Lukmanda Evan Lubis Doctoral Thesis Summary

HALAMAN JUDUL UNIVERSITAS INDONESIA NOVEL PHANTOM FOR THREE-DIMENSIONAL ROTATIONAL ANGIOGRAPHY (3DRA): DESIGN, TESTING, AND APPLICATION DISSERTATION SUMMARY LUKMANDA EVAN LUBIS 1706123192 FACULTY OF MATHEMATICS AND NATURAL SCIENCES DEPARTMENT OF PHYSICS PHYSICS DOCTORAL PROGRAM DEPOK DECEMBER 2021

UNIVERSITAS INDONESIA NOVEL PHANTOM FOR THREE-DIMENSIONAL ROTATIONAL ANGIOGRAPHY (3DRA): DESIGN, TESTING, AND APPLICATION DISSERTATION SUMMARY Submitted to be defended in public defense under the auspices of the Rector of Universitas Indonesia, Prof. Ari Kuncoro, S.E., M.A., Ph.D. on Friday, December 10th, 2021 as partial fulfillment for the degree of Doctor in Physics from Universitas Indonesia LUKMANDA EVAN LUBIS 1706123192 FACULTY OF MATHEMATICS AND NATURAL SCIENCES DEPARTMENT OF PHYSICS PHYSICS DOCTORAL PROGRAM DEPOK DECEMBER 2021

PUBLIC DEFENSE COMMITTEE Under the Auspices of Prof. Ari Kuncoro, S.E., M.A., Ph.D. Rector of Universitas Indonesia Chief Committee Dr. Rokhmatuloh, S.Si., M.Eng. Dean, Faculty of Mathematics and Natural Sciences, Universitas Indonesia Promotor Prof. Dr. rer. nat. Terry Mart Department of Physics, Faculty of Mathematics and Natural Sciences, Universitas Indonesia Co-promotor I Prof. Dr. Djarwani Soeharso Soejoko, M.S., FIOMP Department of Physics, Faculty of Mathematics and Natural Sciences, Universitas Indonesia Co-promotor II Prof. Dr. Ir. Hilde Bosmans Department of Imaging and Pathology, Faculty of Medicine, Katholieke Universiteit Leuven iii

Examiner I Dr. Muhammad Aziz Majidi Department of Physics, Faculty of Mathematics and Natural Sciences, Universitas Indonesia Examiner II Dr. Supriyanto Ardjo Pawiro Department of Physics, Faculty of Mathematics and Natural Sciences, Universitas Indonesia Examiner III Dr. Yessie Widya Sari Department of Physics, Faculty of Mathematics and Natural Sciences, IPB University Examiner IV Dr. dr. Prijo Sidipratomo, Sp.Rad.(K), M.H. Department of Radiology, Faculty of Medicine, Universitas Indonesia Examiner V Prof. Dr. Kwan Hoong Ng Department of Biomedical Imaging, Faculty of Medicine, University of Malaya iv

FOREWORD Solemn gratitude and ultimate appraisal shall be addressed to Allah The All-Knowing, for the completion of this dissertation on design, testing, and application of a novel phantom for three-dimensional rotational angiography, prepared to partially fulfill the requirements for the degree of doctor from the Physics Doctoral Program, Department of Physics, Faculty of Mathematics and Natural Sciences, Universitas Indonesia. The author would like to dedicate this dissertation as a token of gratitude to the following individuals with merit: 1. Prof. Dr. rer. nat. Terry Mart as the promoter who is tireless and always available when the author seeks support and guidance. His significant help is instrumental in the completion of this dissertation. 2. Prof. Dr. Djarwani Soeharso Soejoko, FIOMP, who, aside of being a wonderful co-promoter and academic supervisor throughout the author’s career, is an even greater inspiration in the author’s life. Her radiant wisdom shall always be looked upon with awe. 3. Prof. Dr. Ir. Hilde Bosmans, whose broad knowledge and expertise in the field had significantly improve the value of this dissertation. 4. Members of the board of examiners; Dr. Muhammad Aziz Majidi, Dr. Supriyanto Ardjo Pawiro, Dr. Yessie Widya Sari, Dr. dr. Prijo Sidipratomo Sp.Rad (K), M.H., and Prof. Dr. Kwan Hoong Ng, for priceless inputs in improving this work. 5. The Faculty of Mathematics and Natural Sciences, Universitas Indonesia, and Universitas Indonesia for tuition coverage during the entire studentship. 6. The ever-beloved, irreplaceable mamma, Wuryanti Rahayu G. Martowiguno for her compassion, strength, and patience in making sure that all her two children, whom she had raised alone, have a doctoral degree—this wait will soon be over. 7. The sparkling joy of the author’s life, Zikra Yuyun Adenaila, for all her around-the-clock support and sacrifice, and the little one Rinipta Kinarya Lubis, who, by the time this dissertation is concluding, had spent her entire life cheering her father to finish this dissertation. 8. The author’s brother, Dr. Adiella Yankie Lubis, and the author’s dear papa Achmad Zulkifli Lubis for the boundless encouragement. v

9. Anonymous peer-reviewers of Physica Medica and Atom Indonesia, who had taken their task very seriously, resulting in great improvements in the published manuscripts. 10. Research partners, students, and colleagues as co-authors; Ika Hariyati, Indah Lestariningsih, Intan Apriliani S. Mu’minah, Dea Ryangga, Riski Abdul Basith, Farisa Adlina Ihsani, Ashyfa Santosa, Haitsam Shiddiq Siregar, whose energy has been invested in many parts of this dissertation. 11. The SARS-CoV-2 virus and all its variants, without which this dissertation would have been completed a year earlier. This dissertation work was prepared and completed not on a perfect manner. Flaws and imperfections are to be found, upon which inputs and corrections are welcomed for future improvements. Depok, December 10th, 2021 Author vi

ABSTRACT Name : Lukmanda Evan Lubis Course program : Physics, Doctoral degree Title : Novel Phantom for Three-Dimensional Rotational Angiography (3DRA): Design, Testing, and Application Aiming to make available a specific quality control tool for three-dimensional rotational angiography (3DRA), this dissertation presents results and conclusions of preliminary studies as well as primary studies. The first part of the preliminary study was a series of radiation-related tests of material candidates and selections for phantom inserts to produce adequate tools for assessing image quality. The tests include mass density, electron density, effective atomic number, and linear attenuation coefficient evaluation resulting in the use of polymethyl methacrylate (PMMA) as the phantom body and others as insert objects. The next study presents the evaluations of geometry aspects to dose and image quality in CT. The investigated aspects were CT dose coefficients change with the changing shape and size of the phantom as well as modulation transfer function (MTF) and noise homogeneity change under the same situation, resulting in the decision of using standard sized 16 cm phantom as the novel phantom design. The third part of the study tests the constructed in-house phantom including the materials of choice and compared its performance to a commercial phantom, where the phantom’s ability to accommodate CT dose index (CTDI), contrast, and MTF measurement was evaluated, resulting in conformity with the reference commercial phantom. The use of the phantom for 3DRA has been evaluated and performed in the fourth part of the report, showing that the in-house phantom excels in contrast depiction against Catphan while able to perform measurements of MTF and noise power spectrum (NPS). In the final part of the dissertation, the phantom is proven to be instrumental in performing a task-based evaluation of 3DRA imaging modes when applied on an angiography modality with four selectable modes dedicated for cranial application. In this last part, the use and the phantom gave contrast information and, side-by-side with spectrum information, was able to conclude that a particular mode should be recommended for contrast-enhanced 3DRA cranial application. It is concluded that the produced in-house phantom can be recommended for 3DRA use for performance evaluation and imaging mode selection. Keywords : 3DRA, dose, image quality, phantom Bibliography : 77 (1973-2020) vii

LIST OF CONTENTS TITLE PAGE.................................................................................................. i PUBLIC DEFENSE COMMITTEE............................................................... iii FOREWORD................................................................................................. v ABSTRACT ................................................................................................ vii LIST OF CONTENTS................................................................................. viii 1. INTRODUCTION .............................................................................. 1 2. THEORY AND LITERATURE REVIEW ......................................... 3 2.1 Fluoroscopy and Fluoroscopically Guided Intervention (FGI) ........... 3 2.2 Three-Dimensional Rotational Angiography (3DRA) ....................... 5 2.3 Radiation Dose Measurement for 3DRA........................................... 6 2.4 Image Quality Metrics in 3DRA......................................................11 2.5 Quality Control and Tools for 3DRA..............................................13 3. MATERIALS AND METHODS.........................................................15 3.1 Material Selection and Testing .......................................................15 3.1.1 Materials preparation ........................................................16 3.1.2 Mass density measurement.................................................17 3.1.3 Electron density and effective atomic number calculations..18 3.1.4 Linear attenuation coefficient measurement ........................20 3.2 Study on Dose and Image Quality Effects of Phantom Geometry ....22 3.2.1 Experimental phantoms......................................................22 3.2.2 Dose trend study ................................................................23 3.2.3 Image quality study............................................................25 3.3 Phantom Design, Construction, and Evaluation ................................28 3.3.1 Phantom material and design..............................................28 3.3.2 Phantom testing .................................................................29 3.4 Novel Phantom for Performance Evaluation of 3DRA ....................29 3.4.1 Study steps ........................................................................30 3.4.2 Use of the in-house phantom ..............................................30 3.4.3 Image quality evaluation using in-house phantom...............32 3.4.4 Imaging performance measurement methods ......................33 3.5 Selection of 3DRA Task-Based Imaging Modes .............................36 3.5.1 Modality and phantom .......................................................36 3.5.2 Dose metric measurement and spectrum simulation ............36 3.5.3 Image quality quantification...............................................38 4. RESULTS AND DISCUSSIONS ........................................................39 4.1 Material Selection and Testing Result ..............................................39 viii

4.2 Study on Dose and Image Quality Effects of Phantom Geometry ....42 4.2.1 Dose index coefficient........................................................42 4.2.2 Image quality trend ............................................................43 4.2.3 Considerations for the in-house phantom ............................46 4.3 Phantom Design, Construction, and Evaluation Results ..................46 4.3.1 Dose measurement reading.................................................47 4.3.2 Image quality metric evaluation result ................................48 4.4 Novel Phantom Performance Evaluation for 3DRA ........................51 4.4.1 Comparison with Catphan® phantom .................................51 4.4.2 In-house phantom performance for 3DRA ..........................53 4.4.3 In-house phantom performance for CTA ............................55 4.4.4 Overall evaluation..............................................................58 4.5 Selection of 3DRA Task-Based Imaging Modes .............................61 4.5.1 Dose and spectrum analysis................................................61 4.5.2 SDNR assessment result.....................................................65 4.5.3 Qualitative analysis for mode selection...............................67 5. CONCLUSION AND SUGGESTIONS 5.1 Conclusion ......................................................................................70 5.2 Suggestions.....................................................................................70 REFERENCES............................................................................................72 PUBLICATION LIST.................................................................................79 CURRICULUM VITAE .............................................................................80 ix

INTRODUCTION In the last three decades, angiography has been playing an active role in both diagnosis and treatment for multiple types of medical conditions. Its capability to present real-time image acquisition has made it excel among other X-ray-based medical imaging modalities such as general-purpose X-ray and Computed Tomography (CT) (Bushberg et al., 2011; International Atomic Energy Agency, 2014). This specific capability has allowed clinicians to utilize the continuous and live imaging tool to assist the intervention to the patient’s body, e.g., stent placements, embolization, and ischemic relief efforts (Strauss, 2002); hence the common name ‘interventional radiology’. Recently, the expanding technology has allowed three-dimensional rotational angiography (3DRA) to be in place and complimenting the existing planar (two dimensional) angiography. 3DRA has been extensively exploited during the last decade, ranging from visualization of the vascular system and soft tissues (e.g. lesion/neoplasm) to bone structures. With volume rendering and resulting CT-like images as the key feature, it allows for the evaluation of intriguing anatomies to the detection of hemorrhages and blockage that might be missed during two-dimensional imaging (Nollert et al., 2012). It is also deemed as a useful tool for intra-procedural detection of intracranial aneurysm (Ishihara et al., 2000; Li et al., 2010; van Rooij et al., 2008), coronary interventions (Neubauer et al., 2010; Rasche et al., 2002), as well as targeted chemoembolization for hepatocellular carcinoma (Liapi et al., 2005; Tacher et al., 2015). These diagnostic and therapeutic procedures, in fact, can be aided by acquiring tomographic images in the catheterization laboratory. Such potential might enable the 3DRA to replace CT under demanding situations, which essentially requires studies proving the inter-comparability between CT scan and CT-like 3DRA on various task-based imaging protocols. Due to the use of X-rays in interventional radiology procedures, the safety aspect in terms of radiation dose to patients has been a great concern and challenge for medical physicists (Chu et al., 2006; Faulkner, 2005; McFadden et al., 2002; Vano, Gonzalez, et al., 2005). Additionally, ensuring the resulting image quality remains a major effort. The success of these procedures depends almost entirely on the positioning accuracy and imaging performance of the 3DRA system, which requires proper management and regular evaluation through quality control (QC) programs. Despite the extensive clinical use of 1

3DRA, tools and methods for its QC are currently limited. Image quality checks are treated similarly to other 3D (CT) imaging modalities despite the different clinical use of 3DRA. Radiation dose assessment via an output metric is only recently available in the EFOMP-ESTRO-IAEA CBCT quality control protocol (de las Heras Gala et al., 2017, 2019). In addition, QC of 3DRA is also overlooked in the interventional fluoroscopy quality assurance program. Existing protocols suggested either subjective scoring method (Zhang et al., 2010) or using tools dedicated for classical CT scanners (Bai et al., 2012). Although the existing EFOMP-ESTRO-IAEA CBCT QC protocol had covered interventional 3DRA performance check methods, the tools mentioned in the protocol are basically the tools of other CBCT modalities like dental and radiotherapy on- board imaging—none of which are used clinically with Iodinated contrast agents for very specific contrast needs. Consequently, there is a lack in tools and phantoms dedicated for 3DRA QC. This dissertation is aimed to address the above issues by making available a tool for prompt, quantitative evaluation on image quality and a radiation dose metric for 3DRA, from the choice of materials and shape, in-house phantom production, and then test its performance and demonstrate its application as a tool to choose 3DRA imaging modes under specific clinical task. 2

THEORY AND LITERATURE REVIEW Vascular catheterization procedures make use of a specially designed medical imaging modalities. Rather than general-purpose fluoroscopy, catheterization laboratories are typically equipped with specially designed fluoroscopy systems to guide the entire length of the interventional procedure; hence the name ‘fluoroscopically-guided intervention (FGI)’. From the gold standard planar (two dimensional) fluoroscopy, more recent modalities had expanded their capability into allowing for intra-procedure volumetric image acquisition through a feature called three-dimensional rotational angiography (3DRA). The involvement of ionizing radiation on such systems, both for planar and volumetric imaging, increases the need of an extended knowledge of these systems. Basic knowledge on dose-related applications is also required before discussing a study on dose and image quality aspects of 3DRA system. 2.1 Fluoroscopy and Fluoroscopically Guided Intervention (FGI) Fluoroscopic radiology systems are the most common modalities to obtain real-time images of the patient’s anatomical landmarks. The presence of the temporal dimension on fluoroscopic systems allows clinicians to observe the patient’s moving anatomy and pathology. General-purpose fluoroscopy systems are typically equipped with image intensifiers, with analogue or digital camera in the observer’s end to display the acquired images on medical grade monitors. Typical images are acquired at 30 frames/second on general (pelvic and abdominal) procedures, although higher frame rates (up to 120 frames/second) are available for imaging the cardiovascular structures (Strauss, 2002). Whereas in general fluoroscopy 18.000 images can be acquired in 10 minutes of acquisition, interventional procedure produces more images due to typically longer duration. The total duration of an interventional procedure, moreover, is often unpredictable and almost always varies between patients and cases due to the widely varied complexity factors. The high number of images acquired during interventional procedure leads to a relatively high patient dose. In an attempt to reduce patient dose, a very responsive image receptor system is put in place. Image intensifiers (II) are three 3

magnitudes more sensitive than film-screen systems (Bushberg et al., 2011; Hendee & Ritenour, 2002). General fluoroscopy requires 1 to 5 µR on the image intensifier to produce an image, whereas a film-screen system with speed number 400 needs as much as 600 µR to obtain an optical density value of 1.0. More recently used modalities in Indonesia, however, employ the active-matrix flat panel arrays as image receptors, known as Flat Panel Detectors (FPD) for less image quality deterioration. The FPD consists of an active-matrix flat panel coupled with input X- ray detection material, allowing for real-time imaging of the X-ray (photon) fluence. The configuration requires no film rolls since all images are digitally recorded. X-ray detection in FPD system takes place electronically, eliminating the steps of X-ray conversion to visible light and electrons as in the use of conventional image intensifiers. In the FPD, the incident X-rays interact in the photo conducting amorphous silicon layer, a semiconductor material. Electrons of the semiconductor layer, upon receiving the energy from the incident X-ray photons, are excited to the higher-energy conduction band. They are then collected by microscopic local electrode, one set for each picture element (pixels), and connected by an array of silicon-based thin film transistors (TFT) situated in the nearby layer. In the next step, Fourier transformation is applied by the system to reconstruct the image. Whereas in cine mode the images were acquired periodically using pulsed exposure, in fluoroscopic mode the images are acquired continuously. There are multiple frame rates, each with their number of pulses per second for the cine acquisition. In the clinical implementation and development of interventional procedures, there exists a need to overcome issues found on planar, two- dimensional imaging. The primary issue that requires innovation was overlapping anatomical/pathological structure in the resulting planar image. Acquiring CT images requires the patient to be transported out of the catheterization laboratory and, therefore, is inadvisable. This issue encourages the development of an advanced imaging mode; the three-dimensional rotational angiography (3DRA). 4

2.2 Three-Dimensional Rotational Angiography (3DRA) In its core, 3DRA is a relatively novel feature in interventional radiology that presents reconstruction of tomographic slices of a volume of interest in a similar manner to computed tomography (CT), using only single- plane radiographic equipment (De Potter et al., 2014; Stenger et al., 2016). It essentially employs the principle of convolution-back projection for direct reconstruction of a three-dimensional density function from a set of two- dimensional projections as developed by Feldkamp and his colleagues in 1984 (Feldkamp et al., 1984) and based on the foundations by Nalcioglu & Cho (1978) and Minerbo (1979) (Minerbo, 1979; Nalcioglu & Cho, 1978). They introduced also geometrical correction, regarding the position of the isocenter being not always in the central axis connecting the centers of the X-ray focal area and FPD array. In short, the 3DRA modality records multiple planar images while the gantry rotates about the object/patient and a computer performs reconstruction to present volumetric image digitally dissectible into axial, coronal, and sagittal planes the way CT does. Although essentially similar, trade names of a 3DRA feature vary between manufacturers; being known as ‘XperCT’ in Philips Allura Xper series (Philips Medical Systems, Eindhoven, the Netherlands), ‘DynaCT’ in Siemens Artis series (Siemens Medical Solutions Inc, Iselin, New Jersey), and ‘InnovaCT’ in GE Innova series of angiography modality models (GE Healthcare, Waukesha, Wisconsin). The mode of rotation also varies between manufacturers and models. A Philips Allura Xper’s gantry rotates over 240° (starts from -120°, ends at 120°) with the speed selectable at 10°/s and 20°/s. Siemens Artis Zee acquires 3DRA image by rotating the gantry over a 200° rotation (starts from -100°, ends at 100°), also with selectable rotation speed also at 10°/s and 20°/s. For GE Innova, the gantry rotation angle is approximately 210° (starts from -120°, ends at 90°) with fixed rotation speed at 40°/s. Patient image acquisition using 3DRA generally requires the patient to be completely still, preceded by antero-posterior and lateral position checks (for Philips and Siemens systems only) and a gantry clearance check (for all three manufacturers) to prevent gantry collision during acquisition. The acquisition is 5

typically simulated (gantry rotating without beaming) prior to the actual 3DRA acquisition. Figure 2.1 illustrates how a 3DRA image is typically acquired. Figure 2.1. Illustration of (a-d) image acquisition in 3DRA and (e) resulting image displayed in axial, sagittal, and coronal slices. For the three manufacturers, the radiation-related exposure parameters are pre-selected as part of mode/sub-mode/anatomical region selection rather than governed by an Automatic Brightness Control (ABC) commonly functional in planar modes. These standard parameters are tube voltage, tube current, and filtration thickness; and are essential in determining the resulting image’s visibility as well as directly influencing the radiation dose exposed to the patient. 2.3 Radiation Dose Measurements for 3DRA Generally, radiation energy (or ‘dose’) received by materials are expressed by the term ‘absorbed dose’, being a deterministic measure that can be applied for either directly or indirectly ionizing radiation. Absorbed dose is closely related to the stochastic energy unit being deposited in the medium. The absorbed dose is defined as the mean energy imparted e as a consequence of any interaction between incoming ionizing radiation with the medium of mass m that constitutes a specific volume V (Attix, 1986; International Atomic Energy 6

Agency, 2007; Johns & Cunningham, 1983; The International Commission on Radiation Unit and Measurement, 2005), D = de . (1) dm The energy imparted e is the sum of all particular energies entering the volume of interest subtracted by the sum of energies leaving it. Upon surpassing a medium, electrons deposit their energy along their path. The absorption of this deposited energy along the electron trajectory is not included in the term KERMA (Kinetic Energy Released per Unit Mass) in the same volume. The unit of the absorbed dose is expressed in joule per kilogram (J/Kg), commonly known as Gray (Gy). The dose rate, in turn, is applied to express the amount of dose received by a medium in a given duration (International Atomic Energy Agency, 2007; Johns & Cunningham, 1983). As the theoretical incident air kerma and entrance surface air kerma formalism applies directly for planar radiography and fluoroscopy, dosimetry for rotational X-ray beam configurations requires further deduction. A method by Shope, Gagne and Johnson (1981) introduced how a certain manner of measurement and calculation is required to obtain the information on radiation dose level in a rotating X-ray tube beam configuration. The CTDI (computed tomography dose index) was theoretically described as a representation of the photon fluence energy distribution over a circular volume of interest (Shope et al., 1981). The method covers measurement and calculation for axial and helical mode of parallel X-ray beams (also known as ‘fan-beam’ X-ray), and was adopted as the standard method of CT dosimetry by the International Electrotechnical Commission (IEC) (International Electrotechnical Commission, 1999), as well as by the International Commission of Radiation Unit and Measurement (ICRU) (The International Commission on Radiation Unit and Measurement, 2005). The International Atomic Energy Agency (IAEA) later included the methodology in its publication on the ‘international dosimetry code of practice for diagnostic radiology’ (International Atomic Energy Agency, 2007) with a slight alteration of notations. When first introduced, CTDI was mathematically described as 7

CTDI= 1  D(z)dz , (2) nT − where D(z) is the dose profile along the z-axis (centered at z = 0). The number of detector channels and the width of each channel are, respectively, n and T. Thus, nT represents the nominal total collimated X-ray beam width of the CT. In the first edition of IEC 60601–2–44 standard (1999) the CTDI was defined for an integration length of 100 mm as =CTDI100 1 50 mm . (3) nT D(z)dz −50 mm To represent volumetric photon fluence distribution, the weighted CTDI was then described as CTDI100,w = 1 + 2 CTDI100, p (4) 3 CTDI100,c 3 Figure 2.2. Illustration of (a) CTDI measurement in phantom, with (b) actual photograph of CTDI100,c measurement. Where the subscripts c and p denote pencil-type ion-chamber measurement results of the CTDI100 in the central and peripheral holes of the CTDI reference phantoms, respectively. In practice, the CTDI100,p is the mean CTDI100 from the four peripheral holes in the reference CTDI phantoms (Figure 2.2). 8

The issue emerging afterwards was the inability of this formulation to address the concept of cone-beam CT (CBCT), which later serves as a technological basis for the 3DRA imaging mode. The formal CTDI method did not include all axial-direction scattered radiation generated from the primary beam whenever X-ray beam z-direction width exceeds 40 mm, since the method was developed during the times when wide-beam CT was not available. Therefore, the IEC amended the dosimetry standard through the IEC 60601–2– 44 standard (2012) with a correction to address wide-beam scatter by the given formulation CTDI100,nT = CTDI100,ref   CTDI air ,nT  (5)  CTDI air ,ref  , where the term ‘nT’ refers to any use of detector channel number (n) and width (T) that exceeds 40 mm, and ‘ref’ refers to the reference configuration utilizing either equal to or less than 40 mm. The CTDIair is a metric similar to CTDI100,c, differing only by the absence of a phantom during the measurement, and, thus, is described as =CTDIair 1 L/2 (6) nT D(z)dz , −L/2 where L is at least the nT of a single scan with an additional 40 mm, and the X- ray beam centered at z = 0. In short, this formulation utilizes an understanding that in-phantom measurement and in-air measurement share the identical ratio of primary and scattered radiation between the use of the wide X-ray beam (>40 mm) and parallel X-ray beam (≤ 40 mm). The marvel of this formulation for CT, however, could not be extended to other modalities involving CBCT, such as dental CBCT, radiotherapy CBCT, and 3DRA. This issue was raised mainly due to the fact that these modalities mostly do not evolve a full-circle during image acquisition—therefore losing their eligibility to be included for CT-based dosimetry formalism. It is not until recently that the international medical physics communities agreed on publishing an international quality control guideline for non-CT CBCT modalities, within which a radiation dosimetry method for 3DRA is also discussed. 9

The quality control protocol for CBCT was composed by the international working group of the European Federation of Organization for Medical Physics (EFOMP), collaborating with the Society for Radiotherapy and Oncology (ESTRO) and IAEA in 2017 (de las Heras Gala et al., 2017). On its second edition, it introduces also the method of measuring radiation dose output metric based on DIN 6868-161 protocol (de las Heras Gala et al., 2019), with the metric being described as “dose to the isocenter” or DFOV. The DFOV is essentially the average radiation output from the X-ray tube corrected by geometrical aspects to arrive at an averaged dose over the diameter of the FOV (field of view). A solid-state based dosimeter is suggested to be used on the measurement to first measure accumulated air KERMA at the source- dosimeter distance, Ka,i(SDD). Geometrical correction should then be performed using the following mathematical formula, DFOV = K a,i (SDD)  b  d , (7) a c Figure 2.3. Illustration of DFOV measurement geometry. where a being the distance between the X-ray tube focal spot to the isocenter, b is the distance from X-ray tube focal spot to the point of measurement, c is the 10

horizontal diameter of the scanned volume, and d being the horizontal diameter of the radiation field at the measurement point. The geometry of the measurement is illustrated in Figure 2.3. 2.4 Image Quality Metrics in 3DRA For performance evaluation or mode selection studies, the information of radiation dose, expressed in metrics, should be accompanied by a degree of perception of the resulting image quality. Since the general term ‘quality’ is, by itself, qualitative and subjective, there is a need to employ quantitative metrics. Interestingly, any digital image can be considered a mathematical matrix in a spatial domain from which such metrics could be calculated. In the ideal scene of raw digital radiologic imaging, the image signal (S) is linearly correlated with the incoming photon fluence N, while the noise () is present as statistical fluctuation around the mean number of photons. In view of the photons being distributed according to Poisson’s law, the quantity of  can be found as N . The signal-to-noise ratio (SNR) is commonly referred to as the simplest image quality metric denoting how a signal (i.e. pixel value of a homogenous object in an image) stands out against its background value (Bushberg et al., 2011; Richard et al., 2012b; Verdun et al., 2015). Thus, SNR is mathematically expressed as SNR  S = N = N . (8) N In the ideal situation, each photon arriving the image detector will contribute to the image by forming a signal. However, physical factors including quantum interaction probabilities and detector inefficiency hinder the ideal phenomena to be found in experimental setting. This presents us with the terms SNRIdeal and SNRReal, where the latter is always less than the former. Based on the two definitions, one may eventually describe the efficiency of an imaging modality as the ratio between the two SNRs—that is, the ratio between the number of incoming photons to the detector (NReal) and the number of photons that actually contribute to the image forming, (NReal, also called noise- 11

equivalent quanta, NEQ). With a prior knowledge that NReal is equal to the squared SNRReal, this quantity is widely referred to as detective quantum efficiency (DQE), and is described mathematically as DQE = SNRR2eal = NEQ . (9) SNRI2deal N Ideal Another image quality index commonly used is the signal to noise ratio (SNR) and signal difference to noise ratio (SDNR). While the SNR is defined as the ratio of signal intensity between an object of interest and its background, SDNR refers to statistical difference of the SNRS of a particular object and its background and is widely used to identify the object’s detectability. Gislason- Lee with her colleagues (2010, 2013) used SDNR in FGI applications as (Gislason-Lee et al., 2013; Gislason et al., 2010) SDNR = NB − NO , (10) ( )SDO2 + SDB2 2 where NB is the background pixel value defined as the mean pixel value of the ROI (Region of Interest) outside the object of interest, NO is the object pixel value, which is the mean pixel value of the ROI drawn inside the object of interest, SDB is the background standard deviation, which is the standard deviation of the background’s pixel value NB, and SDO is the object standard deviation, which is the standard deviation of the object of interest’s pixel value ROI, NO. While the above metrics quantify an object’s detectability in an image in terms of contrast (i.e., contrast resolution), the spatial resolution is defined as the ability to distinguish two separate objects and is better described in Fourier domain. Whereas it is extremely challenging to measure spatial resolution in an analogue image (e.g. a film), objective metrics representing spatial resolution of digital images are present. The modulation transfer function (MTF) has become one of the standard performance and image quality metrics employed for X-ray based medical imaging systems (Friedman et al., 2013). The MTF is a metric describing the resolution of linear and shift-invariant imaging systems in the Fourier domain, and the three-dimensional (3D) MTF, MTF(u, v, w), is defined 12

in terms of the 3D system point-spread function, PSF(x, y, z) as in equation (11) (Rossmann, 1969) MTF (u,v, w) = PSF ( x, y, z) . (11) The PSF itself describes the response of an imaging system to a point source or point object. In practice, MTF can be calculated from the image of an edge (edge spread function, ESF), a line (line spread function, LSF), or a point (point spread function, PSF) (Verdun et al., 2015). In CT applications, MTF measurements were first carried out by differentiating an edge spread function (ESF) into a line- spread function (Judy, 1976). Afterwards, many works have introduced manners of CT MTF measurement methods, including software-based automated methods (Friedman et al., 2013; Grimmer et al., 2008; Mieville et al., 2010; Nakaya et al., 2012; Richard et al., 2012a; Shilfa et al., 2019; Takenaga et al., 2015). Since 3DRA is 3D imaging like CT, these image quality metrics described above apply also for 3DRA and are explored in this dissertation. 2.5 Quality Control and Tools for 3DRA Although separate studies regarding dose measurement and image quality evaluation for 3DRA have been performed in the last several years, literature addressing QC protocols specifically for 3DRA is currently very scarce. In some cases, QC for 3DRA is described within the QC protocol for the interventional fluoroscopy modality, but not separately. One protocol that has addressed this issue was the EFOMP-ESTRO-IAEA CBCT QC protocol in 2017 as described in the previous section (de las Heras Gala et al., 2017). The protocol also incorporates image quality checks as part of interventional CBCT QC methods, where the Catphan® 600 was mentioned as one of the suggested tools besides the QUART DVT_AP phantom and vendor-provided phantoms. The CBCT electron density and image quality phantom system, although being promulgated primarily for radiotherapy CT, is even mentioned. However, there was no mention of the specific needs associated with the use of interventional 3DRA, and more in particular the use of Iodinated contrast agents that does not exist in other CBCT modalities such as dental CBCT and radiotherapy on-board 13

imaging systems. The presence of Iodinated contrast agents in interventional 3DRA is, in fact, crucial in determining the exposure parameters as some systems employ the AERC (automatic exposure rate control). The present high contrast due to Iodinated contrast agent would drive the exposure factor into certain specific parameters unique to interventional systems, therefore requiring a specific phantom to check the performance. Another important issue with the available phantoms is that they might not provide adequate image representation for low contrast in 3DRA since they are dedicated for CT with difference in X- ray spectra, rotation extent, and beam collimation. 14

MATERIALS AND METHODS The dissertation consists of five studies, each with specific aims towards the general aim. The first part addresses the material selection and material characterization to be used in the phantom. The second work is a study on dose and image quality effects from phantom geometrical variations. The results will be used to decide upon the next generation phantom design in terms of shape. While the two works essentially serve as preliminary studies, a subsequent primary study is conducted for the phantom’s manufacturing and testing on a CT modality. In the next part, the phantom’s use as QC tool for 3DRA is investigated. Finally, the in-house phantom is used as a tool for mode selection among various head-dedicated protocols for 3DRA. As these studies involved different sets of materials and methods, each study will be described separately in this chapter. All medical X-ray modalities involved in this dissertation had been declared fit for use by regulatory authorities. 3.1 Material Selection and Testing This part of the dissertation was focused on characterizing materials mimicking the following body tissues: adipose tissue, muscle, brain (white and grey matter), liver, as well as typical phantom materials: polymethyl methacrylate (PMMA) and solidated water. The study was generally aimed to select materials to be used in the designed phantom. The candidate surrogate materials were composed using raw materials of traditional Javanese Batik cloth production, namely rice flour, full-refined paraffin, cecek wax and gondhorukem wax (resina colophonium). The study first measured and/or calculated mass density (), Hounsfield Unit (HU), electron density (e), and effective atomic number (Zeff) of the materials. Subsequently, for water- and PMMA-equivalent materials, additional linear attenuation coefficient () measurements were performed to confirm findings from e and Zeff calculations. 15

3.1.1 Materials preparation The raw materials (full-refined paraffin, cecek, and gondhorukem wax) were obtained from standardized suppliers for Batik cloth factories (Adhi Batik, Yogyakarta, Indonesia), while rice flour was a standard factory product for home cooking supplies (Rose Brand, Jakarta, Indonesia). To produce the phantom material candidates, cecek wax, full-refined paraffin and gondhorukem wax were heated separately to a temperature of 78°C as shown in Figure 3.1(a). All materials were then mixed with rice flour and with each other with various ratios to produce solid compounds after cooling down to room temperature, resulting in slabs as shown in Figure 3.1(b). Table 3.1 presents the mixtures of organic materials as well as their ratios and a given phantom label (name). It should be noted that not all combinations and ratios were possible due to clotting during mixing and brittleness after cooling down. These materials were disregarded as no further casting or shaping would be possible in a later stage. The preliminary study was performed by scanning all compounds using the CT modality of a Siemens Intevo 6 SPECT-CT scanner in the “Dharmais” National Cancer Center, with the aim to identify suitable mixtures to mimic the tissues of interest. Figure 3.1. (a) melting process of gondhorukem wax, and (b) samples of cooled, mixed material cast into slabs. 16

The phantom material candidates for further testing were selected based on their HU compared with literature values. In order to study the reproducibility of the production process, each sample was produced three times and the mean CT number of the three samples was calculated along with the standard deviation. 3.1.2 Mass density measurement Evaluation of mass densities of the produced samples was performed with a simple Archimedes’ method. Weighed samples (mass m) were submerged in a measuring cup filled with water to measure the volume. The mass density  is defined as the ratio of the mass m and volume V as expressed in Eq. (12). =m (12) V Table 3.1. Overview of the candidate phantom material compounds from full-refined paraffin (FP), cecek wax (CW), gondhorukem (G) and rice flour (RF) with their mixing ratios. Compound Raw Materials materials ratio A1 FP/CW 80/20 A2 FP/RF 100/0 B1 CW/RF 100/0 B2 CW/RF 90/10 B3 CW/RF 85/15 C1 CW/RF 82/18 C2 CW/RF 72/28 C3 CW/RF 70/30 C4 CW/RF 60/40 D1 G/RF 100/0 D2 G/RF 90/10 D3 G/RF 80/20 17

D4 G/RF 60/40 E1 G/CW/RF 70/20/10 E2 G/CW/RF 50/10/40 3.1.3 Electron density and effective atomic number calculations As described by Saito and Sagara (2017), this method is performed by CT scanning the materials using two energies (Möhler et al., 2017; Saito & Sagara, 2017). The two resulting images contain photon attenuation coefficient information from two different energies, and a simple one-parametric linear superposition algorithm of DECT images, or -blending, were utilized. The CT number information from the two resulting images was used to calculate the attenuation coefficients by using  = (ˆs −1)1000HU , (13) with  being the CT number (in Hounsfield Units, HU) generated from the resulting CT images and ˆs being the attenuation coefficient relative to water, namely µs/µwater. The index s represents {h, l} with h referring to the data from images obtained using higher energy and l from lower energy. The method starts with calculating the constant α that is a calibration factor correlating the  difference of two images (  , which is equivalent with ˆh − ˆl ) and the electron density e . Thus, obtaining the value of  only requires a correlation curve between  and e acquired by scanning a phantom containing objects with known e , with higher and lower energies. With this  parameter measured, the e of the phantom material candidates can be collected by scanning them with two energy levels, which are identical to the ones used for calibration and then using e = ˆh + (1− ) ˆl . (14) 18

In this work, calibration was performed using the CIRS 062M electron density phantom (CIRS, Norfolk, USA). To subsequently calculate the effective atomic number (Zeff), Saito and Sagara introduced the reduced CT number uk given by uk = HU k +1. (15) 1000 Similar to the index s, the index k represents {h, l} with h referring to the data from images obtained using higher energy, and l from lower energy. The Zeff was subsequently calculated using  Zeff m −1= ( L ) uk − 1 , (16)  Zeff ,w  e which requires the material-independent proportionality constants,  L , a fitting parameter m, and effective atomic number of water, Zeff,w. Similar to , the proportionality constant,  L was obtained from a linear regression approach, i the gradient slope of the calibration curve where (uk /e – 1) and {(Zeff / Zeff,w)m – 1} are on respectively the x- and y-axis. As suggested by Möhler et al. (2017), the optimum m value of 3.3 was used in this work (Möhler et al., 2017). The schematic workflow of the DEEDZ method is shown in Figure 3.2. Since the aim was to fund a combination mimicking human tissues, the calculated e and Zeff of all phantom material candidates were then compared with the reference data obtained from the calibration phantom manual and the ICRU Report No. 46, respectively (International Commission on Radiation Units and Measurement, 1992). Two particular phantom material candidates were then selected, namely materials with respectively high and low deviations from their reference values, for confirmation with the linear attenuation coefficient () measurements. 19

Figure 3.2. Workflow of DEEDZ method as proposed by Saito and Sagara (2017) (Saito & Sagara, 2017) Further investigation of e and Zeff could use stochiometric methods (Yohannes et al., 2011), but this method cannot be applied as we have no information on the hydrogen content of the in-house test material. Therefore, linear attenuation coefficient measurements were performed to verify eventual deviations between candidate materials and test materials. 3.1.4 Linear attenuation coefficient measurement Measurements were performed with an Ajex X-ray tube model OX/110- 15 with maximum tube voltage of 125 kV and focal spot size of 1.8 mm (Gyeonggi-do, Korea). Inherent filtration on the X-ray tube was equivalent to 1.5 mm Al. A 2 mm-thick lead collimator with 1 mm circular orifice was made in- house using a computer numerical control (CNC) milling system and was attached to the X-ray tube opening to produce a narrow-beam configuration. Edges, produced from the phantom material candidates and with ten thickness steps of 10 mm increment, were positioned such that the beam exit surface was kept at a constant 500 mm distance from the focus. For in-air dose measurements, a calibrated Radcal® 10X6-180 with cross-section of 100 cm2 and volume of 180 cm3 was positioned on the phantom beam exit, i.e., at 1500 mm from the X-ray tube’s focal spot. A laser pointer was used to ensure that the collimated X-ray beam coincided with the center of the ion chamber. There was a gap between the slabs and the ion chamber to ensure that the radiation quantity arriving at the ion chamber was scatter-free. The schematic measurement apparatus is shown in Figure 3.3. 20

Exposures were made with constant tube current and varied material thickness (0 – 100 mm with 10 mm increment) for a particular tube voltage. This is to fulfill the Lambert-Beer’s Law I = I0e−x , (17) and provide information on the ratio between measured dose with (I) and without materials presence ( I0 ) versus material thickness (x). With simple algebraic operations, Eq. (17) can be written as ln  I  = − x , (18)  I0    and the linear attenuation coefficient  can be obtained as the correlating factor on the plot where x and ln(I/I0) serve as the x- and y-axis, respectively. Since  is energy-dependent, the evaluation was repeated for a range of diagnostic X-ray mean energies. Therefore, the tube voltage was varied from 40 kVp to 120 kVp with increments of 10 kVp and the tungsten anode spectral model using the interpolating polynomial (TASMIP) method by Boone and Seibert (1997) was employed to estimate the mean energy of each beam (Boone & Seibert, 1997). The results of measurements on the materials selected for this test were compared with those of the standard phantoms. Figure 3.3. Schematic apparatus for linear attenuation coefficient measurement. 21

3.2 Study on Dose and Image Quality of Elliptical Phantoms The purpose of this part of study is to highlight how dose and image quality metrics behave on CT for phantoms with various sizes and shapes to consider for the subsequent design of the 3DRA phantom. Six physical phantoms (three cylindrical and three elliptical phantoms) were constructed to accommodate the dose and image quality evaluation. As for dosimetry, measurements were performed with a method similar to weighted computed tomography dose index (CTDI100,w) measurements over varied tube voltages and elliptical phantom aspect ratios. Dose coefficients were calculated for each elliptical phantoms to accompany its future use for dose estimation. Image quality was evaluated from the modulation transfer function (MTF), the signal to noise ratio (SNR), and noise homogeneity (H). 3.2.1 Experimental phantoms Three elliptical phantoms and three cylindrical phantoms (Figure 3.4) made of polymethyl methacrylate (PMMA) with identical thickness of 15 cm were constructed with different diameters as listed in Table 3.2. For the elliptical phantom, the effective diameters were determined by (Tudor & Thomas, 2004) Deff = 2 AB . (19) A+ B with A the length of the minor axis and B the major axis. On each phantom, four holes with 9 mm diameter were drilled to accommodate a Radcal® 10X6-3CT ionization chamber on four IEC standard positions (i.e., 12 o’clock, 3 o’clock, 6 o’clock, and 9 o’clock positions). The phantom allows for 8.5 cm × 8.5 cm square image quality test insert and a dose module. The image quality module contains resin-casted copper wire of 0.19 mm diameter for the MTF measurement, while the dose module is made of stacked PMMA with a hole for the ionization chamber in a central position. 22

Table 3.2. Geometrical information of the constructed phantoms. Phantom Minor axis Major axis Deff (cm) Aspect ratio, no. length, A (cm) length, B (cm) (A/B) Cylindrical 16 1. 16 16 21 1 2. 21 21 26 1 3. 26 26 1 Elliptical 23.2 4. 21 26 28.3 1.24 5. 26 31 33.3 1.19 6. 31 36 1.16 3.2.2 Dose trend study The study on dose trend in elliptical phantom was performed by using a Siemens Somatom 16 CT scanner (Siemens Medical Systems, Erlangen, Germany) operated in the Radiology Department of the Cibinong Regional General Hospital, Cibinong, Indonesia. Throughout the study, axial scanning was applied with 10 mm beam collimation at 200 mA and 1 s gantry rotation speed. The tube voltage was varied at all available voltage options, i.e., 80 kVp, 110 kVp, and 130 kVp. A calibrated Radcal® ionization chamber type 10X6-3CT, was used to measure CTDI100 at all positions and weighted dose index, CTDI100,w with measurement method illustrated in Figure 2.3 and Equation (4) (International Atomic Energy Agency, 2007; Shope et al., 1981; The International Commission on Radiation Unit and Measurement, 2005). For the purpose of calculation, the peripheral CTDI (CTDI100,p) were renamed according to their positions (Figure 3.4) with CTDI100,lat and CTDI100,ave given by CTDI100,lat = CTDI100,3 + CTDI100,9 (20) 2 and 23

CTDI100,ave  CTDIW = C1CTDI100,c + C2CTDI100,12 +C3CTDI100,lat + C4CTDI100,6 . (21) The CTDI100,w coefficients (namely C1, C2, C3, and C4) were chosen as comparative measure among phantoms and tube voltage variations. For coefficients to be applied on 12 o’clock and 6 o’clock positions (i.e. C2 and C4), different coefficients were evaluated to anticipate the impact of scattering from patient table presumed to exist. To calculate the coefficients by using C1 = CTDI100,c , (22) CTDI100,ave (23) (24) C2 = CTDI100,12 , (25) CTDI100,ave C3 = CTDI100,lat , CTDI100,ave C4 = CTDI100,6 , CTDI100,ave the values of cross-sectional average dose, CTDI100,ave, were interpolated from the Monte Carlo simulation results by Markovich et al. (2017) (Markovich et al., 2017). Since the coefficients for cylindrical phantoms are not expected to change (Haba et al., 2017), only the elliptical phantoms were included in this part of study. 24

Figure 3.4. Measurement positions of CTDI100 at the center and peripheral positions. Measurements on each phantom and tube voltage variations were performed three times. Hence, three sets of Eq. (21) were obtained for each variation. With the addition of its boundary condition that the sum of all the coefficient equals unity, a set of equations (Eq. 25) was obtained for every variation and solved by using the simple arithmetical substitution and elimination method with C1, C2, C3, and C4 as unknowns, i.e., CTDI100,ave = C1CTDI100,c(1) + C2CTDI100,12(1) + C3CTDI100,lat(1) + C4CTDI100,6(1) CTDI100,ave = C1CTDI100,c(2) + C2CTDI100,12(2) + C3CTDI100,lat(2) + C4CTDI100,6(2) CTDI100,ave = C1CTDI100,c(3) + C2CTDI100,12(3) + C3CTDI100,lat(3) + C4CTDI100,6(3) 1 = C1 + C2 + C3 + C4 (26) 3.2.3 Image quality study A Philips Ingenuity 128 CT scanner (Philips, Best, the Netherlands) operated in Radiology Department, Cibinong Regional General Hospital, Cibinong, Indonesia, was used in the image quality part of this study. Axial scanning was applied with 8 mm beam collimation at 200 mA and 1 s gantry rotation speed. The tube voltage was varied at all available voltage options, i.e., 80 kVp, 100 kVp, 120 kVp, and 140 kVp. A 512 × 512 matrix was selected to compose the produced image. 25

A placement of ROIs of 2.5 cm in diameter to determine SNRs on the resulting phantom image was performed in the central position as well as four peripheral positions shown in Figure 3.5. The SNRs were calculated by using SNRi = PVmean,i PVstdev,i (27) As a comparative measure, the homogeneity H was selected to represent how the SNRs of all measurement positions differ from one to another and was calculated by using 5 (28) (29)  SNRi SNR = i=1 5 H = CoV = ( )5 2 SNRi − SNR i =1 4 SNR with PV as pixel value. 26

Figure 3.5. ROIs placement for SNR analysis. Additionally, SNRAP and SNRLAT were determined by using SNRAP = SNR2 + SNR4 2 (30) and SNRLAT = SNR3 + SNR5 2 , (31) for separated analysis of noise in antero-posterior and lateral position of the phantoms. To determine the MTF, an automated software, SPICE-CT, was used as plug-in on the Image-J java-based quantitative image analysis software (Loveland, 2011). The 10% cut-off MTF was chosen as a comparative measure among different phantom and tube voltages. 27

3.3 Phantom Design, Construction, and Evaluation The in-house phantom was designed to accommodate the need for dose measurement and image quality evaluation in volumetric imaging, including 3DRA. The material and geometry of the phantom were decided based on the results of previous studies. 3.3.1 Phantom material and design The IEC Standard 6061-2-44 dictates that the CT dosimetry phantom should be made of polymethyl methacrylate (PMMA) (International Electrotechnical Commission, 2009). Therefore, the in-house phantom is designed as a solid PMMA cylinder. All dimensions strictly follow the standard head dosimetry phantom (International Atomic Energy Agency, 2007; International Electrotechnical Commission, 2009; The International Commission on Radiation Unit and Measurement, 2005). Four inserts (one dosimetry insert and three image quality inserts) are designed to be positioned inside the core phantom, filling the 8.5 cm × 8.5 cm square space. For dosimetry use, the dosimetry insert is positioned inside the core phantom, and for image quality evaluation purposes, the three image quality inserts can be stacked and positioned at the same place, using separate acquisitions. The image quality insert for electron density evaluation provides eight materials mimicking water, PMMA, bone, lung (air), adipose, liver, muscle/grey matter brain and white matter brain. Physical test regarding volumetric density, CT number, attenuation coefficients, and effective atomic numbers of these materials are discussed in Section 4.1. All objects are 15 mm in diameter. The contrast resolution insert contains four groups of holes with 10 mm, 8 mm, 6 mm, 4 mm, and 2 mm diameters. Each group was filled with cast resin mixed with 0.25 ml, 0.50 ml, 0.75 ml, and 1.0 ml of iodine contrast agents (Iohexol 350 mg/ml concentration) to produce four different contrast levels in each object sizes. The MTF insert contains a copper wire of 0.19 mm diameter at the center of cast resin. The three image quality inserts can be positioned inside the core phantom with no specific sequence. 28

3.3.2 Phantom testing Both the dose and image quality measurement capability of the in-house phantom were evaluated. Dose measurement results were compared with a standard head CTDI phantom using a calibrated Radcal® 10X6-3CT ionization chamber (Radcal, Monrovia, California, USA), while measured image quality aspects were compared with Catphan® 604 (The Phantom Laboratories, Salem, New York, USA). Both tests were conducted on a Toshiba Aquilion 64 CT scanner (Toshiba Medical System Corp., Ōtawara, Tochigi, Japan) operated in the Radiotherapy Department, Pasar Minggu Regional General Hospital, Jakarta. The CT modality used in this study has passed annual calibration test and compliance test. For the dose measurement capability test, dose readings in five standard measurement positions, i.e., center position, 12 o’clock, 3 o’clock, 6 o’clock, and 9 o’clock, were compared with measurement results using the standard phantom. Exposures were performed on 120 kVp tube voltage, 200 mA tube current, 1 s gantry rotation time, and 8 mm beam collimation. These parameters comply with the CTDI measurement part of typical compliance test protocols for CT scanners. The electron density linearity, contrast linearity and MTF were measured in both the in-house phantom and the reference Catphan® 604 phantom. Evaluation was performed quantitatively by comparing linear regression coefficients from the electron density evaluation and contrast resolution assessment inserts, as well as the MTF values between those obtained from in-house phantom and Catphan®. The linear regression coefficients for the electron density module were obtained by measuring individual pixel values of the objects and plotting them together. For contrast resolution assessment, signal difference to noise ratio (SDNR) calculated by using Eq. (10) were evaluated. 3.4 Novel Phantom for Performance Evaluation of 3DRA In this part we want to explain why dedicated tests are needed for 3DRA testing. An in-house phantom with a specially designed contrast-object module was constructed to address the needs and tested alongside a commercially available phantom for conventional CT testing. Since neurological studies using both CTA and 3DRA often involve challenging radiological tasks such as 29

defining morphological deformations of small intracranial vessels with Iodine- based contrast agents, our in-house phantom was equipped with a module consisting of simulated iodine-filled blood vessels of various contrast levels and sizes for detectability studies in terms of signal difference to noise ratio (SDNR). This study also outlines the 3DRA and CTA image result differences by means of frequency domain imaging performance metrics, such as the modulation transfer function (MTF) for spatial resolution, and the noise power spectrum (NPS) for noise magnitude and texture. 3.4.1 Study steps An in-house phantom has been constructed to address the hypothesized special needs of contrast-enhanced 3D angiographic modalities. The phantom, as explained in the next section, was designed to address 3D angiographic clinical tasks for which we had not found any other dedicated commercial phantom. Reference 3DRA and CTA images were acquired with a Catphan® (Phantom Laboratories, Salem, USA) [15]. Although not being primarily designed for 3DRA, the Catphan® was chosen as comparative tool due to the fact that it is generally used by medical physicists in 3DRA QC due to the absence of task- specific phantoms. Hence, the Catphan® is available in the radiology departments. In the comparison step, quantitative analysis was performed on Catphan® CTP515 module images over three adjacent slices. The 9 mm diameter objects for 0.3%, 0.5%, and 1.0% contrast were evaluated. For 3DRA images of the Catphan®, the ROIs were made in the positions where the objects were expected to be, using the manual book as guidance. Images were acquired using the same modes and exposure parameters and described in Section 3.4.3. Difference in results between Catphan® and in-house phantom was statistically analyzed using student’s t-test. 3.4.2 Use of the in-house phantom The new phantom consists of a PMMA cylinder with a core that is in part detachable. In a first study, a core had been used that had four holes for the pencil type ionization chambers at the locations for standard CT dosimetry positions [16]. This part of the phantom is handy, but not further discussed here. As discussed in Section 3.3, two central modules were specifically constructed for the 3DRA study. The first module is the MTF module, consisting of a resin block with a copper wire of 0.19 mm in diameter for system spatial resolution check (Figure 3.6). The second module is a task-based visibility module, 30

consisting of four groups of artificial vessels for contrast and visibility evaluations. They were made by filling the computer numerical control (CNC)- drilled holes of different diameters with different iodine-based contrast agent concentrations. Each artificial vessel group consists of five artificial vessels of 10 mm, 8 mm, 6 mm, 4 mm, and 2 mm in diameters, filled with a mixture of cast resin (95% resin and 5% catalyst) and varied in contents of iodinated contrast agent, i.e., 0.25 ml, 0.50 ml, 0.75 ml, and 1.00 ml of Ultravist® 370 Iopromide (370 mgI/ml). The iodine contrast agent concentrations were selected by inspecting the Hounsfield units of some prototype artificial vessels. They were adjusted to match iodine-filled vessels in brain CT scans. Table 3.3 shows the nominal iodine concentrations. The nominal iodine concentration is presented as the content of iodine from its contrast agent solution form (370 mg/ml) against the total volume of mixture. Figure 3.6. In-house phantom constructed with the contrast object module (right) presents four groups of iodine contrast levels, each with five object sizes. 31

Table 3.3. Iodine content and concentrations used in the phantom module. Group Iodine content Resin content (ml) Nominal iodine (ml) concentration 9.75 (mg/ml) 1 0.25 9.50 9.25 2 0.50 9.25 18.50 3 0.75 9.00 27.75 4 1.00 37.00 3.4.3 Image quality evaluation using the in-house phantom for 3DRA purposes To acquire the images of the in-house phantom on 3DRA modality, a GE Optima CL323i angiography system with 3DRA capability (GE Healthcare, Chicago, Illinois, USA) operated in the R. Syamsuddin S.H. Regional General Hospital was used, while a Toshiba Aquilion 64 (Toshiba Medical System Corp., Ōtawara, Tochigi, Japan) operated in the Pasar Minggu Regional General Hospital, Jakarta, was used to obtain the CTA images. Both modalities, referred to in this work as '3DRA' and 'CTA', had passed an annual compliance test within less than a year before this study was carried out. Since the 3DRA modality has only a filtered-back-projection (FBP) algorithm, this algorithm was selected as the mode of reconstruction for both modalities. As both modalities have a different mode of acquisition, the parameters and settings were selected from the console-provided neurological protocols, as shown in Table 3.4. These protocols were routinely used both clinically and during onsite quality control. Although both modalities are intended for cranial applications, the 3DRA one makes use of a relatively different beam spectrum (HVL 3.0 mmAl at 71 kVp) than the CTA (HVL 5.1 mmAl at 120 kVp). 32

Table 3.4. List of exposure and image acquisition parameters. Compared parameters Toshiba Aquillon 64 GE Optima Exposure parameters CL323i Protocol Brain routine Cerebral, normal Tube voltage detail 120 kVp 71 kVp Tube current (modulated) Tube focal spot size 200 mA 60 mA Inherent physical filtration 0.8 mm 0.6 mm HVL at selected voltage 1.1 mmAl 0.3 mmCu Pulse width 5.1 mmAl 3.0 mmAl Frame rate - 7 ms Gantry rotation - Continuous 360° 210° (from -120° Rotation time to 90°) Rotation speed 1.00 s 5.25 s Image acquisition parameters 360°/s 40°/s Acquisition mode Beam collimation Axial Cone-beam Field of view (at isocenter) 1×8 mm 300 mm Flat-panel detector size 320 mm 174 mm - 300 mm Recons. Algorithm (diagonal) Matrix size FBP FBP Slice thickness 512×512 512×512 0.50 mm 0.32 mm 3.4.4 Imaging performance measurement methods MTF measurements were carried out using the Spice-CT ImageJ plugin (Loveland, 2011), with regions of interest (ROIs) selection shown in Figure 3.11(a). Noise magnitude and texture in terms of noise power spectrum (NPS) were measured using the Imquest software developed by the Duke University 33

team (Ria et al., 2020; Solomon et al., 2018), as shown in Figure 3.7(b). Four identical, automatically generated ROIs were used over ten images for the 3DRA system and six images for CT, all from repeated acquisitions. In noise texture quantification, the average noise frequency (favg) and peak noise frequency (fpeak) were also generated from the software and employed as a comparison parameter in addition to the noise spectral shape. The fave is the frequency at which noise is found to be the average value across all frequencies, while the fpeak is the frequency at which the largest noise is found according to the measurement. As a surrogate for object visibility, the SDNRs were calculated from the artificial vessels of the in-house phantom using (Gislason-Lee et al., 2013; Gislason et al., 2010; Lubis et al., 2015; Lubis et al., 2018; Vano, Geiger, et al., 2005) SDNRi = PVBG,i − PVO,i , (32) ( )SDO,i2 + SDBG,i2 2 with SDNRi being the signal difference to noise ratio of object i, PVO,i the mean pixel value of the object i, PVBG,i the mean pixel value of the background area adjacent to the object i, SDO,i the standard deviation of the pixel values in object i, and SDBG,i the standard deviation of the pixel value of the background area adjacent to object i. An example of ROI selection to determine vessel visibility in terms of SDNRs is shown in Figure 3.8. The statistical error was determined as the standard deviation from measurements using images of five adjacent slices. 34

Figure 3.7. Example from 3DRA images showing rectangular regions of interest (ROI) selections. Measurements of MTF (a) and NPS (b) were carried out using Spice-CT Image-J plugin using the in-house phantom. Figure 3.8. ROI selection in determining SDNR using the in-house phantom. 35

3.5 Selection of 3DRA Task-Based Imaging Modes This final study was aimed to use the produced and tested in-house phantom to assist with an actual clinical issue in the hospital. On site, the presence of two modes, both intended for cranial applications with similar protocol name (‘cerebral’ and ‘head limited’ with no explanation on what the phrase ‘limited’ represent), had caused some degree of difficulty with the clinicians and operators on deciding which mode to select for which task. This in-house phantom was used in this study to complement the dose descriptor information on the mentioned task-based imaging modes with image quality information for further analysis and selection recommendations. 3.5.1 Modality and phantoms The study made use of the in-house phantoms that had resulted from previous parts of the dissertation. The 3DRA modality investigated was the same modality used in the fourth study part (GE Optima CL323i). The modes in question to be chosen along with detailed exposure parameters are given on Table 3.5. Selected modes were the most clinically relevant and frequently used. Head- dedicated CT-like acquisition modes, namely ‘Cerebral’, and ‘Head limited’ were available for use, often confusing radiologists and operators. While the ‘Cerebral’ mode was manufacturer default, the ‘Head Limited’ mode is an additional mode created by the local vendor technicians. The word ‘Limited’ was reportedly referring to ‘limited dose’. For any modes employed, two image contrast options (‘normal’ and ‘low’) were selectable, making a total of four selectable modes. With the automatic tube modulation on, exposure factors change with the presence of phantom and for the modes enlisted in Table 3.5. The used focal spot size was 0.6 mm, while the selected frame rate was ‘continuous’. By default, the gantry rotates 210° (from -120° to 90°) with rotation speed of 40°/s and a total of 5.25 s of exposure time. 3.5.2 Dose metric measurement and spectrum simulation The DFOV was measured using the method described in the EFOMP- ESTRO-IAEA CBCT quality control protocol (de las Heras Gala et al., 2019). A calibrated Unfors-Xi dosimeter (Raysafe, Uggledalsvägen, Billdal, Sweden) was used as air KERMA measurement tool. The measurement scheme is shown in Figure 3.13(a), while the area of the beam at image detector surface required for 36

geometrical correction to arrive at DFOV was measured using Gafchromic XR- RV3 radiochromic film (Ashland Advanced Materials, New Jersey, USA) as shown in Figure 3.13(b). Figure 3.13. Illustration of distances required in DFOV measurement; a being the distance from tube focal point to isocenter, b being distance from tube focal point to dose detector position, c being the calculated at- isocenter beam width, and d being the at-detector beam width. The value of d was measured using radiochromic films (b). As indicated in the EFOMP-ESTRO-IAEA protocol, the DFOV was measured without phantom and calculated using (de las Heras Gala et al., 2017, 2019) DFOV = K a,i (SDD)  b  d , (33) a c with Ka,i(SDD) being the accumulated air KERMA at the source-dosimeter distance, a being the distance from tube focal point to isocenter, b being distance from tube focal point to dose detector position, c being the calculated at-isocenter beam width, and d being the at-detector beam width. The obtained results were calculated into KAPDFOV by multiplying them with dose area measured from radiochromic film. The KAPDFOV will then be compared with the corresponding KAP value displayed on the console to investigate the agreement between 37

KAPDFOV and displayed KAP and further to decide whether or not the KAP calculated from DFOV can be used as dose metric. X-ray spectra of all 3DRA imaging modes were simulated using the tungsten anode spectral model using the interpolating polynomial (TASMIP) method as described by Boone et al. (1997) (Boone & Seibert, 1997). The information of tube peak voltage (in kVp), filter material and thickness (in cm), as well as tube output (in mAs) was included in the simulation. The calculated average beam energy and spectrum of each mode will be observed together with the mass attenuation coefficient curves of Iodine and ICRU-44 soft tissue from NIST data by Hubbel and Seltzer (1996) (Hubbel & Seltzer, 1996) to analyze the appropriateness of the mode with respect to the produced Iodine-tissue contrast. 3.5.3. Image quality quantification To represent the object visibility, the SDNR was calculated from the artificial vessels on the in-house phantom images using (Gislason-Lee et al., 2013; Gislason et al., 2010; Lubis et al., 2015; Lubis et al., 2018; Vano, Geiger, et al., 2005) SDNRi = PVBG,i − PVO,i , (34) ( )SDO,i2 + SDBG,i2 2 with SDNRi being the signal difference to noise ratio of object i, PVO,i the mean pixel value of the object i, PVBG,i the mean pixel value of the background area adjacent to the object i, SDO,i the standard deviation of the pixel values in object i, and SDBG,i the standard deviation of the pixel value of the background area adjacent to object i. 38

RESULTS AND DISCUSSIONS 4.1 Material Selection and Testing Results The measured CT Numbers for all produced phantom material candidates are shown on Table 4.1 for 130 kVp. Compared to reference values (International Atomic Energy Agency, 2014) almost all of the produced materials are within typical HU ranges of certain phantom materials and tissues. Compounds A1, C1, C4, E2, C2, and E1 were selected to represent adipose, water, liver, PMMA, brain (white matter) and muscle or brain (grey matter), respectively. Apart from A1, C4 and E2, all compounds were selected for closeness of measured CT number with reference mean values. Compounds A1, C4 and E2 were selected because other candidates, i.e. A2, D3 and D4, failed to form robust samples after cooling and had tendency to deform despite their mean  being closer to literature values. In the remainder of the text, the compound names are no longer used. Each compound will be referred to as the material(s) they have been selected to represent. Evaluation of mass density for selected phantom candidate materials are summarized in Table 4.2. The average mass densities of soft tissue surrogate materials (and their deviations against reference ICRU Report 46) were 0.94 g/cm3 (0.96%), 0.97 g/cm3 (2.52%), 1.00 g/cm3 (4.31%), 0.95 g/cm3 (8.86%), 1.03 g/cm3 (2.71%) and 1.18 g/cm3 (0.44%) for adipose, water, muscle, brain and liver, and PMMA-equivalent materials, respectively. With maximum discrepancy being less than 10%, selection of these materials to apply the DEEDZ method is confirmed. For calibration purposes, prior to e and Zeff calculation with the DEEDZ method, the CIRS 062M electron density phantom was scanned,  was measured and e was calculated for each object summarized in Table 4.3. The alpha-blending parameters, α and L, had the values of 1.927 and 9.278, respectively. It can be seen from Table 4.3 that the materials with e lower than water tend to have higher calculated e while higher calculated e was 39

obtained for the materials with e higher than water. Except for PMMA, all test materials had more than 15% deviation against the ICRU Report 46 values. PMMA- and water-equivalent materials were then chosen for further investigation as they are commonly used in the standard phantom studies and they have relatively low (3.3%) and high (22.2%) deviations. Figure 4.1 compares the linear attenuation coefficient measurement of the materials. Standard PMMA slabs were used as reference material for the candidate PMMA- equivalent material, while the candidate water-equivalent material was compared with measurements of Plastic Water® LR slabs (CIRS, Norfolk, USA). Table 4.4. Results of the DEEDZ method for the phantom material candidates. The average electron density (e) and atomic effective number (Zeff) of all materials were compared to reference data of human body tissues from ICRU. The relative deviation (  ) between the measured and reference values of e and Zeff, are also given. Test materials ρe Zeff  Lit. Lit. (%) Calc.  (%) Calc. Adipose (A1) 0.933 0.930 -0.20 6.230 5.120 - ±0.006 ±0.101 17.81 Water (C1) 1.000 1.025 +2.47 7.420 5.771 - ±0.000 ±0.015 22.22 Liver (C4) 1.064 1.086 +2.02 7.740 6.054 - ±0.000 ±0.013 21.79 PMMA (E2) 1.147 1.15 +0.20 6.470 6.254 -3.34 ±0.002 ±0.057 Brain, white 1.035 1.058 +2.20 7.580 5.928 - matter (C2) ±0.000 ±0.004 21.80 Muscle/brain, 1.041 1.065 +2.27 7.590 5.807 - grey matter ±0.000 ±0.007 23.49 (E1) 40

Figure 4.1(c) shows that the produced PMMA-equivalent material shows a small deviation as a function of photon mean energy retrieved using TASMIP method. This makes the material a good surrogate candidate, presumably due to proper simulation of the photoelectric regime in the diagnostic X-ray energy range. This also confirms the Zeff calculation result on Table 4.4, where the PMMA-equivalent material demonstrates a low deviation from the ICRU Report 46 values. On the other hand, water-equivalent material, with >22% Zeff discrepancy in Table 4.4, is shown to have more than 20% deviation from the reference material in terms of  in the lower photon mean energy portion. Materials with higher deviation of calculated Zeff were identified as the materials with higher HU deviation in lower energy CT images. This result was supported by  measurement result presenting higher deviation (>15%) on lower energy. It should not escape one’s attention that the Zeff calculation result can be used not only to indicate the suitability of produced phantom material candidates, but also to distinguish the potential use of such materials in terms of energy range. In our case, DEEDZ method had provided the information that the produced PMMA-equivalent material is shown to be representative of general PMMA in terms of radiation-related properties, and that the water-equivalent material could be suggested for the higher-order diagnostic energy range, bit is inadvisable for lower energy use, i.e., mammography. With these results at hand, it was decided to use PMMA for the primary material of the phantom while other materials are proposed as the phantoms’ insert for a variety of tissue-mimicking objects. 41