matrix. For vertical spatial resolution summing over all columns of the sum matrix is required. When image analysis is performed with the SPM12 toolbox and MatLab software, the script in the appendix 9.1 may be used to obtain the results presented in Fig. 4.31.

✺✹ ❈❤❛♣❡ ✹✳ ❘❡✉❧

Figure 4.31 – Example of a results page of FWHM analysis of point source measurements using the recommended MatLab script in appendix 9.1. In this example, a point source measurement with the LEHRHS collimator is analyzed

❘❡♣♦✐♥❣

The mean of horizontal and vertical resolution shall be plotted against the axial or radial position of the point source and may be considered the spatial resolution at the speciﬁc location. In cases where horizontal and vertical resolution differs signiﬁcantly from each other both shall be plotted and reported separately.

✺✺ ❈❤❛♣❡ ✹✳ ❘❡✉❧

✹✳✻ ❆♥❛❧②✐ ♦❢ ❛①✐❛❧ ❛♥❞ ❛❞✐❛❧ ❡♥✐✐✈✐② ♣♦✜❧❡

The standard procedure proposed in section 4.5 has been applied to all measurements of the axial and radial sensitivity proﬁle of MPH SPECT. The results are plotted in Fig. 4.32, Fig. 4.33, and Fig. 4.34. All KPIs together with their respective standard deviation and 95% conﬁdence interval are listed in Tab. 4.3, Tab. 4.4, and Tab. 4.5.

700 700 600

600 600 500

500 500 400 400 400 300 300 300 200

200 200

sensitivity (cps/MBq) sensitivity (cps/MBq) sensitivity (cps/MBq) sensitivity

100 100 100

0 0 0 -80 -60 -40 -20 0 20 40 60 80 -80 -60 -40 -20 0 20 40 60 80 -100 -50 0 50 100 z (mm) z (mm) z (mm)

20 15 6

10 4 10 5 2

0 0 0

-5 -2 -10

-10 -4

residuals (cps/MBq) residuals (cps/MBq) residuals (cps/MBq) residuals -20 -15 -6

-30 -20 -8 -80 -60 -40 -20 0 20 40 60 80 -80 -60 -40 -20 0 20 40 60 80 -100 -50 0 50 100 z (mm) z (mm) z (mm)

(a) td = 0mm (b) td = 40mm (c) td = 80mm

Figure 4.32 – Analysis of axial sensitivity profiles of striatum MPH: Part (a): Fit of axial sensitivity profile for circular scanning (td = 0mm). Part (b): Fit of axial sensitivity profile for helical scanning with td = 40mm. Part (c): Fit of axial sensitivity profile for helical scanning with td = 80mm.

Striatum MPH - analysis of axial sensitivity proﬁle

td = 0mm td = 40mm td = 80mm

KPI Mean SD 95% -CI Mean SD 95% -CI Mean SD 95% -CI

peak sensitivity [cps/MBq] 641.0 6.6 626.3 - 655.8 608.0 4.8 598.2 - 619.5 530.7 1.5 527.5 - 534.0

peak position [mm] 1.3 1.2 -1.4 - 4.1 0.3 1.0 -1.8 - 2.4 0.0 0.4 -0.9 - 0.9

FWHM inferior [mm] 45.6 1.5 42.2 - 48.9 44.5 1.1 42.1 - 47.0 50.8 0.4 49.9 - 51.8

FWHM superior [mm] 40.8 1.3 37.9 - 43.8 44.2 1.1 41.9 - 46.6 51.2 0.5 50.3 - 52.2

full width at 140 cps/MBq [mm] 123.4 3.2 116.3 - 130.5 129.0 3.2 122.0 - 136.0 140.9 1.3 138.1 - 143.6

Table 4.3 – KPI overview - Striatum MPH: Analysis of axial sensitivity proﬁles (td = 0,40,80mm)

✺✻ ❈❤❛♣❡ ✹✳ ❘❡✉❧

500 500

450 400 400

300 350 300

200 250 sensitivity (cps/MBq) sensitivity sensitivity (cps/MBq) sensitivity 200 100 150

0 100 -150 -100 -50 0 50 100 -100 -50 0 50 100 z (mm) z (mm)

30 30

20 20 10 10 0

-10 0

-20

-10 residuals (cps/MBq) residuals residuals (cps/MBq) residuals -30 -20 -40

-50 -30 -150 -100 -50 0 50 100 -100 -50 0 50 100 z (mm) z (mm)

(a) td = 0mm (b) td = 40mm

Figure 4.33 – Analysis of axial sensitivity profiles of Alzheimer MPH: Part (a): Fit of axial sensitivity profile for circular scanning (td = 0mm). Part (b): Fit of axial sensitivity profile for helical scanning with td = 0mm.

Alzheimer MPH - analysis of axial sensitivity proﬁle

td = 0mm td = 40mm

KPI Mean SD 95% -CI Mean SD 95% -CI

peak sensitivity [cps/MBq] 478.9 7.2 463.6 - 494.2 462.2 4.4 452.9 - 471.6

peak position [mm] 0.0 3.3 -7.0 - 7.0 3.8 2.2 -0.8 - 8.3

FWHM inferior [mm] 80.3 3.8 72.3 - 88.3 88.3 3.1 81.6 - 95.0

FWHM superior [mm] 66.5 4.3 57.6 - 75.5 61.2 2.5 55.9 - 66.5

full width at 140 cps/MBq [mm] 188.9 7.7 172.6 - 205.2 178.1 3.5 170.6 - 185.6

Table 4.4 – KPI overview - Alzheimer MPH: Analysis of axial sensitivity proﬁles (td = 0,40mm)

✺✼ ❈❤❛♣❡ ✹✳ ❘❡✉❧

700 500

450 600 400

500 350

300 400

250

sensitivity (cps/MBq) sensitivity (cps/MBq) sensitivity 300 200

200 150 0 20 40 60 80 100 0 20 40 60 80 100 z (mm) z (mm)

20 20

15 10 10

0 5

0 -10

-5

residuals (cps/MBq) residuals (cps/MBq) residuals -20 -10

-15 -30 0 20 40 60 80 100 0 20 40 60 80 100 z (mm) z (mm)

(a) MPH APT71 (b) MPH APT59

Figure 4.34 – Analysis of radial sensitivity profiles (td = 0mm): Part (a): Fit of radial sensitivity profile for Striatum MPH. Part (b): Fit of radial sensitivity profile for Alzheimer MPH.

Analysis of radial sensitivity proﬁles

Striatum MPH Alzheimer MPH

KPI Mean SD 95% -CI Mean SD 95% -CI

FWHM [mm] 80.6 0.9 78.6 - 82.6 88.8 2.6 82.9 - 94.7

full width at 140 cps/MBq [mm] > 200 > 200

Table 4.5 – KPI overview - Analysis of radial sensitivity proﬁles (td = 0mm) for Striatum MPH and Alzheimer MPH

✺✽ ✺ ❉✐❝✉✐♦♥

The aim of this thesis was to propose a new standard for performance evaluation of gamma cameras with MPH collimators. To increase the chance for universal application of this standard, it was developed during a clinical performance evaluation of a triple-head general purpose SPECT camera with MPH collimators specifically designed for DAT SPECT. This novel collimator was designed to first achieve a high peak sensitivity at the striatum for recovery of striatal tracer uptake with high spatial resolution and low statistical noise and secondly achieve a sufficiently broad sensitivity profile to cover the whole brain to measure extrastriatal tracer uptake with adequate statistical quality for semi-quantitative analysis (ROI analysis).

The first major finding was that the proposed standard was suitable to comprehensively describe the superior performance of the 20 pinhole striatum MPH collimator designed for DAT SPECT over conventional collimators as discussed in detail over the following paragraphs. The process of optimizing the acquisition and reconstruction protocol is also discussed extensively. The second major finding was that the evaluation method could be applied to describe a second set of Alzheimer MPH collimators with 15 pinholes designed for brain perfusion SPECT. The axial and radial fit error of the sensitivity curve for this collimator suggests a systematic error using the developed fit function while still being able to determine the clinical relevant KPIs reason- ably well. For evaluation of even broader sensitivity curves it might be beneficial to divide the sensitivity profile not just in an inferior and superior part but also introduce a center part of constant sensitivity. In this case, an adjusted fit function could also describe conventional collimator sensitivity profiles using the same model. A third major finding was that the advantages of striatum MPH SPECT technology materialized also in a clinical patient scan. Hence, a standardized approach to clinical performance evaluation of MPH SPECT seems overdue.

The proposed standard for performance evaluation is limited by the small number of test systems it was developed on. During this study only two different MPH collimators were available for testing. Whereas the protocol for measuring the sensitivity profile should be transferable to any other MPH SPECT system, also hybrid systems combining different collimator classes, the analysis of the profile, especially the fit function, depends on the profile’s shape. The profile in turn depends on the exact arrangement and number of pin-holes as well as acquisition parameters like table motion. While the exact fit-function might need to be adjusted, the proposed KPIs are of universal importance for clinical DAT SPECT and could provide a helpful scale for comparing systems for this kind of application. The analysis of reconstructed spatial resolution was limited by the relatively large voxel size of 1.72 mm while testing for a FWHM of 3 mm to

✺✾ ❈❤❛♣❡ ✺✳ ❉✐❝✉✐♦♥

4 mm. If possible, this study recommends to adjust the voxel size to be 0.2 FWHM, which ≤ is also in line with NEMA NU 1-2012, p.26 [38]. In cases where the voxel size practically cannot be changed in clinical routine while keeping an acceptable SNR and reconstruction time, it is strongly recommended to perform additional measurements using a conventional (Derenzo type) hot-rod phantom placed in the CFOV. Unfortunately, overlapping projections of a hot-rod phantom might lead to multi-plexing artifacts (as presented for MD phantom measurements in Fig. 4.8) that are not present in point source measurements. Also, a hot-rod phantom is only able to determine the spatial resolution on a per cross-section basis with the assumption that spatial resolution is constant within the cross-section. In cases where this circumstance cannot be taken for granted and the reconstructed spatial resolution needs to be measured at a speciﬁc location in the FOV, only sampling with a point source can map spatial resolution to a speciﬁc point in the FOV.

The count sensitivity of the triple-head striatum MPH SPECT system was characterized by a profile with central peak and decreasing sensitivity towards the edges of the FOV, very different from the uniform sensitivity of the double-head SPECT system with fan-beam or LEHRHS collimators (Fig. 4.16). The axial striatum MPH sensitivity profile was almost symmetrical around its peak, although the peak was not in the transaxial plane at the geometric center of the gamma detector but was shifted by 7 cm towards the patient in inferior direction (Fig. 3.1a). As a result the axial sensitivity profile is also distorted. For the striatum MPH system the HWHM of the axial sensitivity profile profile is about 5 mm larger in inferior than in superior direction while circular scanning (Tab. 4.3). In helical scanning mode the profile is almost symmetrical around its peak. Proper positioning of the patient’s head in the CFOV (Fig. 3.3) such that the striatum is located in the sensitivity peak is required to make optimal use of the striatum MPH collimators for DAT SPECT.

The full width of the sensitivity profile at 140 cps/MBq (typical sensitivity of a double-head SPECT with parallel-hole LEHR collimators as shown in Tab. 4.1) was 12.9 cm in axial direction and > 20cm in transaxial direction (at the axial peak) (Tab. 4.3 and Tab. 4.5). Maximum linear dimensions of the human brain are about 14 cm full height (inferior-superior), 10 cm height without cerebellum, 15 cm width (left-right), and 18 cm length (anterior-posterior). Thus, the > 20cm transaxial FOV of the striatum MPH SPECT is clearly sufficient to reliably include the whole brain. The 12.9 cm axial FOV does not allow to cover the whole brain in all patients but only the cerebrum, that is, the brain without cerebellum. This is acceptable in DAT SPECT, because most authors recommend the occipital cerebrum [8] or the whole cerebrum without striata, thalamus, and hippocampus as reference region for semi-quantitative analysis in DAT SPECT [58], the cerebellum is rarely used [61]. However, even without the cerebellum, it can be quite challenging in clinical practice to reliably cover the whole cerebrum in the CFOV. In SPECT/CT systems, this might be supported by a sagittal X-ray localizer scan of the patient’s head. For SPECT-only systems, positioning lasers could be beneficial that indicate the center of rotation and the axial position of the sensitivity peak of the MPH collimators. Nevertheless,

✻✵ ❈❤❛♣❡ ✺✳ ❉✐❝✉✐♦♥ it is recommended that next generation MPH collimators for DAT SPECT provide somewhat broader sensitivity profile in axial direction ( 15cm). Additionally, this would extend the ap- ≥ plication capabilities of the same MPH collimator also for brain perfusion SPECT, in which inclusion of the whole cerebellum is important [62]. In order to address the need for a broader axial sensitivity profile, manufacturers might chose to design a sensitivity profile that features multiple axial peaks in circular scanning. Such a multi-peak profile could smooth out in helical scanning and create a broad axial plateau of maximum sensitivity. Using the standard evaluation protocol to compare striatum MPH and Alzheimer MPH collimators it is most obvious that in order to increase peak sensitivity, the number of pin-holes need to be increased (Alzheimer MPH: 15 pin-holes, striatum MPH: 20 pin-holes). Therefore, the second recommendation is to further increase the number of pin-hole in future generation.

Another ﬁnding of the present study was the peak system sensitivity of the triple-head system with striatum MPH collimators to be about 2.8 times higher than the sensitivity of a widely used double-head system with fan-beam collimators (608 5cps/MBq versus 223 2cps/MBq). ± ± Peak sensitivity per detector head was 202 2cps/MBq and 112 1cps/MBq for the stria- ± ± tum MPH and for the fan-beam system, respectively. Thus, peak sensitivity of the striatum MPH collimator was about 80 % higher than the fan-beam sensitivity. This is a highly relevant improvement. It is considerably larger than the 20 % sensitivity improvement provided by fan- beam collimators compared to parallel-hole collimators[13], which made practice guidelines on DAT SPECT recommend fan-beam collimators as the current state-of-the-art collimator technology for DAT SPECT [7, 8]. Therefore, striatum MPH SPECT is an interesting candidate to be the successor of fan-beam SPECT once more widely adapted by industry and evaluated using common performance standards like proposed in this thesis.

Reconstructed spatial resolution of MPH SPECT range from 3 mm to 4 mm FWHM throughout the whole FOV, about 3 times better than spatial resolution of the fan-beam SPECT (Fig. 4.20). A limitation of this comparison is that the acquisition radius was ﬁxed to 140 mm for all collimators while it could be possible for patients with a smaller head to experience minor resolution improvements when the acquisition radius is reduced (Fig. 4.23b), which is not possible for MPH SPECT. It is important to note that the resolution of 3 mm to 4 mm FWHM was achieved with the acquisition and reconstruction parameters optimized for clinical MPH DAT SPECT and, therefore, is representative for DAT SPECT in clinical routine. Comparison of the high count SPECT image of the anthropomorphic striatum phantom with CT-based scatter and attenuation correction (Fig. 4.15 (D)) with a computer simulation of a SPECT image of the phantom with 3 mm FWHM (Fig. 4.15 (E)) conﬁrms excellent spatial resolution in reconstructed MPH DAT SPECT images. Additionally, the comparison with a conventional hot-rod phantom (Fig. 4.21) also suggests to specify reconstructed resolution to range from 3 mm to 4 mm. This resolution is only marginally exceeded by the 2.5 mm to 3.5 mm spatial resolution reported for a clinical MPH SPECT (G-SPECT-I), further discussed in the next paragraph, which is dedicated for brain SPECT only and without support for conventional collimators[32, 63].

✻✶ ❈❤❛♣❡ ✺✳ ❉✐❝✉✐♦♥

The utility of MPH collimators for DAT SPECT in humans has been investigated previously by several groups. King and co-workers proposed a combination of a MPH collimator on one head of a double-head SPECT system and a fan-beam collimator on the other head[51]: The rationale was that the MPH collimator would provide high resolution and high sensitivity for imaging of the interior portion of the brain, particularly the striatum. The fan-beam collimator would provide lower resolution, but complete sampling of the brain addressing data sufficiency and allowing a volume-of-interest to be defined over cortical brain reference regions for semi- quantitative analysis[51]. Lee and co-workers demonstrated that MPH collimators can provide over an order of magnitude enhancement in peak count sensitivity compared to conventional parallel-hole collimators[33]. Neither the system proposed by King and co-workers nor the system proposed by Lee and co-workers has yet been made available for widespread clinical use. Additionally, system performance evaluation may not be transferable to other systems because general performance indicators that describe the full axial and radial count sensitivity and spatial resolution profile have not been proposed. Beekman and co-workers developed a clinical MPH SPECT system (G-SPECT-I) with full angular coverage by stationary detectors and adaptable pin-hole diameter for improved resolution-sensitivity tradeoff[63]. For SPECT imaging of the human brain, the system provides 2.5 mm spatial resolution with about 400 cps/MBq peak sensitivity or 3.5 mm spatial resolution with about 900 cps/MBq [32, 63]. Recently, the acquisition protocol for DAT-SPECT with this system has been optimized[32]. Clinical availability of the system depends on final local registrations, including FDA and CE clearance (https://www.milabs.com/clinical-imaging-systems/, accessed August 31, 2019).

The triple-head SPECT system evaluated in the present study is a conventional general purpose SPECT system that is available today and can also be operated with conventional collimators. This is a big advantage since it allows the user to enter MPH SPECT at lower cost by upgrading existing SPECT systems of this type with a new set of MPH collimators compared to investing in a new SPECT system only dedicated to MPH SPECT. Flexibility in application also can increase utilization rate especially for small and medium sized operations. Three heads not only provide 50 % higher count sensitivity compared to two heads, but also allow for more ﬂexibility in the MPH-design with respect to the combination of multi-pinhole collimators with other collimator types, e.g. fan-beam collimators as proposed by King and co-workers[51]. In addition, 3 heads provide better potential for stationary acquisition (without detector motion) with MPH collimators for fast dynamic imaging[18].

In order to guarantee performance estimates obtained by the proposed evaluation procedure to be representative for DAT SPECT in everyday clinical patient care, acquisition and reconstruction protocol were optimized for clinical DAT SPECT using an anthropomorphic striatum phantom as well as two geometric phantoms prior to the performance evaluation. The measurements to test angular sampling have not been performed on site. Therefore, an angular sampling of 4° (used in this study) has not been tested. Based on visual inspection of the provided SPECT images with 3.75° (96 views) and 5° (72 views) angular sampling with almost no difference in

✻✷ ❈❤❛♣❡ ✺✳ ❉✐❝✉✐♦♥ image quality it can be expected that the 4° angular sampling will also provide "optimal" image quality. However, this is a minor limitation of this study and it is recommended that angular sampling is also tested on site for future acquisition parameter optimization. A cost-effective way could be to build an insert for the MD phantom to test angular sampling besides axial sampling. The "optimal" table displacement identiﬁed using the MD phantom has been a compromise between quantitative analysis of the four central disks with respect to the sensitivity proﬁle, visual inspection of the reconstructed image and peak sensitivity point source measurements for various table displacements. For clinical DAT SPECT 40 mm of table displacement has been considered "optimal", while for assessing other MPH designs used for other applications priorities might be set differently. Another limitation is the phantom size, especially the diameter of the active disk being just 120 mm in order to save costs. According to Van Auden- haege et al. it is advisable to choose a phantom of the same size as the required FOV[12].A larger diameter phantom would increase projection overlap and potentially result in increased multi-plexing artifacts.

Quantitative ROI analysis using the anthropomorphic striatum phantom has been used for optimization of reconstruction parameters. Quality of Monte Carlo simulation did not signiﬁcantly effect CRC of Putamen while a higher quality improved the SNR slightly (Fig. 4.14 (A)). Even though top of the line processing hardware is used, it could be argued that more processing power could be utilized for higher quality simulations. Interestingly, a lower number of subsets decreased SNR (Fig. 4.14 (B)), which has not been fully understood. It can be assumed that processing the entire set at once, while increasing reconstruction times, requires a lower number of iterations to lead to a converging reconstruction. Since the number of iterations was constant for testing the "optimal" number of subsets, it can be assumed that reduced SNR is a consequence of over-iteration. Decreasing SNR for increasing number of iterations has been proven while testing the "optimal" number of iterations. However, the iterative Tera-Tomo™ algorithm showed a relative robustness of the reconstructed SPECT images when the number of iterations was varied over a very wide range up to 498 effective iterations (= maximum number of iterations allowed by the system software). SPECT images were rather robust both with respect to image quality for visual interpretation (Fig. 4.13) and with respect to results of semi- quantitative analysis (Fig. 4.14 (C)). Robustness of iterative reconstruction with respect to the number of iterations is important to achieve the same image quality in all patients, as the rate of convergence of iterative reconstruction in general depends on the image and, therefore, varies between patients. The recommended number of 90 effective iterations was chosen as it provides a good contrast recovery that may only be improved under a substantial increase of noise for higher numbers of effective iterations.

Using optimized parameters it has been shown that CT-based attenuation and scatter correction results in the best possible image, which was expected because Chang’s attenuation correction cannot account for variable attenuation due to patient anatomy or external factors like attenuation of the patient table. The ﬁndings are limited by the fact that an external CT image had to

✻✸ ❈❤❛♣❡ ✺✳ ❉✐❝✉✐♦♥ be used because the evaluated SPECT system did not feature an integrated CT. In order to minimize positioning errors of the phantom, the CT image was co-registered to the reconstructed SPECT image without attenuation correction and then used to reconstruct the SPECT image with CT-ACSC a second time. While alignment of the CT image was very good it should be noted that the external CT is inﬂuenced by external parameters like attenuation in the system speciﬁc patient table.

Measurements of the anthropomorphic striatum phantom with varying scan duration conﬁrmed superior sensitivity and spatial resolution of striatum MPH SPECT compared to fan-beam SPECT (Fig. 4.24 and Fig. 4.25). On the other hand, the signal-to-noise (SNR) ratio in the cerebrum (background ROI) decreases for striatum MPH SPECT for very high number of total counts (very long acquisition times) after reaching a maximum at around 3.5 million total counts, while the SNR for the fan-beam system keeps improving for high-count acquisitions. Not having full-transparency over the TeraTomo™ reconstruction algorithm, this effect cannot be explained to satisfaction. It’s assumed that the higher number of total counts results in a faster convergence of the reconstruction, such that the number of necessary effective iterations is lower. Keeping the number of effective iterations ﬁxed also for high-count images, might have the same noise enhancing effect as increasing the number of effective iterations for a given clinical image as the comparison between a Fig. 4.24 (B) and Fig. 4.14 (C) suggests.

Image quality with very low acquisition time is relevant for future dynamic studies of tracer update and proves to be also superior with striatum MPH SPECT with respect to spatial resolution. However, some reconstruction artifacts have been identiﬁed by assessing a low-count striatum MPH SPECT intensity histogram (Fig. 4.26a) compared to a low-count fan-beam intensity histogram (Fig. 4.26c) that especially reader’s interested quantitative analysis of dynamic striatum MPH SPECT should be aware of.

With the same phantom ﬁlling, the triple-head striatum MPH system acquired about 65 % more counts per minute than the double-head fan-beam system, which can be attributed to the fact that also the detector area is increased by about 50 % due to usage of a third detector head. The difference is smaller than the 180 % difference between peak sensitivity of the striatum MPH system and uniform sensitivity of the fan-beam system. This is due to the fact, that the total number of counts obtained by MPH SPECT of the striatum phantom is proportional to the average of the MPH sensitivity over all locations covered by the phantom weighted by the local activity concentration at these locations. While per head sensitivity was comparable, higher spatial resolution of striatum MPH SPECT compared to fan-beam SPECT resulted in better anatomical delineation of the striatum in the reconstructed images (Fig. 4.25) and higher contrast recovery (Fig. 4.24 B). For very low count striatum MPH SPECT images reconstruction artifacts in the reference ROI result in an underestimation of the reference activity and hence an overestimation of the activity concentration in the Putamen. As a consequence, CRC for putamen in striatum MPH SPECT with less than one million total counts is higher than ex-

✻✹ ❈❤❛♣❡ ✺✳ ❉✐❝✉✐♦♥ pected (Fig. 4.24 A). Image quality based on CRC evaluation reached a plateau at about 5 million counts for both systems. The total scan duration to acquire 5 million counts was 32 min for the triple-head striatum MPH system and 54 min for the double-head fan-beam system. A clear advantage of triple-head SPECT systems from an economical point of view. However, this advantage cannot be fully leveraged, especially in brain imaging, when only conventional collimators are used with a triple-head SPECT system. As presented in section 4.2, the concurrent use of a third head is associated with a larger acquisition radius than necessary for brain imaging and thus decreased resolution.

Some procedure guidelines for DAT SPECT recommend that visual interpretation of the SPECT images is complemented by semi-quantitative analysis of tracer (FP-CIT) uptake in the striatum and its sub-regions[7]. However, striatal FP-CIT uptake in the reconstructed images is sensitive to camera-specific differences of acquisition and reconstruction protocol, which limits the use of semi-quantitative analysis in DAT SPECT. In particular, camera-specific variability of spatial resolution causes variable underestimation of striatal FP-CIT uptake by up to 50 % due to partial volume effects[64]. This complicates the creation of large DAT SPECT normal databases across different camera systems and sites. Large databases are a prerequisite for future computer as- sisted preliminary diagnosis and support systems that could be a great benefit for inexperienced physicians. Thus, knowing the "true" value of striatal tracer binding in DAT SPECT would have direct benefit, namely the transfer-ability of the results across cameras and sites. Follow- up DAT SPECT[65] using different cameras would also benefit if true semi-quantitative values could be provided every time. MPH DAT SPECT is a step towards the true value, as was shown by improved CRC of striatum MPH SPECT (Fig. 4.24 (A)).

Superior imaging characteristics of MPH DAT SPECT compared to conventional collimator DAT SPECT was conﬁrmed in a patient scan (Fig. 4.27). Image quality of the MPH DAT SPECT in the patient was almost PET-like, although acquisition time was shorter compared to LEHRHS SPECT (30 min versus 40 min) and the activity concentration was around 10 % lower for the MPH DAT SPECT (performed 2 hours later) due to radioactive decay of 123I.

Another limitation of the present study is that phantom measurements were performed with 99mTc rather than 123I in order to save costs. The identified 2 % lower peak sensitivity for 123I might be explained by the larger sample volume of 123I covering also areas with lower count sensitivity. However, within the limits of accuracy of the measurements, striatum MPH peak sensitivity did not differ clinically significant between 99mTc and 123I, suggesting that results and conclusions of the present study can be translated to DAT SPECT with 123I-FP-CIT. This was confirmed by the 123I-FP-CIT DAT SPECT in a patient (Fig. 4.27).

✻✺ ✻ ❈♦♥❝❧✉✐♦♥

Key performance indicators relevant for MPH SPECT have been identiﬁed to develop a standardized evaluation protocol for gamma cameras with MPH collimators. Applying this protocol to comprehensively assess a MPH collimator designed for DAT SPECT imaging of the brain proves the technological advantage of MPH collimators compared to conventional collimators including low-energy-high-resolution-high-sensitivity collimators and fan-beam collimators: MPH collimators provide considerable improvement of image quality in DAT SPECT with respect to both spatial resolution and statistical noise. The superior image quality of MPH SPECT over LEHRHS SPECT has been veriﬁed by a patient scan.

As visual interpretation of DAT SPECT performed with conventional collimators by experienced readers already provides high diagnostic accuracy ( 90% sensitivity and specificity[9]), ≥ it is expected that improved image quality in MPH DAT SPECT is particularly useful in border- line cases and for less experienced readers. Additionally, improved image quality allows for a more realistic determination of tracer update which simplifies the creation of normal databases across different camera systems and sites. However, the largest benefit of improved image quality in DAT SPECT by MPH technology today is expected for early detection (or exclusion) of PD before the loss of putaminal DAT reaches 50 % and PD-characteristic movement problems occur. These hypotheses might also be tested in prospective clinical studies, preferably by comparing MPH SPECT and parallel-hole/fan-beam SPECT acquired in randomized order after a single injection of FP-CIT in the same patients.

The protocol has been successfully applied to a second set of MPH collimators designed for brain perfusion SPECT. Application of this protocol to MPH SPECT of other organs, across multiple camera systems and different manufacturers is needed and might be tested in prospective multi-center clinical studies before the proposed standard might be included in existing evaluation protocols like NEMA NU-1.

✻✻ ✼ ❆❝❦♥♦✇❧❡❞❣♠❡♥

My special thanks go to University of Hamburg professor Erika Garutti for kindly supervising this external master thesis and being ﬁrst reviewer.

I want to express great gratitude towards Prof. Susanne Klutmann for giving me the opportunity to work in her department of nuclear medicine and granting me unrestricted access after hours to conduct my research with all available tomography systems. I learned a lot controlling the equipment outside their standard application in patient care.

Especially, I want to thank Dr. Ralph Buchert for sharing his great experience in the ﬁeld of brain imaging in nuclear medicine. I am grateful for his guidance and determination in pushing this project forward, enabling me to participate in multiple scientiﬁc conferences both domestic and internationally.

Many thanks to both the German and Hungarian entity of Mediso Medical Imaging Systems for supporting my project and providing instant advice in application and technical related ques- tions, namely by Stefan Wieser and Gerald Hohendorf and Andras Wirth. I owe a great deal to Prof. Janos Mester for establishing this cooperation.

My very special gratitude goes to Thomas Schoch for his unconditional support, building the multi-disk phantom precisely to speciﬁcation in his workshop and helping removing many little device related obstacles during the entire research phase.

A big thank you to Attila Forgács from Scanomed Nuclear Medicine Center Debrecen for sharing the experimental RAW data of his star phantom and for many fruitful Skype discussions during which new troubleshooting ideas developed.

I want to thank Stephan Po-Sing Tai for being the best ofﬁce mate imaginable while establishing a great work atmosphere. I also want to thank our intern Louisa Göpfert for supporting my measurements regardless of the time of day.

A special mention to Marie Wegner for her support with the CAD modeling software, Tom Sokolinski and Thore Dassow for their support with python and the entire scientiﬁc staff in radiotherapy for some great conversations during the coffee break.

Finally, I am very grateful to my friends and family for their great support and encouragement, without this thesis would not have been possible.

✻✼ ✽ ❇✐❜❧✐♦❣❛♣❤②

[1] S. G. Mueller, M. W. Weiner, L. J. Thal, R. C. Petersen, C. Jack, W. Jagust, J. Q. Tro- janowski, A. W. Toga, and L. Beckett, “The Alzheimer’s Disease Neuroimaging Initia- tive”, Neuroimaging Clinics of North America 15, 869–877 (2005). [2] D. Twelves, K. S. Perkins, and C. Counsell, “Systematic review of incidence studies of Parkinson’s disease”, Movement Disorders 18, 19–31 (2003). [3] W. R. Gibb and A. J. Lees, “The relevance of the Lewy body to the pathogenesis of idiopathic Parkinson’s disease.”, Journal of Neurology, Neurosurgery & Psychiatry 51, 745–752 (1988). [4] J. Levin, A. Kurz, T. Arzberger, A. Giese, and G. U. Höglinger, “The Differential Diag- nosis and Treatment of Atypical Parkinsonism”, Deutsches Aerzteblatt Online 113, 61– 69 (2016). [5] NICE, “Parkinson’s disease in adults: diagnosis and management (NICE guideline NG71)”, National Institute for Health and Care Excellence, 47–51 (2017). [6] A. Berardelli, G. K. Wenning, A. Antonini, D. Berg, B. R. Bloem, V. Bonifati, D. Brooks, D. J. Burn, C. Colosimo, A. Fanciulli, J. Ferreira, T. Gasser, F. Grandas, P. Kanovsky, V. Kostic, J. Kulisevsky, W. Oertel, W. Poewe, J.-P. Reese, M. Relja, E. Ruzicka, A. Schrag, K. Seppi, P. Taba, and M. Vidailhet, “EFNS/MDS-ES recommendations for the diagnosis of Parkinson’s disease”, European Journal of Neurology 20, 16–34 (2013). [7] J. Darcourt, J. Booij, K. Tatsch, A. Varrone, T. Vander Borght, Ö. L. Kapucu, K. Nå- gren, F. Nobili, Z. Walker, and K. Van Laere, “EANM procedure guidelines for brain neurotransmission SPECT using 123I-labelled dopamine transporter ligands, version 2”, European Journal of Nuclear Medicine and Molecular Imaging 37, 443–450 (2010). [8] D. S. Djang, M. J. Janssen, N. Bohnen, J. Booij, T. A. Henderson, K. Herholz, S. Mi- noshima, C. C. Rowe, O. Sabri, J. Seibyl, B. N. Van Berckel, and M. Wanner, “SNM practice guideline for dopamine transporter imaging with 123I-ioﬂupane SPECT 1.0”, Journal of Nuclear Medicine 53, 154–163 (2012). [9] J. T. O’Brien, W. H. Oertel, I. G. McKeith, D. G. Grosset, Z. Walker, K. Tatsch, E. Tolosa, P. F. Sherwin, and I. D. Grachev, “Is ioﬂupane I123 injection diagnostically effective in patients with movement disorders and dementia? Pooled analysis of four clinical trials”, BMJ Open 4, e005122 (2014).

✻✽ ✽ ❇✐❜❧✐♦❣❛♣❤②

[10] G. Andringa, B. Drukarch, J. G. Bol, K. de Bruin, K. Sorman, J. B. Habraken, and J. Booij, “Pinhole SPECT imaging of dopamine transporters correlates with dopamine transporter immunohistochemical analysis in the MPTP mouse model of Parkinson’s disease”, NeuroImage 26, 1150–1158 (2005). [11] S. Geisler, N. Beindorff, M. Cremer, K. Hoffmann, W. Brenner, P. Cumming, P. T. Meyer, K.-J. Langen, E. Fuchs, and R. Buchert, “Characterization of [ 123 I]FP-CIT binding to the dopamine transporter in the striatum of tree shrews by quantitative in vitro autoradio- graphy”, Synapse 69, 497–504 (2015). [12] K. Van Audenhaege, R. Van Holen, S. Vandenberghe, C. Vanhove, S. D. Metzler, and S. C. Moore, “Review of SPECT collimator selection, optimization, and fabrication for clinical and preclinical imaging”, Medical Physics 42, 4796–4813 (2015). [13] Y. Akiyama and N. Yui, “Performance of a multislice fan beam collimator for SPECT imaging of the head”, Annals of Nuclear Medicine 5, 117–121 (1991). [14] A. Azazrm, E. Gharapapagh, J. Islamian, and B. Mahmoudian, “Advances in Pinhole and Multi-Pinhole Collimators For Single Photon Emission Computed Tomography Imag- ing”, World Journal of Nuclear Medicine 14, 3 (2015). [15] F. Beekman and F. van der Have, “The pinhole: gateway to ultra-high-resolution three- dimensional radionuclide imaging”, European Journal of Nuclear Medicine and Molecu- lar Imaging 34, 151–161 (2007). [16] M. H. Pirenne, “The Scientific Basis of Leonardo da Vinci’s Theory of Perspective”, The British Journal for the Philosophy of Science 3, 169–185 (1952). [17] C. Scherfler, E. Donnemiller, M. Schocke, K. Dierkes, C. Decristoforo, M. Oberladstätter, C. Kolbitsch, F. Zschiegner, G. Riccabona, W. Poewe, and G. Wenning, “Evaluation of striatal dopamine transporter function in rats by in vivo beta-[123I]CIT pinhole SPECT.”, NeuroImage 17, 128–41 (2002). [18] C. Lange, I. Apostolova, M. Lukas, K. P. Huang, F. Hofheinz, B. Gregor-Mamoudou, W. Brenner, and R. Buchert, “Performance Evaluation of Stationary and Semi-Stationary Ac- quisition with a Non-Stationary Small Animal Multi-Pinhole SPECT System”, Molecular Imaging and Biology 16, 311–316 (2014). [19] I. Apostolova, A. Wunder, U. Dirnagl, R. Michel, N. Stemmer, M. Lukas, T. Derlin, B. Gregor-Mamoudou, J. Goldschmidt, W. Brenner, and R. Buchert, “Brain perfusion SPECT in the mouse: Normal pattern according to gender and age”, NeuroImage 63, 1807–1817 (2012). [20] I. Apostolova, D. Niedzielska, T. Derlin, E. J. Koziolek, H. Amthauer, B. Salmen, J. Pahnke, W. Brenner, V. F. Mautner, and R. Buchert, “Perfusion Single Photon Emis- sion Computed Tomography in a Mouse Model of Neurofibromatosis Type 1: Towards a

✻✾ ✽ ❇✐❜❧✐♦❣❛♣❤②

Biomarker of Neurologic Deﬁcits”, Journal of Cerebral Blood Flow & Metabolism 35, 1304–1312 (2015). [21] G. Bal, P. D. Acton, F. Jansen, and B. H. Hasegawa, “Revolving multipinhole SPECT for small animal imaging”, in 2008 ieee nuclear science symposium conference record (Oct. 2008), pp. 5577–5584. [22] T. E. Peterson and S. Shokouhi, “Advances in Preclinical SPECT Instrumentation”, Jour- nal of Nuclear Medicine 53, 841–844 (2012). [23] K. Vunckx, J. Nuyts, B. Vanbilloen, M. De Saint-Hubert, D. Vanderghinste, D. Rattat, F. M. Mottaghy, and M. Defrise, “Optimized multipinhole design for mouse imaging”, IEEE Transactions on Nuclear Science 56, 2696–2705 (2009). [24] J. Nuyts, K. Vunckx, M. Defrise, and C. Vanhove, “Small animal imaging with multi- pinhole SPECT”, Methods 48, 83–91 (2009). [25] F. P. DiFilippo, “Design and performance of a multi-pinhole collimation device for small animal imaging with clinical SPECT and SPECT–CT scanners”, Physics in Medicine and Biology 53, 4185–4201 (2008). [26] C. Vanhove, M. Defrise, T. Lahoutte, and A. Bossuyt, “Three-pinhole collimator to improve axial spatial resolution and sensitivity in pinhole SPECT”, European Journal of Nuclear Medicine and Molecular Imaging 35, 407–415 (2008). [27] C. Si, G. S. Mok, L. Chen, and B. M. Tsui, “Design and evaluation of an adaptive multipinhole collimator for high-performance clinical and preclinical imaging”, Nuclear Medicine Communications, 1 (2015). [28] P. Yan, L. Chen, B. M. W. Tsui, and G. S. P. Mok, “Evaluation of Stationary and Semi- stationary Acquisitions from Dual-head Multi-pinhole Collimator for Myocardial Perfu- sion SPECT”, Journal of Medical and Biological Engineering 36, 675–685 (2016). [29] J. Bae, S. Bae, S.-y. Lee, K. Lee, Y. Kim, J. Joung, M. Kim, and K. M. Kim, “Two- level multi-pinhole collimator for SPECT imaging using a small-ﬁeld-of-view gamma camera”, Journal of the Korean Physical Society 70, 192–200 (2017). [30] L. Chen, B. M. W. Tsui, and G. S. P. Mok, “Design and evaluation of two multi-pinhole collimators for brain SPECT”, Annals of Nuclear Medicine 31, 636–648 (2017). [31] D. Salvado, K. Erlandsson, A. Bousse, M. Occhipinti, P. Busca, C. Fiorini, and B. F. Hut- ton, “Collimator Design for a Brain SPECT/MRI Insert”, IEEE Transactions on Nuclear Science 62, 1716–1724 (2015). [32] Y. Chen, B. Vastenhouw, C. Wu, M. C. Goorden, and F. J. Beekman, “Optimized image acquisition for dopamine transporter imaging with ultra-high resolution clinical pinhole SPECT”, Physics in Medicine & Biology 63, 225002 (2018).

✼✵ ✽ ❇✐❜❧✐♦❣❛♣❤②

[33] T.-C. Lee, J. R. Ellin, Q. Huang, U. Shrestha, G. T. Gullberg, and Y. Seo, “Multipinhole collimator with 20 apertures for a brain SPECT application”, Medical Physics 41, 112501 (2014). [34] M. E. Spilker, G. Bal, J. Uribe, D. Henderson, L. Thurfjell, C. Tan Hehir, X. Tao, A. Can, B. Sarachan, and F. Jansen, “Evaluation of different multi-pinhole imaging geometries for SPECT imaging of Parkinsonian disorders”, in 2008 ieee nuclear science symposium conference record (Oct. 2008), pp. 4022–4024. [35] Z. Cao, G. Bal, R. Accorsi, and P. D. Acton, “Optimal number of pinholes in multi- pinhole SPECT for mouse brain imaging—a simulation study”, Physics in Medicine and Biology 50, 4609–4624 (2005). [36] L. C. Johnson, S. C. Moore, and S. D. Metzler, “Effect of pinhole shape on projection resolution”, Physics in Medicine and Biology 61, 2003–2013 (2016). [37] M. C. Rentmeester, F. Van Der Have, and F. J. Beekman, “Optimizing multi-pinhole SPECT geometries using an analytical model”, Physics in Medicine and Biology 52, 2567–2581 (2007). [38] NEMA, “Standards Publication NU 1-2012”, National Electrical Manufacturers Associ- ation (2013). [39] NEMA, “Standards Publication NU 1-2018”, National Electrical Manufacturers Associ- ation (2019). [40] T. E. Peterson and L. R. Furenlid, “SPECT detectors: the Anger Camera and beyond”, Physics in Medicine and Biology 56, R145–R182 (2011). [41] N. R. C. Committee on the Mathematics and Physics of Emerging Dynamic Biomedical Imaging, Mathematics and Physics of Emerging Biomedical Imaging, tech. rep. (National Academy of Sciences, 1996), p. 261. [42] S. R. Cherry, J. A. Sorenson, and M. E. Phelps, Physics in Nuclear Medicine, 4th ed. (Elsevier, 2012). [43] A. C. Kak, M. Slaney, and G. Wang, Principles of Computerized Tomographic Imaging, Vol. 29, 1 (2002), pp. 107–107. [44] R. Hofstadter, “The Detection of Gamma-Rays with Thallium-Activated Sodium Iodide Crystals”, Physical Review 75, 796–810 (1949). [45] S. R. Cherry, J. A. Sorenson, and M. E. Phelps, “Interaction of Radiation with Matter”, in Physics in nuclear medicine (Elsevier, 2012), pp. 63–85. [46] K. Vunckx, D. Beque, M. Defrise, and J. Nuyts, “Single and Multipinhole Collimator Design Evaluation Method for Small Animal SPECT”, IEEE Transactions on Medical Imaging 27, 36–46 (2008).

✼✶ ✽ ❇✐❜❧✐♦❣❛♣❤②

[47] N. Ukon, N. Kubo, M. Ishikawa, S. Zhao, N. Tamaki, and Y. Kuge, “Optimization of helical acquisition parameters to preserve uniformity of mouse whole body using multi- pinhole collimator in single-photon emission computed tomography”, Results in Physics 6, 659–663 (2016). [48] R. Yao, X. Deng, Q. Wei, T. Dai, T. Ma, and R. Lecomte, “Multipinhole SPECT helical scan parameters and imaging volume”, Medical Physics 42, 6599–6609 (2015). [49] P. C. Huang and C. H. Hsu, “Fast iterative reconstruction for helical pinhole SPECT imaging”, Bio-Medical Materials and Engineering 26, S1371–S1380 (2015). [50] K. Vunckx, P. Suetens, and J. Nuyts, “Effect of overlapping projections on reconstruction image quality in multipinhole SPECT”, IEEE Transactions on Medical Imaging 27, 972– 983 (2008). [51] M. A. King, J. M. Mukherjee, A. Konik, I. G. Zubal, J. Dey, and R. Licho, “Design of a Multi-Pinhole Collimator for I-123 DaTscan Imaging on Dual-Headed SPECT Systems in Combination with a Fan-Beam Collimator”, IEEE Transactions on Nuclear Science 63, 90–97 (2016). [52] H. M. Hudson and R. S. Larkin, “Accelerated Image Reconstruction Using Ordered Sub- sets of Projection Data”, IEEE transactions on medical imaging 13, 601–609 (1994). [53] S. R. Cherry, J. A. Sorenson, and M. E. Phelps, “Nuclear Counting Statistics”, in Physics in nuclear medicine (Elsevier, 2012), pp. 125–140. [54] L.-t. Chang, “A Method for Attenuation Correction in Radionuclide Computed Tomog- raphy”, IEEE Transactions on Nuclear Science 25, 638–643 (1978). [55] M. Magdics, L. Szirmay-Kalos, B. Toth, D. Legrady, A. Cserkaszky, L. Balkay, B. Domonkos, D. Volgyes, G. Patay, P. Major, J. Lantos, and T. Bukki, “Performance evaluation of scatter modeling of the GPU-based "Tera-Tomo"3D PET reconstruction”, in 2011 ieee nuclear science symposium conference record (Oct. 2011), pp. 4086–4088. [56] H. K. Tuy, “An Inversion Formula for Cone-Beam Reconstruction”, SIAM Journal on Applied Mathematics 43, 546–552 (1983). [57] G. T. Gullberg, G. L. Zeng, F. L. Datz, P. E. Christian, C. .-.-H. Tung, and H. T. Morgan, “Review of convergent beam tomography in single photon emission computed tomography”, Physics in Medicine and Biology 37, 507–534 (1992). [58] D. Kupitz, I. Apostolova, C. Lange, G. Ulrich, H. Amthauer, W. Brenner, and R. Buchert, “Global scaling for semi-quantitative analysis in FP-CIT SPECT”, Nuklearmedizin 53, 234–241 (2014). [59] W. Koch, P. Bartenstein, and C. la Fougère, “Radius dependence of FP-CIT quantiﬁca- tion: a Monte Carlo-based simulation study”, Annals of Nuclear Medicine 28, 103–111 (2014).

✼✷ ✽ ❇✐❜❧✐♦❣❛♣❤②

[60] E. Bombardieri, J. Buscombe, G. Lucignani, and O. Schober, Advances in Nuclear On- cology: Diagnosis and Therapy, Vol. 1 (2007). [61] W. Koch, P. E. Radau, C. Hamann, and K. Tatsch, “Clinical testing of an optimized software solution for an automated, observer-independent evaluation of dopamine transporter SPECT studies.”, Journal of nuclear medicine : ofﬁcial publication, Society of Nuclear Medicine 46, 1109–18 (2005). [62] M. Takasawa, M. Watanabe, S. Yamamoto, T. Hoshi, T. Sasaki, K. Hashikawa, M. Mat- sumoto, and N. Kinoshita, “Prognostic value of subacute crossed cerebellar diaschisis: single-photon emission CT study in patients with middle cerebral artery territory infarct.”, AJNR. American journal of neuroradiology 23, 189–93 (2002). [63] F. J. Beekman, F. van der Have, M. C. Goorden, P. E. B. Vaissier, J. van Roosmalen, H. During, and B. Vastenhouw, “G -SPECT-I: a full ring high sensitivity and ultra- fastclinical molecular imaging system with < 3mm resolution”, European Journal of Nu- clear Medicine and Molecular Imaging 42, S209 (2015). [64] C. Lange, A. Seese, S. Schwarzenböck, K. Steinhoff, B. Umland-Seidler, B. J. Krause, W. Brenner, O. Sabri, J. Kurth, S. Hesse, and R. Buchert, “CT-Based Attenuation Correction in I-123-Ioﬂupane SPECT”, PLoS ONE 9, edited by Q. Zhang, e108328 (2014). [65] I. Apostolova, D. S. Taleb, A. Lipp, I. Galazky, D. Kupitz, C. Lange, M. R. Makowski, W. Brenner, H. Amthauer, M. Plotkin, and R. Buchert, “Utility of Follow-up Dopamine Transporter SPECT With 123I-FP-CIT in the Diagnostic Workup of Patients With Clini- cally Uncertain Parkinsonian Syndrome”, Clinical Nuclear Medicine 42, 1 (2017).

✼✸ ✾ ❆♣♣❡♥❞✐①

✾✳✶ ▼❛▲❛❜ ❝♦❞❡ ✲ ♦✐♥✲♦✉❝❡ ❛♥❛❧②✐ ✭❋❲❍▼✮

1 %function result = fwhm_KT

3 % fp, lp = first, last plane to average

5 % fit gauss f(x) = a*exp(-(x-x0)^2/s^2);

7 fname = spm_select(Inf,'any','select reconstructed point source ... image, please');

8 result = cell(size(fname,1),3);

10 for n = 1:size(fname,1)

11 % get header info

12 hdr = dicominfo(fname(n,:));

13 % matrix_size = hdr.Columns; % Columns = Rows assumed up to 255px

14 matrix_size = double(hdr.Columns); % Columns = Rows assumed for 256px ... onwards

15 pixel_size = hdr.PixelSpacing(1); % quadratic pixels assumed

16 Series_Description_RAW = hdr.SeriesDescription;

18 % get image

19 matrix = dicomread(fname(n,:));

20 matrix = sum(matrix,4);

21 %matrix = rot90(flipud(matrix));

22 marker = 1;

24 while (marker == 1)

25 % subtract background

26 omatrix = matrix;

27 close;

28 colormap(hot(128));

29 subplot(1,2,1);

30 imagesc(matrix);

31 axis equal;

32 bg_fraction = input('\n\nSpecify fraction of background voxels in %: ');

33 v = sort(reshape(matrix,matrix_size*matrix_size,1));

✼✹ ❈❤❛♣❡ ✾✳ ❆♣♣❡♥❞✐①

34 threshold = v(round(bg_fraction/100*matrix_size*matrix_size));

35 mask = (matrix threshold); ≤ 36 bg = mean(mean(matrix(mask)));

37 matrix = matrix - bg;

38 matrix(mask) = 0;

39 subplot(1,2,2);

40 imagesc(matrix);

41 axis equal;

42 % fprintf('\n\nPress any key to continue\n\n');

43 % pause;

44 close;

46 f = figure;

47 subplot(2,3,1);

48 imagesc(omatrix);

49 axis equal;

50 title('original');

51 subplot(2,3,4);

52 imagesc(matrix);

53 axis equal;

54 title(string);

56 % vertical projection

58 projection = mean(matrix);

59 subplot(2,3,2);

60 plot(projection,'.k');

61 a = axis;

62 a(1) = 1;

63 a(2) = max(size(projection));

64 axis(a);

65 title('vertical projection');

66 xlabel('pixel');

67 ylabel('intensity');

69 [maximum, index] = max(projection);

70 lb = index;

71 while ( projection(lb)/maximum > 0.5 )

72 lb = lb - 1;

73 end

74 ub = index;

75 while ( projection(ub)/maximum > 0.5 )

76 ub = ub + 1;

77 end

78 FWHM0 = ub - lb;

79 if ( (index - lb) < 2 )

80 lb = lb - 1;

81 end

✼✺ ❈❤❛♣❡ ✾✳ ❆♣♣❡♥❞✐①

82 if ( (ub - index) < 2 )

83 ub = ub + 1;

84 end

85 y = projection(lb:ub);

87 num_points = max(size(y));

88 y = reshape(y,num_points,1);

90 x = 1:num_points;

91 x = x';

93 [maximum, index] = max(y);

94 start(1)= log(maximum);

95 start(2) = index;

96 start(3) = FWHM0 / (2*sqrt(log(2)));

98 opts = optimset('fminsearch');

99 %opts = optimset(opts,'Display', 'off', 'TolX', 0.001, 'TolFun', ... 0.001, 'MaxFunEvals', 5000, 'MaxIter', 1000);

100 solution = fminsearch('chi2', start, opts, x, y);

101

102 a = exp(solution(1));

103 x0 = solution(2);

104 s = solution(3);

105

106 FWHM = 2*sqrt(log(2))*s; % in pixels; 107 FWHMmm_H = FWHM*pixel_size; %no semicolon-->display data in Command ... Window

108

109 subplot(2,3,3);

110 x = -1 : num_points + 2;

111 x = x';

112 y = projection(lb-2:ub+2);

113 y = reshape(y,num_points+4,1);

114 xp = -1:0.1:num_points + 2;

115 dxp = xp - x0*ones(size(xp)); 116 dxp2 = dxp .* dxp; 117 s2 = s*s; 118 fit = a * exp(-dxp2 / s2);

119 x1 = x0-FWHM/2;

120 x2 = x0+FWHM/2;

121 plot(x,y,'k+',xp,fit,'k-',[x1 x2], [a/2 a/2], 'k.-');

122 a = axis;

123 a(1) = min(x);

124 a(2) = max(x);

125 axis(a);

126 title(sprintf('horizontal FWHM = %6.1f mm', FWHMmm_H));

127 xlabel('pixel');

✼✻ ❈❤❛♣❡ ✾✳ ❆♣♣❡♥❞✐①

128 ylabel('intensity');

129

130 % horizontal projection

131

132 projection = mean(matrix');

133 subplot(2,3,5);

134 plot(projection,'.k');

135 a = axis;

136 a(1) = 1;

137 a(2) = max(size(projection));

138 axis(a);

139 title('horizontal projection');

140 xlabel('pixel');

141 ylabel('intensity');

142

143 [maximum, index] = max(projection);

144 lb = index;

145 while ( projection(lb)/maximum > 0.5 )

146 lb = lb - 1;

147 end

148 if ( (index - lb) < 2 )

149 lb = lb - 1;

150 end

151 ub = index;

152 while ( projection(ub)/maximum > 0.5 )

153 ub = ub + 1;

154 end

155 if ( (ub - index) < 2 )

156 ub = ub + 1;

157 end

158 y = projection(lb:ub);

159

160 num_points = max(size(y));

161 y = reshape(y,num_points,1);

162

163 x = 1:num_points;

164 x = x';

165

166 [maximum, index] = max(y);

167 start(1)= log(maximum);

168 start(2) = index;

169 start(3) = num_points / 8;

170

171 opts = optimset('fminsearch');

172 %opts = optimset(opts,'Display', 'off', 'TolX', 0.001, 'TolFun', ... 0.001, 'MaxFunEvals', 5000, 'MaxIter', 1000);

173 solution = fminsearch('chi2', start, opts, x, y);

174

✼✼ ❈❤❛♣❡ ✾✳ ❆♣♣❡♥❞✐①

175 a = exp(solution(1));

176 x0 = solution(2);

177 s = solution(3);

178

179 FWHM = 2*sqrt(log(2))*s; % pixels 180 FWHMmm_V = FWHM*pixel_size; %no semicolon-->display data in Command ... Window

181

182 subplot(2,3,6);

183 x = -1 : num_points + 2;

184 x = x';

185 y = projection(lb-2:ub+2);

186 y = reshape(y,num_points+4,1);

187 xp = -1:0.1:num_points + 2;

188 dxp = xp - x0*ones(size(xp)); 189 dxp2 = dxp .* dxp; 190 s2 = s*s; 191 fit = a * exp(-dxp2 / s2);

192 x1 = x0-FWHM/2;

193 x2 = x0+FWHM/2;

194 plot(x,y,'k+',xp,fit,'k-',[x1 x2], [a/2 a/2], 'k.-');

195 a = axis;

196 a(1) = min(x);

197 a(2) = max(x);

198 axis(a);

199 title(sprintf('vertical FWHM = %6.1f mm', FWHMmm_V));

200 xlabel('pixel');

201 ylabel('intensity');

202

203 Series_Description = ... (regexprep(regexprep(regexprep(Series_Description_RAW,' ... ','_'),'?',''),'_Recon.','')) %cleanup spaces/questionmarks in ... series description

204 saveas(f,Series_Description,'fig')

205 result(n,1) = {Series_Description};

206 FWHMmm_H_r = round(FWHMmm_H,2);

207 FWHMmm_V_r = round(FWHMmm_V,2);

208 result(n,2:3) = {FWHMmm_H_r,FWHMmm_V_r};

209 data = [FWHMmm_H_r FWHMmm_V_r]

210

211 marker = input('\n\nTry a different background fraction ... (1=yes/0=no)?: ');

212 end

213 end

214 xlswrite('result.xlsx',result)

Listing 9.1 – MatLab script for point-source analysis in batch mode without ﬁt function

✼✽ ❈❤❛♣❡ ✾✳ ❆♣♣❡♥❞✐①

1 function residuum = chi2(parameters,x,y)

3 % chi2 for Gauss

5 lna = parameters(1);

6 x0 = parameters(2);

7 s = abs(parameters(3));

9 dx = x - x0*ones(size(x)); 10 dx2 = dx .* dx; 11 s2 = s*s;

13 lny0 = lna - dx2 / s2;

14 lny = log(y);

16 dy = lny - lny0;

18 residuum = dy' * dy;

Listing 9.2 – MatLab script for point-source analysis in batch mode - ﬁt function

✾✳✷ ❆❞❞✐✐♦♥❛❧ ▼❛▲❛❜ ❝♦❞❡

Additional small bits of code that might be worth sharing to be used in future projects.

✾✳✷✳✶ ❆♥♦♥②♠✐③❡ ✳❞❝♠ ❞❛❛ ✐♥ ❝❛❡ ♦❢ ♣❛✐❡♥ ❝❛♥

1 p = spm_select(Inf,'any','Please select images to be anonymized!'); ... %select input files

2 for i = 1:size(p,1)

3 p_clean = regexprep(p(i,:),'.dcm','_anonymized.dcm') %output file ... extension

4 dicomanon(p(i,:),p_clean,'keep',{'PatientAge','PatientSex', ... 'StudyDescription'}) %delete any personal data but keep age, sex ... and study description

5 end

Listing 9.3 – MatLab script to anonymize .dcm data in batch mode

✼✾ ❈❤❛♣❡ ✾✳ ❆♣♣❡♥❞✐①

✾✳✷✳✷ ❙❧✐❝❡ ✉♠♠✐♥❣ ♦♦❧

1 p = spm_select %input file in .nii format

2 v = spm_vol(p)

3 y = spm_read_vols(v);

4 z = sum(y(:,:,63:78),3); %select slices to be summed up

5 v.fname = 'C:\Users\...\folder\filename.nii' %output file

6 v.dim = [128 128 1]

7 spm_write_vol(v,z) % save image

Listing 9.4 – MatLab script for slice summing

✾✳✷✳✸ ▼✉❧✐✲❞✐❦ ❛♥❛❧②✐ ✲ ✐♥❡♥✐② ♣♦❥❡❝✐♦♥ ♦♥ ❈❖❘✲❛①✐

1 p = spm_select(Inf,'any','Please select reconstructed MD Phantom ... images!'); % load input images

2 figure

3 a = 2

4 b = round(size(p,1)/2,0)

5 for i = 1:size(p,1)

6 n = deblank(p(i,:));

7 v = spm_vol(n);

8 y = spm_read_vols(v);

9 z = sum(sum(y,1),2);

10 z = reshape(z,max(size(z)),1);

11 z = z/max(z);

12 m(:,i) = z;

13 subplot(a,b,i);

14 plot(z);

15 name = extractBefore(extractAfter(n, 'APT71_'), '_');

16 axis([30 160 0 1.15]);

17 title(name);

18 xlabel('pixel');

19 ylabel('rel. intensity');

21 end

22 save 'MD_profiles.txt' m -ascii % save profiles

Listing 9.5 – Multi-disk analysis

✽✵ ❊❦❧✉♥❣

Hiermit bestätige ich, dass die vorliegende Arbeit von mir selbstständig verfasst wurde und ich keine anderen als die angegebenen Hilfsmittel - insbesondere keine im Quellenverzeichnis nicht benannten Internet-Quellen - benutzt habe und die Arbeit von mir vorher nicht in einem anderen Prüfungsverfahren eingereicht wurde. Die eingereichte schriftliche Fassung entspricht der auf dem elektronischen Speichermedium.

Ich bin damit einverstanden, dass die Masterarbeit veröffentlicht wird.

Hamburg, den 1. September 2019

(Kristian Tecklenburg)

✽✶