Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Patient-specific internal dose calculation techniques for clinical use in targeted radionuclide therapy Grimes, Joshua 2013

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2013_spring_grimes_joshua.pdf [ 3.93MB ]
Metadata
JSON: 24-1.0073590.json
JSON-LD: 24-1.0073590-ld.json
RDF/XML (Pretty): 24-1.0073590-rdf.xml
RDF/JSON: 24-1.0073590-rdf.json
Turtle: 24-1.0073590-turtle.txt
N-Triples: 24-1.0073590-rdf-ntriples.txt
Original Record: 24-1.0073590-source.json
Full Text
24-1.0073590-fulltext.txt
Citation
24-1.0073590.ris

Full Text

PATIENT-SPECIFIC INTERNAL DOSE CALCULATION TECHNIQUES FOR CLINICAL USE IN TARGETED RADIONUCLIDE THERAPY by Joshua Grimes  B.Sc., The University of Guelph, 2006  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Physics)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  February 2013  © Joshua Grimes, 2013  Abstract The objective of this thesis was to investigate and develop a set of methods for accurate dose calculation that would be practical to implement into routine clinical use. Towards this aim, a graphical user interface (GUI) was developed in order to handle the large body of data associated with internal dose calculations and to perform each step in the dose calculation procedure. Furthermore, an iterative adaptive thresholding method for determining volumes and activities of objects in single photon emission computed tomography images was developed and the accuracy and reproducibility of this method was investigated. Next, organ level and voxel S value dose calculations were compared to results from Monte Carlo simulation. This comparison included the assessment of various aspects of OLINDA/EXM dose calculation such as the use of stylized reference phantoms used to represent the average patient and the sphere model used to estimate tumour dose. Finally, the internal dosimetry techniques investigated in this thesis were applied to estimate patient-specific absorbed doses in patients imaged for suspected neuroendocrine tumours and in patients treated with  188  Re  microspheres. The proposed iterative adaptive thresholding method was found to accurately and reproducibly determine object volumes and activities in phantom and patient studies, regardless of the image reconstruction method used. In the comparison of OLINDA/EXM to Monte Carlo dose calculations, variation in patient-specific anatomy was found to lead to large differences between cross-organ dose estimates based on stylized phantoms in OLINDA and corresponding cross-organ dose estimates calculated for each patient with Monte Carlo. However, total organ doses agreed within 6%. Monte Carlo and voxel S value dose calculations were found to produce nearly identical 3-dimensional dose distributions within source organs. Analysis of the patients scanned for neuroendocrine tumours revealed large dose variations in tumours and normal organs between patients. The ratio of tumour-to-kidney dose ranged from 0.13 to 2.9, demonstrating the importance of determining patient-specific dose estimates. Finally, response data in the patients treated with  188  Re microspheres did not  correlate with mean or maximum tumour dose, but it was found to correlate with the minimum dose received by 90% of the tumour volume (D90).  ii  Preface A version of Chapter 3 was published as a book chapter: Celler A, Grimes J, Shcherbinin S, Piwowarska-Bilska H and Birkenfeld B. (2013) Personalized Image-Based Radiation Dosimetry for Routine Clinical Use in Peptide Receptor Radionuclide Therapy: Pretherapy Experience. In Baum RP and Rosch F (eds.), Theranostics, Gallium-68, and Other Radionuclides, Recent Results in Cancer Research (pp. 497-517) Springer-Verlag Berlin Heidelberg. Dr. Anna Celler wrote the introduction of this book chapter, Dr. Sergey Shcherbinin wrote the section on “Image Reconstruction” and I wrote the remainder of the book chapter. The work presented in Chapters 5 and 6 was published as a single paper: Grimes J, Celler A, Shcherbinin S, Piwowarska-Bilska H and Birkenfeld B. (2012) The Accuracy and Reproducibility of SPECT Target Volumes and Activities Using an Iterative Adaptive Thresholding Technique. Nucl Med Commun 2012; 33:1254-1266. I wrote most of the manuscript with contributions from Dr. Anna Celler and Dr. Sergey Shcherbinin. A version of Chapter 8 has been published: Grimes J, Celler A, Birkenfeld B, et al. (2011)  Patient-Specific  Radiation  Dosimetry  of  99m  Tc-HYNIC-Tyr3-Octreotide  in  Neuroendocrine Tumors. J Nucl Med 2011; 52:1474-1481. I wrote most of the manuscript and coauthored the introduction with Dr. Bozena Birkenfeld. The patient studies used for the work described in Chapters 5-8 were performed by collaborators in Szczecin, Poland at the Pomeranian Medical University, and these studies were approved by the Ethics Review Board at that institution. Use of the electronic data associated with these patient studies was approved by the University of British Columbia’s Clinical Research Ethics Board (ID: H08-01642). Additionally, the work described in the following chapters will be submitted for publication, starting with a version of Chapter 4: Grimes J and Celler A. A Graphical User Interface for Internal Dose Calculations. I developed the software described in this work and wrote the manuscript. A version of Chapter 7 will be submitted for publication: Grimes J and Celler A. Calculation of Patient Absorbed Doses Calculated Using Three Methods: Organ Level,  iii  Voxel S Values and Monte Carlo Simulation. I planned and performed this experiment and wrote the manuscript. A version of Chapter 9 will be submitted for publication: Shcherbinin S, Grimes J, Bator A, Cwikla JB and Celler A. Three-Dimensional Personalized Dosimetry for  188  Re Hepatic  Selective Internal Radiation Therapy Based on Quantitative Post-Treatment SPECT Studies. Dr. Jaroslaw Cwikla and Dr. Andrzej Bator performed the phantom and patient studies at the Department of Radiology and Diagnostic Imaging of the Hospital of Ministry of Internal Affairs & Administration in Warsaw, Poland, and the patient studies were approved by the Clinical Ethics Committee at this institution. Use of the electronic data associated with these patient studies was approved by the University of British Columbia’s Clinical Research Ethics Board (ID: H08-01642). Dr. Sergey Shcherbinin analyzed the phantom studies to assess the camera dead time and performed the image reconstructions. I performed dose calculations based on these reconstructed activity distributions. Furthermore, the work described throughout this thesis has been presented at several international meetings including the 2012 European Association of Nuclear Medicine (EANM) congress in Milan, Italy (Eur J Nucl Med Mol Imaging. 2012; 39 (Supplement 2):S322); the 2011 joint meeting of the American Association of Medical Physicists and the Canadian Organization of Medical Physicists in Vancouver, BC (Med Phys. 2011; 38(6):3472); the 2010 EANM congress in Vienna, Austria (Eur J Nucl Med Mol Imaging. 2010; 37 (Supplement 2):S367); the 2010 Society of Nuclear Medicine (SNM) meeting in Salt Lake City, UT (J Nucl Med. 2010; 51 (Supplement 2):1439); and the 2009 SNM meeting in Toronto, ON (J Nucl Med. 2009; 50 (Supplement 2):1874).  iv  Table of Contents Abstract .................................................................................................................................... ii Preface ..................................................................................................................................... iii Table of Contents .................................................................................................................... v List of Tables ........................................................................................................................... x List of Figures ........................................................................................................................ xii List of Abbreviations ........................................................................................................... xix Acknowledgements .............................................................................................................. xxi Dedication ............................................................................................................................ xxii Chapter 1: Introduction ........................................................................................................ 1 1.1  Aim ........................................................................................................................... 2  1.2  Outline of Dissertation .............................................................................................. 3  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy............................ 4 2.1  Historical Perspective ............................................................................................... 4  2.2  Physics of Nuclear Medicine Imaging and Therapy ................................................. 7  2.3  Radioactive Decay .................................................................................................... 9  2.3.1  Nuclear decay mechanisms ................................................................................... 9  2.3.2  Atomic emissions ................................................................................................ 10  2.3.3  Decay equations .................................................................................................. 11  2.4  Interaction of Radiation with Matter....................................................................... 11  2.4.1  Charged particle interactions .............................................................................. 12  2.4.2  Photon interactions with matter .......................................................................... 13  2.5  Production of Radionuclides ................................................................................... 17  2.6  Radionuclides Used in SPECT Imaging and Therapy ............................................ 18  2.7  Components of the Gamma Camera ....................................................................... 21  2.7.1  Collimator ........................................................................................................... 21  2.7.2  Radiation detection system ................................................................................. 22  2.7.3  Positioning and energy analysis .......................................................................... 24  2.7.4  Data storage and image display .......................................................................... 25  2.8  SPECT/CT Image Acquisition ................................................................................ 25  v  2.8.1  Hardware ............................................................................................................. 25  2.8.2  Data acquisition parameters ................................................................................ 27  2.9  Image Reconstruction ............................................................................................. 27  2.9.1  Image formation process ..................................................................................... 28  2.9.2  Reconstruction algorithms .................................................................................. 29  2.10  Image Degrading Factors ........................................................................................ 32  2.10.1  Attenuation and scatter ................................................................................... 32  2.10.2  Collimator detector response .......................................................................... 33  2.10.3  Noise ............................................................................................................... 34  2.10.4  Partial volume effect ....................................................................................... 36  2.10.5  Camera dead time ........................................................................................... 37  2.11  Quantitative Corrections in SPECT Imaging .......................................................... 38  2.11.1  Attenuation correction .................................................................................... 38  2.11.2  Scatter correction ............................................................................................ 39  2.11.3  Collimator detector response compensation ................................................... 40  2.11.4  Partial volume effect correction ...................................................................... 40  2.11.5  Dead time correction ....................................................................................... 41  2.12  Absolute Quantification .......................................................................................... 41  2.13  Summary ................................................................................................................. 42  Chapter 3: Internal Dose Calculations .............................................................................. 43 3.1  General Concepts in Internal Dosimetry................................................................. 43  3.2  Acquiring the Time-Integrated Activity ................................................................. 44  3.2.1  Imaging protocol ................................................................................................. 45  3.2.2  Segmentation of nuclear medicine images ......................................................... 46  3.2.3  Time-activity curves ........................................................................................... 48  3.3  Dose Estimation Method......................................................................................... 53  3.3.1  Methods for organ level dose estimation ............................................................ 54  3.3.2  Voxel S values .................................................................................................... 55  3.3.3  Monte Carlo simulation ...................................................................................... 56  3.4  Software Tools for Internal Dose Calculations ....................................................... 57  3.5  Patient-Specific Dose Calculation Protocol ............................................................ 59  vi  3.6  Summary ................................................................................................................. 61  Chapter 4: A Graphical User Interface for Internal Dose Calculations......................... 63 4.1  Introduction ............................................................................................................. 63  4.2  Overview of the Dosimetry Tool ............................................................................ 63  4.3  Description of the Major GUI Functions ................................................................ 65  4.3.1  Main GUI ............................................................................................................ 65  4.3.2  Planar region selection GUI............................................................................... 69  4.3.3  Planar image registration GUI........................................................................... 70  4.3.4  Process 3D image data GUI ............................................................................... 71  4.3.5  Organ level dose calculation GUI ...................................................................... 75  4.4  Results and Discussion ........................................................................................... 76  4.5  Conclusion .............................................................................................................. 79  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities in SPECT Images .......................................................................................... 80 5.1  Introduction ............................................................................................................. 80  5.2  Materials and Methods ............................................................................................ 81  5.2.1  Calibration phantom experiment......................................................................... 81  5.2.2  SPECT image reconstruction and activity quantification ................................... 82  5.2.3  Processing calibration phantom data .................................................................. 83  5.2.4  Description of the iterative adaptive thresholding technique ............................. 85  5.2.5  Validation phantom experiment .......................................................................... 88  5.2.6  Patient studies ..................................................................................................... 89  5.2.7  Dependence of volume and activity estimates on the initial parameters ............ 89  5.3  Results ..................................................................................................................... 91  5.3.1  Analysis of the calibration phantom experiment ................................................ 91  5.3.2  Validation of the iterative adaptive thresholding technique ............................... 91  5.3.3  Influence of initial parameters on the final volume and activity estimates ........ 92  5.4  Discussion ............................................................................................................... 94  5.5  Conclusion .............................................................................................................. 97  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates Derived From SPECT Images ............................................................................................. 98  vii  6.1  Introduction ............................................................................................................. 98  6.2  Methods................................................................................................................... 98  6.2.1  Patient studies ..................................................................................................... 98  6.2.2  Repeatability of volume and activity estimates .................................................. 99  6.2.3  Reproducibility of different reconstruction methods .......................................... 99  6.2.4  Dosimetry.......................................................................................................... 100  6.3  Results ................................................................................................................... 100  6.3.1  Repeatability of volume and activity estimates ................................................ 100  6.3.2  Reproducibility of volume, activity and dose estimates ................................... 101  6.4  Discussion ............................................................................................................. 103  6.5  Conclusion ............................................................................................................ 103  Chapter 7: Comparison of Dose Estimates Obtained Using Organ Level, Voxel S Value and Monte Carlo Techniques ............................................................................................. 104 7.1  Introduction ........................................................................................................... 104  7.2  Methods................................................................................................................. 105  7.2.1  Patient studies ................................................................................................... 105  7.2.2  Calculation of TIACs ........................................................................................ 105  7.2.3  Absorbed dose calculation ................................................................................ 106  7.2.4  Evaluation of dose estimation methods ............................................................ 107  7.3  Results ................................................................................................................... 109  7.3.1  Comparison of organ level S values ................................................................. 109  7.3.2  Total organ and tumour dose assessment.......................................................... 113  7.3.3  Paired organs ..................................................................................................... 116  7.3.4  Monte Carlo and voxel S value comparison ..................................................... 117  7.4  Discussion ............................................................................................................. 118  7.5  Conclusion ............................................................................................................ 120  Chapter  8: Patient-Specific Dosimetry of  99m  Tc-HYNIC-Tyr3-Octreotide in  Neuroendocrine Tumours .................................................................................................. 122 8.1  Introduction ........................................................................................................... 122  8.2  Background to Neuroendocrine Tumour Imaging ................................................ 122  8.3  Methods................................................................................................................. 123  viii  8.3.1  Patient studies ................................................................................................... 123  8.3.2  Determination of the time-activity curves and effective half-lives .................. 124  8.3.3  Activity quantification and dose estimation ..................................................... 125  8.4  Results ................................................................................................................... 126  8.4.1  Time-activity curves ......................................................................................... 126  8.4.2  Organ segmentation .......................................................................................... 129  8.4.3  Dose calculation ................................................................................................ 130  8.5  Discussion ............................................................................................................. 132  8.6  Conclusion ............................................................................................................ 134  Chapter 9: Patient-Specific Dosimetry for Radioembolization with  188  Re Microspheres  ............................................................................................................................................... 135 9.1  Introduction ........................................................................................................... 135  9.2  Radioembolization Overview ............................................................................... 135  9.3  Methods................................................................................................................. 136  9.3.1  Patient studies ................................................................................................... 136  9.3.2  Image reconstruction ......................................................................................... 137  9.3.3  Activity quantification ...................................................................................... 138  9.3.4  Dose calculation ................................................................................................ 139  9.3.5  Data analysis ..................................................................................................... 139  9.4  Results ................................................................................................................... 140  9.4.1  Dose estimates .................................................................................................. 140  9.4.2  Dose-response ................................................................................................... 141  9.5  Discussion ............................................................................................................. 142  9.6  Conclusion ............................................................................................................ 143  Chapter 10: Conclusions and Future Work .................................................................... 145 10.1  Conclusions ........................................................................................................... 145  10.2  Future Work .......................................................................................................... 146  Bibliography ........................................................................................................................ 149  ix  List of Tables Table 2.1 Common radionuclides for SPECT imaging and therapy ...................................... 20 Table 4.1 Dosimetry protocol options offered in dosimetry toolkit presented in this work. .. 64 Table 4.2 Example output from organ level absorbed dose calculation. ................................ 78 Table 4.3 Organ dose estimates obtained using the Organ level dose calculation GUI and Monte Carlo simulation with the Process 3D image data GUI. ........................... 79 Table 5.1 Summary of parameters for the different reconstruction methods. ........................ 83 Table 5.2 Curve fit parameters for threshold-SBR data from the calibration phantom experiment reconstructed using three different algorithms. .................................. 91 Table 5.3 Interobserver comparison of tumour and left kidney volume and activity estimates. ............................................................................................................................... 94 Table 6.1 Summary of region volumes and the number of samples segmented for each region. .................................................................................................................. 100 Table 6.2 Repeatability analysis showing percentage differences in volume and activity estimates for kidneys, liver and spleen derived from two consecutive SPECT images of patients in group 1. ............................................................................. 101 Table 6.3 Reproducibility analysis showing percentage differences in volume, activity and dose estimates for organs (kidneys, liver, spleen and thyroid) and tumours derived from SPECT images of patients in groups 1 and 2 reconstructed using different methods. .............................................................................................................. 102 Table 7.1 Time-integrated activity concentrations for investigated regions in patients 1 to 6. ............................................................................................................................. 109 Table 7.2 Summary of percentage differences between the  99m  Tc patient-specific S values  calculated using Monte Carlo and the corresponding S values used by OLINDA/EXM and those calculated using voxel S values for each source and target region pair. ................................................................................................ 110 Table 7.3 Summary of percentage differences between the  131  I patient-specific S values  calculated using Monte Carlo and the corresponding S values used by OLINDA/EXM and those calculated using voxel S values for each source and target region pair. ................................................................................................ 111  x  Table 7.4 Summary of percentage differences between the  177  Lu patient-specific S values  calculated using Monte Carlo and the corresponding S values used by OLINDA/EXM and those calculated using voxel S values for each source and target region pair. ................................................................................................ 111 Table 7.5 Percentage differences between  99m  Tc patient-specific S values calculated by  Monte Carlo simulation and the corresponding reference S values used by OLINDA/EXM and those calculated using voxel S values for patients 2 and 4. 112 Table 7.6 Percentage differences between  131  I patient-specific S values calculated by Monte  Carlo simulation and the corresponding reference S values used by OLINDA/EXM and those calculated using voxel S values for patients 2 and 4. 113 Table 7.7 Percentage differences between 177Lu patient-specific S values calculated by Monte Carlo simulation and the corresponding reference S values used by OLINDA/EXM and those calculated using voxel S values for patients 2 and 4. 113 Table 7.8 Percentage difference between right and left kidney doses  calculated by Monte  Carlo for three radionuclides with corresponding biological half-lives in each patient. ................................................................................................................. 117 Table 8.1 Effective and biologic half-lives determined from monoexponential fits through tumours and normal organs. ................................................................................ 128 Table 8.2 Normal organ masses for males, females and corresponding reference phantom values. .................................................................................................................. 129 Table 8.3 Organ time-integrated activity coefficients. ......................................................... 131 Table 8.4 Relative absorbed doses. ....................................................................................... 131 Table 8.5 Tumour masses and doses as well as normal organ doses for patients with pathological uptake. ............................................................................................. 132 Table 9.1 Radioembolization patient information and dose parameters............................... 140  xi  List of Figures Figure 2.1 General overview of nuclear medicine procedures from the production of radioisotopes to the acquisition of nuclear medicine images and delivery of radionuclide therapy. ............................................................................................... 8 Figure 2.2 Mass attenuation coefficients for classical scattering, photoelectric effect, Compton scattering and pair production in H2O, Pb and NaI(Tl). (Data used for generating  these  plots  was  obtained  at:  http://www.nist.gov/pml/data/xcom/index.cfm). .................................................. 14 Figure 2.3 Photoelectric effect. ............................................................................................... 15 Figure 2.4 Compton effect. ..................................................................................................... 16 Figure 2.5 Polar plot of the Klein-Nishina distribution of scattering angle cross sections over a range of energies from 10 keV to 10 MeV, including selected energies of particular relevance to nuclear medicine. This distribution demonstrates how low energy photons (~10 keV) have nearly equal probability of being forward or backscattered at 0º and 180º, respectively. At 140 keV, photons are more likely scattered in the forward direction but may still backscatter. At even higher energies (> 1 MeV) photons are almost always forward scattered with little deviation from the incident photon direction. ....................................................... 17 Figure 2.6 A comparison of a high sensitivity and a high resolution collimator. The high sensitivity collimator on the left has a large acceptance angle resulting in relatively poor spatial resolution. The collimator on the right has longer and narrower holes, providing an improved spatial resolution at the cost of sensitivity. The dashed red lines represent photons absorbed by the collimator. .................... 22 Figure 2.7 Summary of events leading to an electric signal produced at the anodes of a photomultiplier tube (PMT) array following the detection of a 140 keV gamma ray. Typical values of scintillation efficiency (of NaI(Tl)), photocathode quantum efficiency and PMT dynode multiplication factors are used. ................................ 24 Figure 2.8 An example dual headed gamma camera with a computed tomography (CT) component from GE Healthcare (Infinia Hawkeye 4). The two detector heads are on a rotating gantry for single photon emission computed tomography (SPECT)  xii  image  acquisition.  (Adapted  from  image  obtained  at:  http://www3.gehealthcare.ca/enCA/Products/Categories/Nuclear_Medicine/SPECTCT_Cameras/Infinia_Hawkeye_4). ....................................................................... 26 Figure 2.9 Example 1-dimensional projection profiles obtained by a gamma camera rotating around a 2-dimensional activity distribution. The value of  at each point  is proportional to the total activity along the line of response (LOR) passing through each collimator hole (ignoring image degrading effects such as attenuation, scatter and collimator detector response, which are discussed in Section 2.10). ......................................................................................................... 28 Figure 2.10 General structure of iterative reconstruction methods......................................... 30 Figure 2.11 Series of projection images of an activity source demonstrating different sources of image degradation. The projection of the source in air (a) differs from the true image due to the poor resolution of the system. When the source is placed in water (b) attenuation reduces the number of counts and scatter reduces image contrast. The impact of image noise due to the random nature of radioactive decay and radiation detection are demonstrated by the statistical fluctuations in the measured projection in (c). .................................................................................................... 32 Figure 2.12 Distance dependent spatial resolution is demonstrated by the count profiles of a point source, which broaden with distance from the detector. The full width half maximum (FWHM) of these profiles can be used to characterize the spatial resolution of the system. In addition, the dashed red lines correspond to gamma rays that have undergone septal penetration, which further degrades the spatial resolution. .............................................................................................................. 34 Figure 2.13 Examples of low-pass filters used to suppress image noise, which dominates at high frequencies. The shape of the Hanning filter (a) is controlled by one parameter: the cutoff frequency  . The Butterworth filter (b) is controlled by  two parameters: the critical frequency  and the order . ................................... 35  Figure 2.14 Example SPECT image slices reconstructed by OSEM with no filter (a), and with Hanning (b) and Butterworth (c,d) postreconstruction filters applied. ......... 36  xiii  Figure 2.15 Comparison of a true object on the left with an image of this object on the right, demonstrating the influence of image sampling. Pixel intensity at the object boundaries is an average value taken from the source and background contributions. ......................................................................................................... 37 Figure 2.16 Observed counting rates functions of the true count rate  for paralyzable and nonparalyzable systems as ......................................................................... 38  Figure 3.1 Planar data with segmentation performed to determine the planar counts in regions of interest over a period of 22 hours after injection. ............................................. 45 Figure 3.2 Sample coronal slices from a SPECT image of the tumour in the abdomen of the patient displayed in Figure 3.1. The SPECT image provides information about the 3D activity distribution. ......................................................................................... 46 Figure 3.3 The trapezoidal method used to find the time-integrated activity. In this example, the time-integrated activity is equal to the sum of four areas labelled in the figure from A1 to A4. Elimination of radioactivity after the last data point (at 22 hours) is assumed to be by physical decay only, which is a conservative estimate that will most likely over-estimate the time-integrated activity. For  99m  Tc, the area under  the curve after the last data point (A4) is small due to the relatively short physical half-life of  99m  Tc. However, for longer lived radionuclides (eg.  177  Lu, T1/2 = 6.73  days) the area under the curve after the last data point can be significant. ........... 49 Figure 3.4 Use of the hybrid planar/SPECT technique to plot a time-activity curve (TAC). The shape of the TAC is first determined from the planar data (a) and then rescaled to pass through the quantitative SPECT-based activity measurement at the time of the SPECT acquisition (b)................................................................... 51 Figure 3.5 A flowchart depicting the general framework of the hybrid planar/SPECT approach. ............................................................................................................... 52 Figure 3.6 Reference female used for traditional organ level dose calculation. ..................... 55 Figure 3.7 Voxel S values are a table of precalculated doses to an array of voxels given activity in a single source voxel. ........................................................................... 56 Figure 3.8 Example coronal slices from a CT phantom and corresponding activity distribution used as input for patient-specific dose calculation with Monte Carlo simulation. ............................................................................................................. 57  xiv  Figure 3.9 Flowchart outlining the necessary steps in patient-specific dose calculation for radionuclide therapies. ........................................................................................... 60 Figure 4.1 Overview of workflow in the dosimetry toolkit between the 5 sub-GUIs, which are the main GUI, the Planar region selection GUI, the Planar image registration GUI, the Process 3D image data GUI and the Organ level dose calculation GUI. ............................................................................................................................... 65 Figure 4.2 Screen capture of the main GUI. ........................................................................... 66 Figure 4.3 Screen capture of the Planar region selection GUI. Anterior (left) and posterior (right) views from the first of three planar scans is displayed. A region of interest delineating the spleen and surrounding background is drawn (solid magenta line) on the posterior view and automatically copied to the anterior. In addition, a background region is drawn (solid green line) to be used for geometric background subtraction. ........................................................................................ 69 Figure 4.4 Screen capture of the Planar image registration GUI. The top left window displays the reference image, which in this case is a posterior view of the spleen in the first of three whole body planar scans. The bottom left window shows the image of the spleen from the second whole body planar scan registered to the reference image. This registration was performed using the semi-automatic algorithm, which shifted the image to register by one pixel in the vertical direction and rotated it by three degrees in the counter clockwise direction. The plots in the top right and bottom right windows are profiles through the reference image and the registered image, which serve to check the accuracy of the registration. ....... 71 Figure 4.5 Screen capture of the Process 3D image data GUI demonstrating how reconstructed SPECT images are analyzed. The user draws regions slice by slice in the top left window, with a choice of using a transaxial, coronal or sagittal view. The remaining two views are displayed in the top right and bottom right windows, which are used to check the slice position. The bottom left window is used for different purposes and in this case it shows outlines of the regions segmented from the nuclear medicine image on the corresponding CT slice. ...... 72 Figure 4.6 A sub-GUI that is called from the Process 3D image data GUI for creating DOSXYZnrc/EGSnrc input files. .......................................................................... 74  xv  Figure 4.7 Screen capture of the Organ level dose calculation GUI. ..................................... 75 Figure 5.1 (a) Transaxial SPECT slice of the image of a bottle, showing region boundaries delineated by thresholds ThV (dashed green line) and ThA (dotted red line). The white solid line corresponds to the region manually drawn by the operator. (b) A plot of the activity concentration profile drawn through this slice with labelled positions of the true volume and activity boundaries. ........................................... 84 Figure 5.2 Flowchart illustrating the iterative adaptive thresholding technique. ................... 87 Figure 5.3 Schematic diagram of validation setup showing an axial view of phantoms 1 and 2 with background concentrations of 12 kBq/mL and 9 kBq/mL, respectively. The numbers besides each container indicate its volume (mL) and activity (MBq) in parentheses. ........................................................................................................... 89 Figure 5.4 Functions relating threshold values ThV and ThA to SBR obtained from the phantom data reconstructed using the Clinical (a), BBACRR (b) and SCACRR (c) algorithms. ............................................................................................................. 91 Figure 5.5 The ratios of the measured to true volumes (a) and activities (b) for the whole range of values investigated in the validation experiment, as determined using ThV and ThA, respectively. The data were reconstructed using the Clinical, BBACRR and SCACRR methods. ........................................................................ 92 Figure 5.6 An example of a plot showing the convergence of the SBR estimated by the iterative adaptive thresholding technique for three different initial values of the SBR. ...................................................................................................................... 93 Figure 6.1 Example coronal slice from images reconstructed using SCACRR (a) and Clinical (b) methods. Segmented regions include the liver, left and right kidneys, spleen and a tumour, which are each delineated using ThV (solid magenta line) and ThA (white dashed line). ............................................................................................. 102 Figure 7.1 Coronal view of maximum intensity projections for patients 2 (a) and patient 4 (b) illustrating differences in patient-specific anatomy and relative uptakes of  99m  Tc-  HYNIC-TOC in the kidneys, liver and spleen. ................................................... 112 Figure 7.2 Total organ and tumour doses calculated by OLINDA/EXM and the voxel S value (VSV) method compared to mean doses from Monte Carlo simulation for  99m  Tc  (a) and (b), 131I (c) and (d), and 177Lu (e) and (f). One data point not visible in part  xvi  (a) is the OLINDA/EXM underestimation of 22% in the tumour dose of patient 2. Similarly in part (b), tumour dose in patient 2 was underestimated by 12% and is not displayed. ....................................................................................................... 115 Figure 7.3 Average percent contributions of self and cross doses to the total organ and tumour doses for simulation with 99mTc (a), 131I (b), and 177Lu (c). Note that the xaxis scale begins at 50%. ..................................................................................... 116 Figure 7.4 Coronal maximum intensity projections of patients 1 (a) and patient 6 (b) who had the largest differences between right and left kidney doses with 99mTc. ............. 117 Figure 7.5 Cumulative dose volume histograms based on 3D dose distributions calculated by Monte Carlo (dashed lines) and voxel S values (solid lines) for organs and a tumour analyzed in patient 3, using  131  I. For each region, the dose volume  histograms determined using each dose estimation method are nearly overlapping, demonstrating the similarity between the Monte Carlo and voxel S value dose distributions. ........................................................................................................ 118 Figure 8.1 Anterior (a, b and c) and posterior (d) whole body planar images acquired at approximately 1.5 hours after injection with pathological uptake indicated by arrows. (a) Neuroendocrine tumour in the right maxilla with metastasis to the left mastoid, both thyroid lobes, left subdiaphragmatic and lower intra-abdominal region in patient 1. (b) Neuroendocrine tumour in the small bowel of patient 24. (c) Neuroendocrine tumour in the small bowel with metastasis to the liver in patient 28. (d). Patient 5 has no visible lesions. In this case the posterior view is displayed to show the kidney uptake. .................................................................. 127 Figure 8.2 Examples of kidney and liver decay corrected time-activity data fitted with a monoexponential for five patients. Spleen time-activity data is plotted for three of these patients, also showing the use of a biexponential in addition to the monoexponential fit. ............................................................................................ 128 Figure 8.3 Dynamic planar plots for a patient with no visible lesions (a) and a patient with pathological uptake (b). ....................................................................................... 129 Figure 8.4 Segmentation of the left kidney on transaxial SPECT (a) and CT (b), and coronal SPECT (c) and CT (d) slices comparing the use of a fixed 40% threshold (white dashed line), ThV (solid magenta line), and ThA (dotted yellow line). .............. 130  xvii  Figure 9.1 Tumour dose volume histograms for ten radioembolization patients treated with 188  Re-HSA microspheres. .................................................................................... 141  Figure 9.2 Overall survival of all ten patients is plotted versus average tumour dose (a), maximum tumour dose (b), and D90. In the bottom row, overall survival data from a subset of seven patients with similar tumour volumes is plotted against average tumour dose (d), maximum tumour dose (e), and D90 (f). Of these data, only the plot of overall survival in the subset of patients versus D90 was found to have a statistically significant correlation of 0.75 (P < 0.05). ............................. 142  xviii  List of Abbreviations 2D  2-Dimensional  3D  3-Dimensional  APDI  Analytical Photon Distribution Interpolated  CT  Computed Tomography  D90  Minimum Dose Received by 90% of the Volume  DVH  Dose Volume Histogram  EBRT  External Beam Radiation Therapy  EC  Electron Capture  EGS  Electron Gamma Shower  FBP  Filtered Backprojection  FWHM  Full Width Half Maximum  GUI  Graphical User Interface  HSA  Human Serum Albumin  HYNIC-TOC  Hydrazinonicotinamide-Tyr3-Octreotide  IMRT  Intensity Modulated Radiation Therapy  IT  Isomeric Transition  LEAP  Low Energy All Purpose  LEHR  Low Energy High Resolution  LET  Linear Energy Transfer  MELP  Medium Energy Low Penetration  MIRD  Medical Internal Radiation Dose  MLEM  Maximum Likelihood Expectation Maximization  MRI  Magnetic Resonance Imaging  NET  Neuroendocrine Tumour  NHL  Non-Hodgkin’s Lymphoma  OSEM  Ordered Subsets Expectation Maximization  PET  Positron Emission Tomography  PMT  Photomultiplier Tube  PRRT  Peptide Receptor Radionuclide Therapy  xix  PSF  Point Spread Function  ROI  Region of Interest  SBR  Source-to-Background Ratio  SPECT  Single Photon Emission Computed Tomography  SSTR  Somatostatin Receptor  TAC  Time-Activity Curve  TIAC  Time-Integrated Activity Coefficient  TRT  Targeted Radionuclide Therapy  WB  Whole Body  xx  Acknowledgements I would like to acknowledge my supervisor, Dr. Anna Celler, for the continuous support she offered during the course of my PhD. Thank you Anna for all of your insight, encouragement and guidance. The hard work and countless hours you dedicate to each and every one of your students is extremely appreciated. I would also like to thank the other members of my PhD committee: Dr. Francois Benard, Dr. Stefan Reinsberg and Dr. Vitali Moiseenko for their valuable feedback and support. Also, thank you to the instructors in the Physics department and in the Medical Physics program for helping to make my time at the University of British Columbia a truly great and enriching experience. I am also grateful to the University of British Columbia for funding my research. Additional thanks goes to Dr. Sergey Shcherbinin. Thank you Sergey for always being there to share your mathematical expertise. It was a tremendous help. And if you weren’t sharing your knowledge of image reconstruction, you had a joke or story to share for every situation. This work would not have been possible without the help of our collaborators at the Pomeranian Medical University in Szczecin, Poland. Thank you to Dr. Bozena Birkenfeld and Hanna Piwowarska-Bilska for all of your effort in this fruitful collaboration. Thank you to my family for always being there. Thank you Mom and Dad for supporting me through all of these years of education. Also thank you to my sister, Meaghan, and grandparents for all of your encouragement. Finally, thank you to my wife, Jennifer, and daughter, Jada. Jen, I could not have done this without your endless love and support. Jada, you were born while I was in the middle of working on this PhD and you have brightened every day ever since.  xxi  Dedication  To Jennifer and Jada  xxii  Chapter 1: Introduction  Chapter 1: Introduction “We must learn to shoot microbes with magic bullets” -Paul Ehrlich (1854-1915) Paul Ehrlich’s search for a chemical compound that targeted disease while leaving the rest of the body unharmed led to his discovery that the synthesized arsenic compound, Salvarsan, could be used to treat syphilis [1]. He was the founder of chemotherapy and also postulated that toxins could be attached to antibodies in order to target and treat tumours. Today, many drugs exist that might be classified as what Ehrlich termed “magic bullets”. In the field of nuclear medicine, unsealed radioactive sources are used to label pharmaceuticals that are designed to target a specific tissue or physiologic function in the body. These radiopharmaceuticals are administered to patients for the diagnosis and treatment of disease. For diagnostic procedures, the radiopharmaceutical biodistribution is imaged using one of two classes of nuclear medicine imaging, either single photon emission computed tomography (SPECT), or positron emission tomography (PET). Each uses a detector positioned outside the patient to image the location of the radiation source inside the body by measuring its electromagnetic emissions (gamma rays or annihilation photons). In therapeutic nuclear medicine procedures, the radioactive source is usually a charged particle emitter. Charged particles have a relatively short range in tissue, which allows for a localized radiation dose to be deposited. There are certain advantages of targeted radionuclide therapy (TRT) compared to the other radiotherapy approaches performed with external photon or particulate beams, or with brachytherapy, which uses sealed radioactive sources placed near or inside cancerous tissues. First of all, TRT has the potential to deliver a radiation dose directly to tumours, while avoiding irradiation of healthy tissues. Furthermore, in addition to already identified tumour sites, the radiopharmaceutical’s affinity to a given receptor type allows it to target other, undiagnosed tumours that are expressing the same receptors and may be located anywhere in the body. An important advantage of TRT over other systemic therapies is the possibility to perform a pretreatment imaging study to plan  1  Chapter 1: Introduction  patient-specific treatment, optimize the dose to be administered and to predict the effectiveness of the therapy for a given patient. The goal of any radiation treatment is to deliver a high radiation dose to the diseased tissue without exceeding safe levels of radiation to the surrounding healthy tissues. To reach this goal, detailed patient-specific treatment plans are routinely performed in external beam radiation therapy (EBRT) and brachytherapy. However, in TRT, patient-specific dose calculations are seldom performed. Patients are usually administered fixed levels of activity, which may be adjusted by patient weight or body surface area. The fixed level of activity is usually based on previous experience with the maximum activity that can be safely injected into the most sensitive of patients. As a result, deleterious effects are usually avoided, but the majority of patients are actually under-dosed. One of the major reasons patient-specific internal dose calculations are not routinely implemented into clinics is due to the perception that these calculations are too difficult and time consuming to perform. They must be based on imaging studies, as the absorbed radiation dose delivered to tissues depends on the fraction of administered activity that will accumulate and be subsequently retained in or cleared from the tissue. This fraction depends on the radiopharmaceutical kinetics and tissue properties, which both are subject to many patient-specific factors that result in interpatient variability of the absorbed radiation dose per unit of administered activity. Personalized dosimetry has to take all of these patient-related variables into account. Additionally, it must consider physics-related factors such as the type (particles or gammas) and the energy of radioactive emissions.  1.1 Aim The objective of this thesis was to investigate and develop a set of methods for accurate dose calculation that would be practical to implement into routine clinical use. Towards this aim, a dosimetry tool in the form of a MATLAB graphical user interface (GUI) was created for performing internal dose calculations. As an important step in the dose calculation process, an innovative segmentation method for determining volumes and activities of objects in SPECT images has been proposed and implemented. The developed dosimetry tool was used to compare the results obtained from a variety of dose calculation techniques, including Monte Carlo simulation, voxel S value and organ level methods. Finally, the GUI  2  Chapter 1: Introduction  was used to estimate patient-specific absorbed doses for patients undergoing both diagnostic and therapeutic nuclear medicine procedures.  1.2 Outline of Dissertation This thesis is structured as follows. In Chapter 1, a brief introduction to the problem and the objectives of the thesis has been provided. Chapter 2 discusses the fundamentals of nuclear medicine imaging and therapy. This chapter begins with a historical overview of radiation therapy, which is intended to shed more light on the reasons individualized dose calculations in TRT are not often performed even though prescribed dose in EBRT often accounts for patient-specific characteristics. The remainder of Chapter 2 discusses the physical principles that make nuclear medicine imaging and therapy possible. Chapter 3 is the final introductory chapter of the thesis and provides background information about internal dose calculations. Here, details of the requirements and procedures for internal dosimetry are outlined. Furthermore, Chapter 3 discusses some of the limitations of internal dose calculations in current practice, which provides the motivation for this thesis. In Chapter 4, the software that was created for use in a wide variety of applications in internal dosimetry is presented. Chapters 5 and 6 focus on the innovative image segmentation technique that was developed as a part of this thesis. Chapter 5 describes the principles of the technique and the experiments that were performed for validation of this method. In chapter 6, the segmentation method is used to assess the repeatability and reproducibility of target volume and activity estimates using different reconstruction algorithms. Comparisons of organ and voxel level dose calculation techniques are explored in Chapter 7. Then, the methods described and developed throughout this thesis are used to perform patient-specific dose estimates in diagnostic and therapeutic procedures and the results from these calculations are reported in Chapters 8 and 9. Finally, conclusions and a discussion of potential future work are included in Chapter 10.  3  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy 2.1 Historical Perspective The birth of diagnostic and therapeutic uses of radiation can be traced back to the late 19th century. It began with the discovery of x-rays by the German physicist Wilhelm Conrad Roentgen in 1895. By the following year, the dermatologist Leopold Freund founded radiotherapy in Vienna, where he began therapeutic irradiation of a hairy nevus in a young girl [2]. Also in 1896, Henri Becquerel had discovered radioactivity, and by 1898, Marie and Pierre Curie had discovered radium. The first therapeutic use of radioactivity shortly followed when brachytherapy was used to treat tuberculous skin lesions with  226  Ra, starting  in 1901. Over the decades to follow, the use of brachytherapy expanded to a number of applications in radiation oncology and to this day remains an important treatment option for cervical, prostate, breast and skin cancers. A report of the first intravenous injection of an unsealed radioactive source was published in 1913 by Frederick Proescher, who injected radium into patients for the treatment of different diseases, including leukemia [3]. The 1920’s marked the first use of radioactive sources as biological tracers. In 1924, Georg de Hevesy used 210Pb and 210Bi to study radiotracer kinetics in animals. Then in 1925, the first diagnostic nuclear medicine procedure was performed by Hermann Blumgart, who measured human blood flow rates using  214  Bi. The radiation detector used by Blumgart in  this monumental work was a cloud chamber [4]. Ernest Lawrence invented the cyclotron in 1931, about a decade before the first nuclear reactor. This invention would lead to the production of usable quantities of radionuclides. In 1936, the two most widely used radionuclides in nuclear medicine,  99m  Tc and  131  I were  32  discovered. Then in the late 1930’s, P became the first artificially produced radionuclide to be employed for therapy when it was used to treat leukemia [5]. However it was found to have limited success. Meanwhile, in the field of external beam radiation therapy (EBRT) during the 1930’s, advancements were made in the understanding of radiobiology, leading to 4  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  widespread use of dose fractionation. Megavoltage EBRT began in 1937, making it possible to treat deep-seated tumours [6]. Radioiodine (131I) treatment of thyroid cancer and hyperthyroidism started in the early 1940’s [7]. The therapeutic window (the difference between tumour and healthy tissue dose) for this treatment is large since iodine localizes well in the thyroid allowing for the delivery of high radiation doses to the thyroid while sparing other tissues in the body. After World War II, nuclear reactors overtook the cyclotron as the primary method for producing radionuclides. At this time, it was anticipated that radionuclide therapies might revolutionize radiotherapy, given their potential to selectively target a diseased tissue with a therapeutic radiopharmaceutical. However, this proved to be a difficult task and for several decades the primary radionuclide therapy remained for diseases of the thyroid. Marinelli and Quimby published the first mathematical treatment of internal dose calculations in 1948 [8]. Around this time, the first attempts were made to prescribe  131  I  activities on an individual patient basis for the treatment of thyroid cancer and hyperthyroidism [9]. However, this individualized treatment was not found to improve patient outcome compared to a fixed administered activity approach because of large errors in the measurements required for dose calculation. As a result, individualized treatments fell out of popularity and thyroid disease has since been treated for decades using a fixed activity approach. Major technological advancements in the field of nuclear medicine were made in the 1950’s, particularly by Hal Anger, who developed the first scintillation camera in 1958. To this day, most commercial gamma cameras are built using a design that is largely based on Anger’s original idea. A significant hardware development was also achieved in EBRT around this time when the first linear accelerator (linac) was used to treat a patient in 1953 [6]. Medical linacs produce megavoltage beams and can be rotated around the patient to deliver radiation from multiple beam positions. In 1960,  99m  Tc generators became commercially available. Then in 1964, the first  radiotracers labelled with 99mTc were introduced. This represented an important development in nuclear medicine because of the desirable imaging properties and useful chemistry of 99m  Tc for studying many organs in the body.  5  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  The dose calculation system that defined how internal dosimetry was performed for many years to follow was developed by the Medical Internal Radiation Dose (MIRD) Committee of the Society of Nuclear Medicine. The MIRD Committee published its first pamphlet in 1968. Now, there are over 20 MIRD pamphlets describing internal dose techniques, models and equations. A significant step towards the goal of producing a “magic bullet” against cancer was achieved in 1975 when a technique was developed to produce monoclonal antibodies, which could be used to target tumour associated antigens. Throughout the 1980’s, radiolabelled monoclonal antibodies were used to treat patients with melanoma, non-Hodgkin’s lymphoma (NHL) and leukemia. Finally, in 2002, the 90Y labelled ibritumomab tiuxetan (Zevalin®) was FDA approved for treatment of patients with NHL. In the following year,  131  I-tositumomab  (Bexxar®) was approved for the same indication. Another approach in which there has been a growing interest since the 1990’s has been the use of radiolabelled peptides for targeting somatostatin expressing tumours. Both 90Y and 177  Lu labelled peptides have produced promising results [10]. However, clinical trials have  demonstrated large interpatient differences in tumour and organ uptake, indicating that more success could be achieved if patients were treated using individualized treatment plans [11]. Meanwhile, significant developments in EBRT were made as computer hardware and software advancements allowed for complex treatment planning and computer controlled radiation delivery. This technology led to the development of intensity modulated radiation therapy (IMRT), which uses an inverse planning approach to deliver nonuniform radiation beams incident on the patient from several beam positions [12]. The beam fluence maps are optimized using constraints placed on volume coverage and normal tissue sparing. IMRT provides the potential for better conformation of the radiation dose to the target and sparing of the surrounding tissues compared to conventional 3-dimesional (3D) conformal radiotherapy, and is an example of the detailed patient-specific treatment planning that is currently performed in EBRT. With the increased computer technology available in the 21st century, more accurate image reconstructions became possible in nuclear medicine imaging. These advanced image reconstruction techniques have provided the opportunity for more accurate dose calculations. Recent studies that have incorporated patient-specific organ masses and anatomy into the  6  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  dose calculation have had greater success at uncovering a dose-response relationship than previous attempts that based dose calculations on standardized reference phantoms [13-15]. These developments have led to the call for routine patient-specific dose calculations from several authors [16-19]. In summary, reasons for not performing routine dose calculations in radionuclide therapy can be identified based on the history of nuclear medicine over the last century. First, there is a long history of successful treatment with radioiodine, even though individualized dose assessments are not generally performed for these procedures. Second, investigators in the past have encountered difficulty identifying a clear dose-response relationship, both for predicting response in tumours and for predicting organ toxicities. Finally, accurate internal dose calculations are difficult to perform and given the perceived lack of a dose-response relationship and the long successful history of radioiodine treatment without patient-specific calculations, it has been difficult to justify the extra effort. However, as already mentioned, with recent advancements in nuclear medicine imaging and reconstruction techniques, it is now possible to make meaningful dose calculations. Dose-response relationships have been observed when internal dosimetry is performed using more sophisticated methods than were performed for early investigations into internally absorbed dose [13-15]. Furthermore, with the increased use of new therapeutic radiopharmaceuticals such as radiolabelled antibodies and radiopeptides, healthy organs such as the red marrow and kidneys can receive high doses. Thus, without careful treatment planning, many patients are under-dosed to avoid deleterious effects to these organs at risk. Not performing personalized dose calculations in radionuclide therapy is like treating every patient with the same beam fluence profiles in EBRT. Clearly, individualized treatment plans can increase the quality of patient care as they have done in EBRT for decades.  2.2 Physics of Nuclear Medicine Imaging and Therapy As described in Chapter 1, targeted radionuclide therapy (TRT) is performed using a pharmaceutical that is usually labelled with a charged particle emitter, which delivers a localized radiation dose. Determination of the dose distribution requires knowledge of the radiopharmaceutical spatiotemporal distribution, which is estimated by acquiring a series of nuclear medicine images.  7  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  The remainder of this chapter summarizes the fundamental principles that are needed to understand how nuclear medicine images are obtained and how therapeutic radiation doses are delivered to tumours using radiopharmaceuticals. This overview begins with a description of what radioactivity is. The radiation transport of photons and particles produced by radioactive decay is then described in a discussion of the interaction of radiation with matter. Next, the entire chain of events that is required to perform a nuclear medicine procedure starting from radioisotope and radiopharmaceutical production is outlined (Figure 2.1). The focus of this chapter then shifts to a description of the nuclear medicine hardware and the image reconstruction techniques used to estimate the 3D distribution of radioactivity in a patient.  Figure 2.1 General overview of nuclear medicine procedures from the production of radioisotopes to the acquisition of nuclear medicine images and delivery of radionuclide therapy.  8  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  2.3 Radioactive Decay The atomic nucleus is composed of protons and neutrons, which together are called nucleons. The notation used to represent a specific nuclide is  Using this notation, Z is  the atomic number of the atom, which indicates the number of protons. The number of nucleons in the nucleus is called the mass number and is equal to A, and the number of neutrons is denoted by N. The chemical element is given by the symbol X. Since the chemical element corresponds to the atomic number and the number of neutrons can be determined from the difference of A and Z, the notation for representing the nuclide can be simplified to  Isotopes of an element correspond to nuclides with the same number of protons, but different mass numbers, corresponding to nuclei with different numbers of neutrons. Out of all the possible combinations of protons and neutrons in a nucleus, only relatively few of these combinations result in a stable nucleus. In fact, out of more than 3000 known nuclides, less than 300 of them are actually stable [20]. The ratio of the number of neutrons to the number of protons is one factor that affects nuclear stability. For light elements, stable nuclides have roughly equal numbers of neutrons and protons. As mass number increases, the ratio of the number of neutrons to the number of protons increases in stable nuclides. Nuclides with an excess of protons or neutrons are unstable nuclei, called radionuclides, and undergo radioactive decay to nuclei with more stable configuration. 2.3.1  Nuclear decay mechanisms  Energy released by radionuclides in the process of radioactive decay results in the emission of photons and particles. There are several modes through which unstable nuclei decay. These different decay modes include:   β- decay,    positron (β+) decay,    electron capture,    nuclear fission,    alpha (α) decay, and    isomeric transition.  9  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  The particular decay mode that a nucleus undergoes depends on the relative number of protons and neutrons, on the size of the nucleus, the energy difference between the parent and daughter nuclei and on the spin and parity of states. Nuclei that have an excess of neutrons emit a β- particle in the process of converting a neutron to a proton. The resulting daughter nucleus has a more favourable ratio of neutrons to protons than the parent nuclide, and is thus more stable. Excess energy released by this reaction is randomly shared between the beta particle and a simultaneously released antineutrino. As a result, the kinetic energy of the emitted beta particle is a continuous spectrum that ranges from zero to the maximum energy released in this decay. Nuclei that are proton-rich can decay through one of two processes. In the first of these two mechanisms a proton is converted to a neutron resulting in the emission of a β+ emitted from the nucleus. Similar to β- emission, the β+ particle randomly shares the excess energy with a neutrino emitted in the process and so takes on a continuous spectrum of energies. A competing process with β+ emission is electron capture, which is the combining of a proton with an atomic electron to form a neutron. Additional decay mechanisms may be involved in the decay of heavy elements. Besides different types of beta decay, unstable heavy elements may become more stable either through the emission of alpha particles (  nuclei) or through nuclear fission, which is the  fragmentation of a heavy nucleus into two lighter nuclei. Isomeric transition is an additional mode through which a nucleus can decay and is an extremely important decay mechanism for diagnostic SPECT imaging. If a daughter nucleus created in a decay that involves particle emissions is not in the ground state and is relatively long lived, it is said to be in a metastable or isomeric state. Metastable states decay through the emission of gamma rays and conversion electrons, and this decay is termed isomeric transition. Internal conversion is the transfer of energy to an orbital electron, which is ejected as a conversion electron instead of the emission of a gamma ray. 2.3.2  Atomic emissions  As discussed in the previous section, radioactive decay may leave the atom in an excited state. For example, electron capture and internal conversion both remove electrons from inner shells with greater probability than from the outer shells. This leaves a vacancy in the inner shell, which is filled by an outer shell electron. Energy is released in the process, equal 10  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  to the difference in binding energy between the two orbitals. This energy is carried away as a photon, called a characteristic x-ray. Alternatively, this energy may be transferred to another orbital electron, which is emitted as an Auger electron. All photons and particles produced by nuclear decay and atomic emissions must be considered in internal dose calculations. 2.3.3  Decay equations  Decay of a radionuclide is a statistical event. It is not possible to predict exactly when an individual nucleus is going to decay, however it is possible to describe radioactive decay in terms of probabilities and average decay rates. The average decay rate (  ) of a  radionuclide sample is proportional to the number of atoms ( ) in that sample: (2.1) where λ is the decay constant, representing the fraction of atoms that decay per unit time. The decay constant has a fixed value for each specific radionuclide. The decay rate is usually called the activity  of the sample. Equation (2.1) can be integrated to obtain: (2.2)  revealing the exponential nature of radioactive decay. Multiplying both sides of Eq. (2.2) by  λ gives the activity: (2.3) The SI unit of activity is the becquerel (Bq), where 1 Bq corresponds to 1 decay/second. Another important quantity is the half-life  of a radionuclide. The half-life is the time it  takes for 50% of a sample to decay. It is related to the decay constant by: (2.4)  2.4 Interaction of Radiation with Matter A description of the physics governing the interactions of photons and particles with matter is important for understanding how the energy released during radioactive decay is deposited in tissue to deliver a radiation dose. Knowledge of photon interactions with matter is also required to understand the process of detecting these photons outside the body to image the spatial distribution of radioactivity using a gamma camera. 11  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Particles and photons produced by radioactive decay are both capable of exciting and ionizing atoms and are thus called ionizing radiations. It is through these excitations and ionizations that radiation transfers energy to an absorber, hence depositing a radiation dose. The mechanisms that charged particles and photons actually interact with and ionize atoms through are very different and so each will be discussed independently. 2.4.1  Charged particle interactions  Charged particles such as electrons and alphas are called non-penetrating radiation because they do not travel very far before giving up all of their energy and coming to a stop. Alpha particles emitted by radioactive decay typically have energies between 5 and 8 MeV, and only have a range in tissue on the order of a few cell diameters [21]. High energy β emissions (~2 MeV) used in radionuclide therapy have a maximum range of about 1 cm in tissue. There are two types of interactions of charged particles with matter. First, as charged particles traverse through a medium, they undergo collisions with orbital electrons. These collisions are actually electrostatic interactions between the charged particle and orbital electron and don't involve physical contact. In the process, the charged particle loses some of its energy. If this energy is greater than the binding energy of the orbital electron, then this electron can be ejected from the atom with a kinetic energy equal to the energy lost by the incident charged particle in the collision minus the binding energy. The ejected electron may even have enough energy to create secondary ionizations, in which case the secondary electrons are called delta rays. The second type of charged particle interaction involves an encounter with the electric field of the nucleus. This interaction causes a change in direction and deceleration of the charged particle. The loss of energy associated with this deceleration results in the emission of electromagnetic radiation called bremsstrahlung ("braking radiation"). As charged particles pass through a material, they constantly experience the electric field from surrounding atomic electrons and nuclei. Thus, they continuously lose energy through collisional and radiative losses (bremsstrahlung) as they pass through matter. The energy a charged particle transfers along its track is called the linear energy transfer (LET). The LET is typically measured in units of keV/μm and its value for charged particles depends on the mass, charge and energy of the particle. When comparing an electron and alpha particle of 12  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  equal kinetic energy, the electron travels at a much higher velocity because of its lower mass. As a result, the electron spends a shorter time in the neighbourhood of any particular atom and is less likely than the alpha particle to interact. Furthermore, the greater charge on the alpha particle results in stronger electrical interactions with orbital electrons. In summary, as the mass and charge of a particle increase, the frequency of interactions increases. Conversely, as the kinetic energy of the particle increases, the density of ionizations decreases. 2.4.2  Photon interactions with matter  Unlike charged particles, photons do not continuously lose energy as they pass through matter. Thus, electromagnetic radiation can pass through much greater distances than charged particles before losing its energy and is called penetrating radiation. A single photon interaction may result in that photon losing a fraction of its energy or it can even result in total energy loss. When a photon does interact it is said to be attenuated. For a number of photons attenuated photons  incident on a material with thickness  , is found to be proportional to  , the number of  : (2.5)  where μ is the linear attenuation coefficient, measured in units of inverse length. Similarly, in terms of photon beam intensity , the differential Eq. (2.5) becomes: (2.6) which in an analogous fashion to the description of radioactive decay, can be solved to reveal the exponential nature of photon attenuation: (2.7) The value of the attenuation coefficient depends on the photon energy as well as the type of the absorbing material. Furthermore, the value of μ represents the total attenuation coefficient, which is the sum of individual coefficients corresponding to different interaction types. It can be useful to factor out density effects to obtain the mass attenuation coefficient . The possible interactions in the range of photon energies produced by radioactive decay, from a few keV to greater than 1 MeV, are classical scattering (Rayleigh), photoelectric effect, Compton scattering and pair production. The mass attenuation coefficients for each of  13  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  these interaction types in different materials are plotted as a function of incident photon energy in Figure 2.2.  Figure 2.2 Mass attenuation coefficients for classical scattering, photoelectric effect, Compton scattering and pair production in H2O, Pb and NaI(Tl). (Data used for generating these plots was obtained at: http://www.nist.gov/pml/data/xcom/index.cfm).  Classical scattering involves the interaction of a photon with an atom as a whole. The interaction results in a scattered photon that has the same energy as the incident photon. For photons with an energy greater than 100 keV, the chance of classical scattering in tissue drops to less than 3% and therefore has little significance in nuclear medicine. Pair production is an interaction that occurs between a photon and the nucleus of an atom. In the process, the photon loses all of its energy and an electron-positron pair is created. The total kinetic energy of the electron and positron is equal to the energy of the incident photon minus the rest mass of the electron-positron pair (1.02 MeV). Thus, pair production is only 14  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  possible when the incident photon energy is at least 1.02 MeV, which again is not important for nuclear medicine applications. The most relevant photon energies in nuclear medicine are between 70 and 511 keV. The two dominating interactions at these energies are the photoelectric effect and Compton scattering. Photoelectric effect In the photoelectric effect, an incident photon transfers all of its energy to a bound electron. The electron is subsequently ejected from the atom and is called a photoelectron (Figure 2.3).  Figure 2.3 Photoelectric effect.  The electron involved in this process must be bound so that momentum can be conserved through recoil of the atom. Furthermore, the incident energy of the photon than the binding energy of the electron the photoelectron  must be greater  for this interaction to occur. The kinetic energy of  is: (2.8)  The likelihood of a photon interacting by the photoelectric effect is greatest when  is  equal to or slightly higher than the electron binding energy. This gives rise to the discontinuities visible in the plot of  in Figure 2.2, which correspond to binding energies  of the absorbing materials ( is the individual attenuation coefficient for the photoelectric effect).  15  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  The probability of photoelectric effect depends on  and is inversely proportional to  ,  such that: (2.9) Compton scattering Compton scattering is an interaction that occurs between an incident photon and an outer shell electron. This electron is considered to be free or unbound as long as the incident energy and momentum of the photon are much greater than the binding energy and momentum of the electron. In the Compton effect, the photon scatters at some angle θ and in the process loses some of its energy to the electron, which is ejected from the atom (Figure 2.4).  Figure 2.4 Compton effect.  The final energy  of the photon can be derived from conservation of energy and  momentum considerations: (2.10) where 0.511 MeV is the rest mass of the free electron. Unlike the photoelectric effect, the probability of Compton scattering is independent of Z. This is due to the fact that Compton effect occurs with a free electron. Thus, the chance of Compton scattering only depends on the number of electrons per gram, which is relatively  16  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  constant between elements. This can be observed in Figure 2.2 where the dependence of Compton mass attenuation coefficient on energy is similar in water, lead and NaI(Tl). The probability of photons scattering at different angles can be determined using the Klein-Nishina formula, which calculates the differential cross section for Compton scattering as a function of scattering angle and incident photon energy (Figure 2.5).  Figure 2.5 Polar plot of the Klein-Nishina distribution of scattering angle cross sections over a range of energies from 10 keV to 10 MeV, including selected energies of particular relevance to nuclear medicine. This distribution demonstrates how low energy photons (~10 keV) have nearly equal probability of being forward or backscattered at 0º and 180º, respectively. At 140 keV, photons are more likely scattered in the forward direction but may still backscatter. At even higher energies (> 1 MeV) photons are almost always forward scattered with little deviation from the incident photon direction.  2.5 Production of Radionuclides Since most naturally occurring radionuclides are very long-lived, radionuclides used in nuclear medicine are artificially produced in a nuclear reactor or a cyclotron [22]. In addition, some radionuclides can be obtained from radionuclide generators that contain long-lived parent radionuclides, which decay to short-lived daughters.  17  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Nuclear reactor production of radionuclides involves the fission of heavy elements into two lighter fragments. These fission fragments are neutron-rich since the neutron to proton ratio is higher for heavier elements than it is for lighter ones. As a result, nuclear reactor produced radionuclides tend to be β- emitters. In addition, radioisotopes can be produced in the reactor using large neutron fluxes created during fission to initiate reactions in a target material placed inside the reactor. In cyclotron production of radionuclides, a target is bombarded with charged particles such as protons, deuterons, and alpha particles. These charged particles have been accelerated to very high energies allowing for nuclear reactions to take place when they strike the target. Since cyclotron production typically involves addition of positive charges to the target nucleus, the radionuclide product is usually proton-rich and thus decays by β+ emission or electron capture. After being produced using one of these methods, the radionuclide is chemically attached to a compound to produce the radiopharmaceutical that is injected into the patient.  2.6 Radionuclides Used in SPECT Imaging and Therapy The selection of a radionuclide for a particular imaging or therapy application must take into consideration each of the features of radioactive decay, interaction of radiation with matter and production methods described thus far [23]. The application will dictate the appropriate emission type and half-life that is required for the specific clinical situation. For example, a radioisotope to be used for diagnostic imaging should have a physical decay constant that is similar to the rate of uptake and washout of the radiopharmaceutical. If the half-life is too short, then most of the radionuclide will have decayed before reaching the target. If the half-life is too long, then the radionuclide may remain in the patient for longer than necessary, delivering a radiation dose long after the image is acquired. In addition, a radionuclide to be used for SPECT imaging is ideally a pure gamma emitter. This is because any radiation emitted besides gammas will contribute to the radiation burden on the patient without any diagnostic benefit. On the other hand, radionuclides used for therapy are typically β- emitters, but can also be alpha or Auger electron emitters. Each of these emissions results in the release of charged particles, which leads to locally deposited radiation doses. As described above, the LET of  18  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  these different particles varies as a function of their mass, charge and energy. Thus, the choice of radioisotope for therapy has a big impact on the density of ionizations that will be produced. For example, the LET of β- particles emitted by the commonly used radionuclide 90  Y is about 0.2 keV/μm, whereas, the LET associated with alpha particles is approximately  100 keV/μm [24]. An additional important consideration is the production method for a particular radioisotope, since this will influence the cost and availability of the radionuclide. Furthermore, the actual element that is being used has an impact on the actual pharmaceuticals that the radionuclide can be combined with. Table 2.1 lists some commonly used radionuclides in imaging and therapy. As mentioned, a notable therapeutic radionuclide listed in this table is  90  Y, which emits β-  particles with a maximum energy of 2.28 MeV and a maximum range in tissue of 11.3 mm. No gammas are emitted by  90  Y making it a challenge to image the 3D distribution of this 177  radionuclide. On the other hand,  Lu is an example of a radionuclide used for therapy that  features both β- and gamma emissions. Lutetium-177 is a low energy beta emitter compared to  90  Y, with a maximum beta energy of 0.497 MeV and a maximum range in tissue of 1.8  mm. The differences in energies and associated ranges of the β- particles emitted by these two radionuclides gives each of them particular advantages in different applications. The relatively long range of  90  Y betas is useful for smoothing out the dose in situations of  heterogeneous uptake of radioactivity. The short range of treating micrometastases. Furthermore,  177  177  Lu betas is well suited for  Lu has been found to result in less frequent renal  impairment than 90Y labeled agents [25]. For diagnostic imaging,  99m  Tc is the most important radionuclide currently used in  SPECT imaging. Favourable characteristics of  99m  Tc include the fact that it decays by  isomeric transition (through the emission of a 140 keV gamma ray), which approximates the ideal of using a pure gamma emitter. In addition, the energy of 140 keV is high enough to readily escape the body, but not too high to escape detection by the nuclear medicine camera.  19  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Table 2.1 Common radionuclides for SPECT imaging and therapy  Nuclide  Decay mode  Half-life  Main photon  Photon  Beta Emax  Max range  energies (keV)  abundance (%)  (MeV)  (mm)  Application  32  β-  14.26 d  —  —  1.71  8.2  Therapy  67  EC  3.26 d  93, 185, 300  39, 21, 17  —  —  Imaging  90  P Ga  β-  64.1 h  —  —  2.28  11.3  Therapy  99m  Tc  IT  6.01 h  140  89  —  —  Imaging  111  In  EC  2.80 d  171, 245  90, 94  —  —  Imaging/therapy  123  I  EC  13.27 h  159  83  —  —  Imaging  131  Y  I  β-  8.02 d  364  82  0.606  2.3  Imaging/therapy  177  Lu  β-  6.73 d  113, 208  6, 11  0.497  1.8  Therapy  188  β-  17.0 h  155  15  2.12  10.4  Therapy  201  EC  3.04 d  69, 71, 80 (x-rays)  27, 46, 20  —  —  Imaging  Re Tl  EC, electron capture principal IT, isomeric transition  20  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  2.7 Components of the Gamma Camera After injection into the patient, the distribution of the radiopharmaceutical in the body is imaged by detecting gamma rays using a gamma camera [26,27]. The types of images that a gamma camera used for single photon imaging is designed to acquire are planar images, whole body (WB) planar images, dynamic images and tomographic SPECT images. Only gamma camera acquisitions that are relevant to this thesis will be discussed. Planar images and WB planar images are 2-dimensional (2D) projections of the 3D radioactivity distribution, often acquired from the anterior and posterior view, while SPECT images are 3D reconstructions of the spatial distribution of activity in a volume. Since a single projection does not contain any depth information, the raw data required for SPECT imaging are equivalent to a set of 2D planar images acquired at multiple angles surrounding the patient. The major components of the gamma camera that make acquisition of this data possible are the collimator, radiation detection system, electronics for positioning and energy analysis, and a computer for data storage, processing and image display. 2.7.1  Collimator  Gamma rays are too energetic to be refracted using a lens the way visible light photons are in a photographic camera. Thus, another mechanism is required to determine the point of origin of high energy gammas in order to produce an image. In single photon imaging this is accomplished using a collimator [28]. A collimator is designed so that only gamma rays traveling in certain directions can actually pass through it and reach the radiation detector. All other photons incident on the collimator should be absorbed. This process of absorptive collimation requires the collimator to be made of a heavy material with a high attenuation coefficient, such as lead. The channels that define the direction that gammas are allowed to pass through are called holes and the walls between these holes are called septa. The most common collimator design is the parallel hole collimator, which ideally only accepts gammas that are moving perpendicular to the face of the detector. In reality, the finite size of the holes allows photons traveling in directions at an angle to the normal of the detector surface to pass through the collimator (Figure 2.6). This acceptance angle degrades the spatial resolution of the system. The resolution can be improved by lengthening the holes 21  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  and narrowing their diameter, however, this reduces the count sensitivity of the system. This trade-off between sensitivity and spatial resolution means that different collimators are needed for specific applications.  Figure 2.6 A comparison of a high sensitivity and a high resolution collimator. The high sensitivity collimator on the left has a large acceptance angle resulting in relatively poor spatial resolution. The collimator on the right has longer and narrower holes, providing an improved spatial resolution at the cost of sensitivity. The dashed red lines represent photons absorbed by the collimator.  If a high number of counts is more important than good spatial resolution, then a low energy all purpose (LEAP) or high sensitivity collimator can be used. On the other hand, if the clinical situation requires better spatial resolution, then a high resolution collimator, such as a low energy high resolution (LEHR) collimator, with longer and/or narrower holes can be employed. An additional consideration is that not all radiation is perfectly absorbed within the septa. Septal penetration is particularly a concern for high energy radionuclides in which case medium energy and high energy collimators designed with thicker septa are required. 2.7.2  Radiation detection system  Gamma rays that pass through the collimator reach the radiation detection system of the gamma camera, which generally consists of a scintillation crystal coupled to an array of  22  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  photomultiplier tubes (PMTs). A scintillator is a material that emits light in the ultraviolet/visible range after absorbing high energy radiation. There are several characteristics that the ideal scintillator used for radiation detection should have [29]:   The scintillator should have relatively high scintillation efficiency, which is the fraction of incident photon energy that gets converted into detectable light.    This conversion should be linear so that the amount of light produced is proportional to the deposited energy.    It should be practical to produce detectors in sizes that are useful for clinical applications.    The scintillator should be transparent to its own scintillation light.    Decay time of the scintillation process should be short so that high count rates are possible.    The scintillating material should have a high atomic number and density to increase the probability of total gamma ray absorption by the photoelectric effect.  For SPECT imaging, the scintillator of choice that best meets these criteria is thalliumactivated sodium iodide [NaI(Tl)]. Large NaI(Tl) crystals with rectangular dimensions up to 40 cm x 60 cm are available, with a typical thickness of 0.95 cm. At this thickness, the intrinsic detection efficiency, which is a measure of how efficiently the detector absorbs incident radiation, is about 90% for the 140 keV gammas emitted by 99mTc. Furthermore, the scintillation efficiency of NaI(Tl) is 11.3% [30]. Thus, total absorption of a 140 keV gamma yields approximately 5000 scintillation photons in the energy range of 3-4 eV. The purpose of the PMTs that are coupled to the backside of the scintillation crystal is to amplify the relatively weak light output from the scintillator and convert it into an electrical signal. The PMT accomplishes this through the following components:   a photocathode at the beginning of the PMT, which converts photons into photoelectrons,    a series of dynodes that amplify the electric signal in the PMT,    an anode where the electric signal is collected, and    a high voltage power supply to maintain each consecutive dynode at a higher potential than the previous dynode.  23  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  The quantum efficiency of the photocathode, which is a measure of the number of photoelectrons emitted per number of incident photons, is typically in the range of 10-30% [31]. The potential difference between successive dynodes accelerates the electrons from one dynode to the next and at each dynode the number of electrons is multiplied by a factor of 3 to 6. Typically there are around 10 dynode stages, which results in a total electron multiplication factor in the range of 104-107. An illustration of the radiation detection process is depicted in Figure 2.7. A standard gamma camera will have between 30 and 100 PMTs coupled to the scintillation crystal, arranged in a hexagonal array.  Figure 2.7 Summary of events leading to an electric signal produced at the anodes of a photomultiplier tube (PMT) array following the detection of a 140 keV gamma ray. Typical values of scintillation efficiency (of NaI(Tl)), photocathode quantum efficiency and PMT dynode multiplication factors are used.  2.7.3  Positioning and energy analysis  The electrical signal produced by each PMT passes through a series of components including a preamplifier, amplifier and pulse height analyzer. These electronics are used to shape and amplify the electrical pulse and ultimately to determine the deposited gamma ray energy and location of the scintillation event. The position of the scintillation event on the crystal is calculated based on the differences between signals from each PMT and is performed by a dedicated onboard computer. In general, there is an inverse relationship between the light received by each PMT and the 24  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  distance to the point of interaction in the crystal. Non-uniformities in the detector system output are compensated for using the data acquired in a calibration scan to form precalculated lookup tables. The total energy deposited by the absorbed gamma ray is computed by adding up the signals from all PMTs. A multichannel analyzer is used to measure the energy spectrum of the detected photons. Using the multichannel analyzer, an energy window is set on the photopeak(s) of the radioisotope. To deal with the limited energy resolution of the scintillation detector, the upper and lower level discriminators of the energy window are typically set to ± 10% from the primary gamma energy [32]. Although separation is not perfect, this energy window helps to distinguish between primary and scattered photons. Photons detected with energies below the energy window have scattered and can no longer be used to provide information about the site of radioactive decay where they originated from. Scattered photons with an energy that falls within the energy window will still be counted. 2.7.4  Data storage and image display  Events within the selected energy window are accepted and added to a 2D histogram of counts versus position, which is stored in computer memory. Each element in this 2D histogram corresponds to a location on the scintillation crystal determined using the positioning logic. A projection image is obtained after a set period of time or preset number of counts in the 2D histogram is reached. This projection can be viewed on a computer monitor, stored, or processed in combination with an entire set of projections to reconstruct a SPECT image.  2.8 SPECT/CT Image Acquisition 2.8.1  Hardware  The SPECT camera contains all of the components listed in the previous section. The camera must be engineered so that the radiation detection system is part of a rotating gantry to allow for acquisition of projections at multiple angles around the patient. Furthermore, to increase detection efficiency, it is common for a SPECT camera to have multiple heads so  25  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  that projections can be acquired from two or more projection angles simultaneously. An image of a typical modern SPECT system is provided in Figure 2.8.  Figure 2.8 An example dual headed gamma camera with a computed tomography (CT) component from GE Healthcare (Infinia Hawkeye 4). The two detector heads are on a rotating gantry for single photon emission computed  tomography  (SPECT)  image  acquisition.  (Adapted  from  image  obtained  at:  http://www3.gehealthcare.ca/en-CA/Products/Categories/Nuclear_Medicine/SPECTCT_Cameras/Infinia_Hawkeye_4).  In addition, many modern SPECT cameras are hybrid systems that include a computed tomography (CT) component. The CT included with SPECT/CT systems is often a low dose (non-diagnostic) CT scanner, which is useful for lesion localization and attenuation map generation for attenuation correction. The reduction in dose leads to poor image quality compared to images produced using standard diagnostic CT scanners. Using a hybrid SPECT/CT system, the SPECT and CT scans are acquired sequentially while the patient remains on the same imaging table.  26  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  2.8.2  Data acquisition parameters  There are several parameters that can be adjusted before acquiring the projections needed for SPECT image reconstruction. These parameters include selection of the:   matrix size,    number of projections,    angular range of detector(s) rotation,    frame duration of each projection, and    orbit type.  The selected matrix size determines the number of discrete elements in the projection matrix. Matrix dimensions are usually a power of 2 and can range anywhere from 32x32 to 1024x1024. For patient SPECT studies, typical matrix sizes used are 64x64 or 128x128. Higher matrix sizes are used for planar imaging and acquisitions using 256x1024 matrices are common for WB scans. The number of projections is another important parameter since angular undersampling can lead to image blurring and artifacts. As the number of projection angles is increased, sharper images can be reconstructed. Most SPECT acquisitions use between 60 and 128 projections. These projections can be obtained over an angular range of 180 degrees or 360 degrees. A 360 degree arc is less prone to image artifacts caused by attenuation and spatial resolution effects varying with distance from the camera. The frame duration of each camera stop determines the number of counts and ultimately the noise in the projection. Finally, the orbit shape can be chosen to follow a circular, elliptical or noncircular orbit. The noncircular orbit follows the patient contours using infrared sensors. The advantage of this orbit type is that it minimizes depth dependent spatial resolution loss by acquiring each projection from as close to the patient as possible.  2.9 Image Reconstruction After the dataset of projection images is acquired, tomographic image reconstruction can be performed to create an estimate of the 3D distribution of activity. In this section, details of the image formation process are outlined and the most common algorithms used to reconstruct this data into a 3D image are described.  27  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  2.9.1  Image formation process  The situation that confronts SPECT image reconstruction can be demonstrated with a 2D object containing an activity distribution  . A set of 1-dimensional projections can be  measured around this object using a rotating gamma camera (Figure 2.9). This situation is a simplified example, which will explain how the activity distribution in a single 2D slice can be reconstructed. The function  is introduced to represent the counts detected at point  s on the detector at projection angle θ. For now, it is assumed that only gammas traveling perpendicular to the collimator surface are detected and effects of attenuation and scatter are ignored. Under these assumptions, the value of each element in  is proportional to the  total activity found along the lines of response extending from each position on the detector and running through the whole thickness of the object.  Figure 2.9 Example 1-dimensional projection profiles obtained by a gamma camera rotating around a 2dimensional activity distribution. The value of  at each point  is proportional to the total activity along  the line of response (LOR) passing through each collimator hole (ignoring image degrading effects such as attenuation, scatter and collimator detector response, which are discussed in Section 2.10).  The line integral operation that turns  is called the Radon transform.  This projection operation can be summarized by the equation: (2.11)  28  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  where C is called the system matrix or the forward projection operator. Essentially, the imaging problem is to find the activity distribution f, using the projection measurements g. 2.9.2  Reconstruction algorithms  Backprojection techniques The most straightforward approach for reconstructing an image estimate from the projection dataset is to use simple backprojection [33]. The basic idea of backprojection is to smear the counts obtained in  back into image space. In simple backprojection, since  there is no knowledge of the original activity distribution, these counts are spread evenly into all voxels falling along the lines of response extending from each position on the detector. Clearly, this puts counts in voxels where they don't belong and the result is a very blurry image. Filtered backprojection (FBP), which uses a ramp filter to correct for this blurring, can be employed. However, even though FBP is still sometimes used in clinical studies, it cannot be used to reconstruct quantitative images. This is because FBP does not compensate for image degrading factors such as photon scatter and depth dependent spatial resolution. An approximate attenuation correction with the Chang method can be applied [34], but this approach is limited by the assumption that the linear attenuation coefficient within the object is uniform. Furthermore, use of the ramp filter, which enhances high frequency components corresponding to fine details in the data, also seriously enhances noise in the data. The noise can be reduced using a smoothing filter, but this degrades the spatial resolution. For dosimetry calculations, advanced algorithms that more accurately model image degrading effects must be employed. Iterative reconstruction techniques It is possible to reconstruct quantitative images using iterative techniques, which allow for the modelling of physical effects that degrade image quality [35]. Iterative algorithms use successive estimates of the activity distribution to find a solution to Eq. (2.11). The general structure of iterative algorithms is outlined in Figure 2.10. First a starting estimate of the activity distribution  is created. This starting estimate can be very simple, such as a  uniform image, or it could be a FBP reconstructed image (modified to exclude pixels with zero or negative values). In any case, the current estimate  is forward projected to obtain 29  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  a set of projection estimates. These projection estimates are compared to the actual measured projections to form an error estimate. Next, the error estimate is backprojected into image space and used to update estimate  and the process begins again starting from the updated image  . Iterative techniques usually produce superior images to FBP, but this comes  at the cost of computation time.  Figure 2.10 General structure of iterative reconstruction methods.  The most commonly used iterative algorithm is maximum likelihood expectation maximization (MLEM) [36]. The MLEM algorithm is derived from Poisson statistics and works to find the activity distribution that is most likely to produce the set of measured projections. The MLEM equation is: (2.12) which fits the general structure of iterative reconstruction algorithms as follows:   The current estimate projections  is forward projected to obtain a set of estimated , where M is the total number of image voxels. The value  is the system matrix element that represents the probability of a photon emitted from voxel k contributing to a count in projection pixel . Physical effects such as attenuation and collimator blurring can be modeled and included in the system matrix.  30  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy    An error estimate is formed by taking the ratio of the measured counts over the estimated counts in projection pixel :    . , where N is  The backprojection of this ratio is given by the total number of projection pixels.    Finally, the backprojected error estimate is normalized by by  to obtain the updated image estimate  and multiplied  .  Since the MLEM algorithm accounts for the Poisson statistical nature of the emission data, images reconstructed using this approach have improved noise characteristics compared to those reconstructed with FBP. An additional problem with FBP images is that they may contain unphysical negative values. With MLEM, multiplying the error and image estimate keeps the image values non-negative, as long as the initial estimate is non-negative. A disadvantage of MLEM is that it is slow to converge and may take up to 50-200 iterations before reaching a usable solution. Furthermore, as the number of iterations increases, the reconstructed image noise increases. This noise can be reduced using a postreconstruction lowpass filter. The slow convergence of MLEM can be accelerated by using the ordered subsets expectation maximization (OSEM) algorithm [37]. The modification that OSEM makes to the MLEM algorithm to speed up convergence is to break up the projection data into subsets. For example, if 60 projections were acquired, then these could be divided into 6 subsets with 10 projections per subset. The MLEM algorithm is then used to update the image one subset at a time. One iteration is completed after all of the subsets have been used. In the given example of 60 projections, the image estimate will have been updated 6 times within one iteration. A single subset containing all of the projections is equivalent to standard MLEM, which uses all of the projection data to perform one update. In general, OSEM speeds up MLEM by a factor that is approximately equal to the number of subsets used. The OSEM equation is: (2.13) where the image update  is obtained from the current estimate  through all of the projections in subset  after passing  . 31  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  2.10 Image Degrading Factors Various image degrading physical factors have been alluded to in the previous sections. The impact of some of these effects, including resolution loss, attenuation, scatter and noise is illustrated in Figure 2.11. Here the true image is defined as the projection image that is unaffected by any of the physical factors that will be discussed.  Figure 2.11 Series of projection images of an activity source demonstrating different sources of image degradation. The projection of the source in air (a) differs from the true image due to the poor resolution of the system. When the source is placed in water (b) attenuation reduces the number of counts and scatter reduces image contrast. The impact of image noise due to the random nature of radioactive decay and radiation detection are demonstrated by the statistical fluctuations in the measured projection in (c).  2.10.1 Attenuation and scatter Attenuated photons are those that are absorbed or scattered and are subsequently undetected. This leads to a reduction in the number of counts relative to the ideal projection image. The ratio of transmitted to incident photons along path S through a heterogeneous material that has a distribution of linear attenuation coefficients  is given by the  transmitted fraction, TF: (2.14)  32  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Clearly, the degree of attenuation will depend on the thickness and type of material between the source and the detector (Figure 2.12). Scattered photons are photons that interact through classical or Compton scattering and are still detected. As discussed in Section 2.7.3, Compton scattered photons will still be counted if their energy falls within the selected energy window. This adds counts to the projections in the wrong locations since these photons have scattered at some angle and are not detected along the line of response corresponding to the site of the radioactive source. As a result, scattered photons reduce image contrast. 2.10.2 Collimator detector response An additional source of image degradation is the collimator detector response, which is the principal factor that determines SPECT image resolution. There are actually four components that make up the collimator detector response: collimator resolution, detector intrinsic resolution, septal penetration and septal scatter [38]. The collimator resolution is determined by the geometrical acceptance angle of the collimator holes, which results in spatial resolution that degrades with distance from the collimator (Figure 2.12). The intrinsic resolution refers to the resolution of the detection system without the collimator and is limited by the resolution of the crystal and by the accuracy of the positioning electronics. Septal penetration and scatter are the components of the collimator response corresponding to photons that pass through the collimator septa, further degrading the spatial resolution (Figure 2.12). The effects of septal penetration and scatter are especially important for medium and high energy gamma rays.  33  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Figure 2.12 Distance dependent spatial resolution is demonstrated by the count profiles of a point source, which broaden with distance from the detector. The full width half maximum (FWHM) of these profiles can be used to characterize the spatial resolution of the system. In addition, the dashed red lines correspond to gamma rays that have undergone septal penetration, which further degrades the spatial resolution.  As illustrated in Figure 2.12, the spatial resolution can be characterized by the full width half maximum (FWHM) of the point-spread function (PSF). The PSF is the profile of a point source projected onto the detector. 2.10.3 Noise Statistical fluctuations in the number of photons detected arise from the fact that radioactive decay and interactions of photons with matter are random processes. In general, the counting statistics of nuclear medicine studies are limited by the camera sensitivity, as well as the amount of injected activity and the length of image acquisition time that is reasonable for the patient. The resulting statistical noise gives images a mottled appearance. Since noise usually dominates at high frequencies, it can be reduced through the use of low-pass filters. Two of the more commonly used low-pass filters in nuclear medicine are Hanning and Butterworth [39]. The Hanning filter is just a cosine function:  34  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  (2.15) The filter is defined to be zero for spatial frequencies  above the cutoff frequency  . The  Butterworth filter features two parameters that can be selected, making it a more versatile filter than Hanning. The two parameters are the critical frequency  and the order , which  controls the steepness of the roll-off as the filter approaches zero. The Butterworth filter is given by: (2.16) The Hanning and Butterworth filters are plotted in Figure 2.13, showing the shape of these filters when different parameters are selected.  Figure 2.13 Examples of low-pass filters used to suppress image noise, which dominates at high frequencies. The shape of the Hanning filter (a) is controlled by one parameter: the cutoff frequency filter (b) is controlled by two parameters: the critical frequency  . The Butterworth  and the order .  Fine details in the image also correspond to the high frequency region of the power spectrum, which means that suppressing noise comes with the trade-off of degrading image resolution. Thus, when estimating the total activity in a region of interest it is not always desirable to apply a low-pass filter [40]. However, if the aim is to use a reconstructed image to calculate a 3D dose distribution, then it is recommended to apply a filter to reduce the noise. The influence of low-pass filters on a reconstructed SPECT image slice can be visualised in Figure 2.14 where both Hanning and Butterworth filters have been applied using some of the parameters used to plot the functions in Figure 2.13.  35  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Figure 2.14 Example SPECT image slices reconstructed by OSEM with no filter (a), and with Hanning (b) and Butterworth (c,d) postreconstruction filters applied.  2.10.4 Partial volume effect Another important effect that impacts image quantitation is partial volume effect (PVE) [41]. PVE is actually the consequence of two separate phenomena that affect the accuracy of reconstructed activity concentrations. The first component of PVE is related to the blurring caused by the limited spatial resolution of the imaging system. This blurring causes the spillout of activity from the source into the surrounding background and the spill-in of activity from the background to the source. This effect is most severe for small objects with a size that is less than three times the FWHM of the system. The second component of PVE is image sampling. Reconstructed images assume that each voxel contains uniformly distributed activity. However, the boundaries of an object do  36  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  not actually follow the shape of the voxels (Figure 2.15). Thus, each voxel contains an average intensity taken from each of the tissues actually included in that voxel.  Figure 2.15 Comparison of a true object on the left with an image of this object on the right, demonstrating the influence of image sampling. Pixel intensity at the object boundaries is an average value taken from the source and background contributions.  2.10.5 Camera dead time The last factor that will be considered here is camera dead time (τ), which has to do with the time it takes a counting system to process an individual detected event [42]. Counting systems can usually be described as being paralyzable or nonparalyzable. For paralyzable systems, every event that occurs during the time it takes the system to process individual events adds to the total resolving time without getting counted. In nonparalyzable systems, every event that occurs during the processing time is ignored, but does not add to the total resolving time. The observed count rate  as a function of the true count rate  in the  nonparalyzable case is: (2.17) which approaches the asymptote  . For paralyzable systems, the observed count rate is: (2.18)  which reaches a maximum value of  and then approaches zero as the true counts  increase to very high rates (Figure 2.16). A typical value of τ for a scintillation detection system with NaI(Tl) is about 5 μsec.  37  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Figure 2.16 Observed counting rates count rate  for paralyzable and nonparalyzable systems as functions of the true  .  Dead time is not a factor in diagnostic imaging where count rates are relatively low, but can become significant in therapeutic applications where the injected activities are high, especially for activities greater than about 4 GBq. Another issue at high count rates is the mis-positioning of events. If two photons are detected roughly simultaneously, the combined scintillation light from each event will affect the ability of the positioning electronics to accurately identify the positions of these events.  2.11 Quantitative Corrections in SPECT Imaging Internal dose calculations require quantitative SPECT images in order to produce accurate estimates of the radiation dose distribution in the patient. However, the image degrading factors described in the previous section introduce image artifacts and affect the quantitative accuracy. In this section, the methods used to correct for the most important effects are described. 2.11.1 Attenuation correction An attenuation map of the distribution of linear attenuation coefficients in the patient’s body is needed for accurate attenuation correction. Using modern dual modality SPECT/CT systems, the attenuation map can be obtained from the CT image [43]. This first requires transformation of the CT data, which are in Hounsfield units, into a map of linear attenuation coefficients that correspond to the gamma ray energy used to acquire the SPECT image. 38  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Once the attenuation map is obtained, attenuation is accounted for using iterative reconstruction by incorporating transmission factors calculated using Eq. (2.14) into the system matrix [44]. Specifically, each system matrix element  becomes the product of the  transmission factor and the fractional contribution of voxel k to pixel i. This demonstrates how physical effects are modeled using iterative reconstruction by forward projecting the current image estimate with the system matrix. 2.11.2 Scatter correction If attenuation correction is applied and scatter effects are not accounted for, then the reconstructed counts will be an overestimate. This is because the linear attenuation coefficients used to calculate the transmission factor assume that a photon is removed from the beam once it has been fully absorbed or scattered. These linear attenuation coefficients are called narrow beam attenuation coefficients. The problem is that in nuclear medicine there is a significant fraction of scattered photons that are still counted by the detector. The simplest way to account for scatter is to use a broad beam attenuation map when performing attenuation correction [45]. Broad beam attenuation coefficients are lower than narrow beam attenuation coefficients. This leads to an undercorrection for attenuation and balances out the inclusion of scattered photons in the projections. However, this method is only approximate and does not compensate for the loss of contrast caused by scattered photons. A second option for scatter compensation is to use an energy based method such as the dual energy window [46] or the triple energy window technique [47,48]. Using these approaches, counts obtained in additional energy windows placed around the photopeak, which should only contain scattered photons, are scaled and subtracted from the photopeak counts. The disadvantages of these energy based methods are that they tend to increase image noise and they wrongly assume that the spatial distributions of scattered photons of different energies are the same. Methods that employ scatter modelling are the most accurate scatter correction techniques, but they can also be very computationally expensive. In this category, Monte Carlo and analytical techniques can be used to model scatter starting from an estimate of the activity distribution and a patient-specific attenuation map. Monte Carlo based techniques model photon transport stochastically using the known physics of photon interactions. Using 39  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Monte Carlo, scatter is modeled in the forward projection step of iterative reconstruction to transport photons from each estimated image voxel to estimate the distribution of scattered photons in the projections [49]. Alternatively, noise-free scatter projections can be determined analytically using the Klein-Nishina formula [50,51]. 2.11.3 Collimator detector response compensation In general there are two methods of collimator detector response compensation. The first method involves restoration filtering. Degradation of the true image can be described by convolution of the object with the PSF. Restoration filters, such as the Metz or Wiener filters [52], aim to recover the true image by applying an inverse of the PSF of the system. However, these inverse filters amplify the noise and wrongly assume that system resolution is spatially invariant and are therefore limited in how much spatial resolution compensation they can achieve. The second method of collimator detector response compensation is to use an iterative technique. Here, the effects of distance dependent spatial resolution are modeled by iterative reconstruction by incorporating effects of the collimator detector response into the system matrix [53]. Essentially, the data is projected along lines that are not just perpendicular to the face of the detector. This blurring of the data is usually described by a distance dependent Gaussian function. The standard deviation of this Gaussian function  is the system  spatial resolution, which can be expressed as a function of distance intrinsic resolution of the camera  and the collimator resolution  from the camera, the : (2.19)  2.11.4 Partial volume effect correction It may be necessary to apply PVE correction in radionuclide therapy applications, particularly when calculating the dose to small tumours. Since PVE is due to limited spatial resolution, this effect can be reduced by including resolution recovery in the iterative reconstruction algorithm as described in Section 2.11.3. However, collimator detector response compensation does not completely eliminate PVEs, especially in the case of small objects.  40  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  Traditionally, one of the most popular strategies to correct for PVE is the use of recovery coefficients [54]. A recovery coefficient is defined as the ratio of activity concentration measured in the SPECT image to the true activity concentration. Physical phantoms involving spheres of different sizes filled with known activity concentrations can be used to obtain a set of recovery coefficients that can be applied to approximately spherical objects of similar sizes. However, this simple correction method will be inaccurate when applied to irregularly shaped tumours surrounded by nonuniform background activity. Recently, more sophisticated methods have been developed. For example, voxel based methods, such as the iterative template based technique, have been proposed [55]. Advanced PVE correction is still an active area of research and no widely accepted method exists. 2.11.5 Dead time correction A simple method for dead time correction involves the use of a reference source that is scanned with and without the patient present [56]. The counts from the reference source without the patient present divided by the counts from the scan performed with the patient provides a dead time correction factor. However, the issue with this method is that when the patient is present, counts from the reference source can get contaminated by counts from the patient. Alternatively, mathematical models can be used to correct for dead time losses. For nonparalyzable systems, Eq. (2.17) can be rearranged to solve for the true count rate analytically. For paralyzable systems, Eq. (2.18) cannot be solved analytically, but graphical or numeric methods can be used to approximate the true count rate [57]. Gamma cameras generally behave as paralyzable or combined paralyzable-nonparalyzable systems. The exact behaviour should be investigated using a calibration phantom experiment that mimics the patient study and that can be used to relate the observed count rate to the true count rate.  2.12 Absolute Quantification The goal of quantitative SPECT imaging is to be able to relate the reconstructed counts in a voxel to the absolute activity concentration in the corresponding volume of the object under investigation. In addition to applying the quantitative corrections described in Section 2.11, this absolute quantification requires a camera calibration factor. This calibration factor  41  Chapter 2: Fundamentals of Nuclear Medicine Imaging and Therapy  should be obtained by taking a measurement of a source of known activity (determined by a dose calibrator) with the gamma camera. One method for performing this measurement is to acquire a planar scan of the source in air and measure the count rate by summing all of the counts duration  [58]. The calibration factor  dividing the count rate by the known activity  acquired in a scan with  (in cpm/kBq) of the camera is determined by of the point source. (2.20)  Use of this method requires very accurate correction for scatter and attenuation in the SPECT reconstruction. A more robust calibration method is to perform a SPECT acquisition of a source geometry that better represents the patient [59]. Unlike the planar scan in air, the counts acquired using this method are affected by attenuation and scatter. The SPECT image should be reconstructed using the same parameters that are to be used for patient imaging. The reconstructed counts in the source volume can then be divided by the activity of the source to obtain the calibration factor. This SPECT-based calibration is superior to the in-air planar method because it partially compensates for inaccuracies in the attenuation and scatter correction. However, the SPECT-based method is not as practical to perform on a routine basis.  2.13 Summary The different requirements for producing quantitative 3D images of radioactivity distributions have been described. These quantitative images are needed for accurate calculation of 3D dose distributions. Details of internal dose calculations are explained in the next chapter.  42  Chapter 3: Internal Dose Calculations  Chapter 3: Internal Dose Calculations 3.1  General Concepts in Internal Dosimetry The radiation absorbed dose is defined as the amount of energy deposited per unit mass.  To estimate the dose, one must first specify the mass of the object of interest (  ), the  cumulated activity in that object, as well as the activity in the surrounding regions. The frequency n and energy E of each radiation type i emitted per nuclear decay of the administered radionuclide must also be known. Furthermore, a quantity referred to as the absorbed fraction φ is needed, which represents the fraction of energy emitted from the source  that is deposited in the target  . These factors can be combined to form a generic  equation for calculating the absorbed dose in  due to activity in  : (3.1)  where  is the time-integrated activity (total number of nuclear decays integrated over a  specified time period) of the source [60]. This general equation for calculating the absorbed dose can be simplified using the Medical Internal Radiation Dose (MIRD) system, in which the dose to the target summed over all source regions is given by: (3.2) where the S value  represents the mean absorbed dose deposited in the target per  unit of time-integrated activity that is present in the source. Equation (3.2) illustrates the two major ingredients of internal dose calculations, which are (i) the total number of decays in each source region given by  , and (ii) the radionuclide specific emission data and  corresponding absorbed fractions given by  . The time-integrated activity  is  often normalized by the administered activity to form the time-integrated activity coefficient (TIAC) so that the estimated dose is reported in units of gray per megabecquerel (Gy·MBq‑1). The TIAC was previously known as the residence time.  43  Chapter 3: Internal Dose Calculations  Dose calculations can be performed at a variety of spatial levels using source and target regions defined at the organ, sub-organ, voxel and cellular levels.   Organ level dose calculations are used to estimate dose to whole organs under the assumptions that activity is uniformly spread through source organs and dose is uniformly deposited in target organs.    Sub-organ dosimetry may be performed in organs where the radiopharmaceutical activity distributes non-uniformly. For example, a multiregion kidney model can be used to perform suborgan dosimetry for the kidneys [61].    Voxelized dose distributions may be estimated from nonuniform activity distributions in voxels of any dimension. Usually, the voxel dimensions correspond to the SPECT image voxel size, which is the minimum scale that can be used for quantifying activity in vivo.    Dosimetry at the cellular level can be investigated through the use of autoradiographic techniques [62] or using cell dosimetry models [63].  Traditionally, dose calculations are performed at the organ level, in which case total organ doses are calculated using S values that have been pre-calculated for reference phantoms representing average patients of a given age and gender. The procedure for performing an internal dose calculation, as summarized by Eq. (3.2) is completed in two distinct steps. First, the time-integrated activity  in each source region  must be determined. In the second step, the time-integrated activity data is combined with the physical data  to form the dose estimate.  3.2 Acquiring the Time-Integrated Activity Questions that need to be answered when attempting to determine  , include: (i) what  are the source regions of interest (ROIs), (ii) how quickly does the activity accumulate in these regions, (iii) how much activity accumulates, and (iv) how long does that activity stay in each source ROI? This information is often called the biodistribution of activity. To answer these questions about the biodistribution of radiopharmaceutical, a series of nuclear medicine scans (planar and/or SPECT) must first be acquired. Then, the planar and/or reconstructed SPECT images are segmented to obtain volume and activity estimates of source ROIs inside the patient. Finally, the acquired time-activity data can be plotted to 44  Chapter 3: Internal Dose Calculations  form time-activity curves (TACs) for each source region, which are integrated to find the corresponding values of 3.2.1  . In this section, each of these steps is described in detail.  Imaging protocol  The first step in estimating time-integrated activities is to acquire a series of nuclear medicine images. This allows for the identification of source regions and the determination of activity in ROIs at different points in time. Traditionally, this step is completed using conjugate view analysis in which time-activity data in source regions is estimated using a 2D imaging protocol [64-66]. An example set of planar images is given in Figure 3.1. There is no depth information available when a 2D imaging protocol is used, so no correction or simplified methods to correct for attenuation, scatter and organ overlap in 2D images are employed [67]. These approximate methods usually lead to inaccurate activity estimates.  Figure 3.1 Planar data with segmentation performed to determine the planar counts in regions of interest over a period of 22 hours after injection.  To increase the accuracy of time-integrated activity estimates for different areas in the body, hybrid planar/SPECT techniques have been proposed [68,69]. In this approach, the temporal changes in radiotracer concentration in each ROI are determined from a coregistered series of 2D whole body (WB) images acquired over several hours or even days after radiotracer injection. Additionally, a single SPECT or SPECT/CT scan is acquired over  45  Chapter 3: Internal Dose Calculations  the same ROIs to provide quantitation for this activity. The data acquired with any standard hybrid SPECT/CT camera and reconstructed with standard software (provided by the camera manufacturer) allows the user to compensate for attenuation, collimator blurring, and to a certain degree – scatter, creating images representing the 3D semi-quantitative distribution of the radiotracer in the body (Figure 3.2). However, to obtain truly quantitative images, accurate scatter modeling, partial volume effect correction and dead-time correction must be implemented [70,71].  Figure 3.2 Sample coronal slices from a SPECT image of the tumour in the abdomen of the patient displayed in Figure 3.1. The SPECT image provides information about the 3D activity distribution.  The final option is to perform a study following a purely SPECT based approach. This is the only imaging protocol that can fully characterize the temporal behaviour of the 3D distribution of activity. 3.2.2  Segmentation of nuclear medicine images  At the next stage, image segmentation is necessary to determine the counts and corresponding activities in ROIs at each imaging time point. When planar images are acquired, ROIs must be delineated in 2D. This is often done manually. Alternatively, a thresholding scheme may be applied, where regions that include the organ or tumour of interest as well as some surrounding background are drawn. Next, segmentation of the organ 46  Chapter 3: Internal Dose Calculations  or tumour is performed by choosing a threshold calculated as a percentage of the maximum counts inside the manually drawn region and all pixels with a value above this threshold are included in the final segmented region. Another option for defining 2D ROIs is to use the projection of volumes of interest that have been segmented in 3D using either a nuclear medicine or anatomical modality. When a hybrid planar/SPECT or purely SPECT approach is being used for organ level dose calculation, ROIs must be segmented in 3D to determine the total activity in these regions and the volume of these regions. Three-dimensional image segmentation is not strictly required when voxelized dosimetry is performed. However, it is useful to summarize 3D dose calculations using quantities such as the mean dose, maximum dose or the minimum dose to 90% of the volume (D90). In these cases, definition of the 3D ROI is necessary. In clinical and research studies, volumes of interest are commonly segmented using manual region delineation. Since drawing region boundaries on low resolution nuclear medicine images can be difficult, high resolution images obtained from an anatomical modality, such as CT or magnetic resonance imaging (MRI) may be used for this purpose. Alternatively, as described above for segmentation of ROIs in planar images, thresholding schemes are often applied to nuclear medicine images. Use of thresholds ranging from 25 to 70% have been reported, while fixed thresholds set at around 40% of the maximum counts are most commonly employed [72,73]. For 3D dose calculations, the dose distribution can be analyzed within the segmented region. For organ level dose calculations, the same segmented region is often used to determine both the object volume and the activity. This approach, however, has important drawbacks related to the fact that the “true-volume” regions delineated by these techniques often do not encompass the true activity, since due to PVE the reconstructed nuclear medicine image is blurred and activity spills out from the object into the surrounding tissues [41]. As a result, one and the same threshold cannot be used for accurate estimation of both activity and volume for any particular object [74]. To resolve this problem, one may try to apply a PVE correction hoping to restore the true total activity within the true total volume. As described in Section 2.11.4, although the use of recovery coefficients is one of the more popular strategies, it is not always suitable for dose calculations where both the size and shape of objects can vary considerably. Even the more  47  Chapter 3: Internal Dose Calculations  advanced methods are unable to model all clinical situations and correct for the full extent of spill-out (and also spill-in) effects. Furthermore, although some of the advanced methods perform better than recovery coefficients, they require substantial processing effort, which makes them impractical for routine clinical use. An additional problem associated with threshold-based segmentation of nuclear medicine images is that fixed thresholds cannot account for the variety of clinical situations where organs and tumours with different sizes and activities are surrounded by other tissues containing different levels of background activity [75-77]. As a solution, adaptive thresholding techniques have been proposed where a threshold is chosen that takes into account the source-to-background ratio (SBR) of activity concentrations, the size of the object to be segmented and the image acquisition and processing methods, all of which cannot be fully accounted for by a fixed threshold approach. A variety of adaptive thresholding methods have previously been proposed. Erdi et al. [78] employed a set of curves corresponding to different SBRs, relating the optimum threshold for segmentation to the object volume. A drawback of this method is that it requires an a priori estimate of the volume. As an alternative, Daisne et al. [77] proposed an approach requiring only a single threshold-SBR curve that can be applied to objects of all volumes. Iterative methods that remove the need for any prior knowledge of the object volume have also been suggested [79-81]. In order to establish the parameters of the adaptive threshold curves, each of these methods requires a calibration phantom experiment. An important shortcoming of adaptive thresholding is that the estimate of background activity needed to calculate the SBR is done by manually selecting background regions outside the source. This procedure is tedious and may lead to inconsistent estimates of the background activity depending on where and how these regions are drawn, especially in situations with inhomogeneous background. In this thesis, an iterative adaptive thresholding technique that improves on current adaptive thresholding techniques by semi-automatically selecting background regions and using separate thresholds for volume and activity estimation has been developed (Chapter 5). 3.2.3  Time-activity curves  Sections 3.2.1 and 3.2.2 have described the different imaging protocols and some of the segmentation methods that can be used to acquire the time-activity data. The next step 48  Chapter 3: Internal Dose Calculations  towards estimating the time-integrated activity in source regions is to plot the activity versus time for each region and fit a curve through this data. The area under this curve represents the total number of decays in the source (the time-integrated activity). One way to find the area under the time-activity data is the trapezoidal method. This method involves adding up the areas of trapezoids formed by each pair of data points (Figure 3.3). The trapezoidal method is simple to implement, but it provides limited information about the biokinetics of the patient and it can be problematic to estimate the time-integrated activity beyond the last data point.  Figure 3.3 The trapezoidal method used to find the time-integrated activity. In this example, the time-integrated activity is equal to the sum of four areas labelled in the figure from A1 to A4. Elimination of radioactivity after the last data point (at 22 hours) is assumed to be by physical decay only, which is a conservative estimate that will most likely over-estimate the time-integrated activity. For 99mTc, the area under the curve after the last data point (A4) is small due to the relatively short physical half-life of 99mTc. However, for longer lived radionuclides (eg. 177Lu, T1/2 = 6.73 days) the area under the curve after the last data point can be significant.  A more robust method for acquiring TACs is to perform multi-exponential fits through the time-activity data: (3.3) where  is the physical decay constant defined in Eq. (2.1) and  is the biological  elimination constant of the jth exponential. The biological elimination constant is analogous to the physical decay constant and describes the rate of radiopharmaceutical disappearance  49  Chapter 3: Internal Dose Calculations  by biological processes. Thus, in an analogous fashion to physical decay, the biological halflife can be defined: (3.4) The physical decay constant and biological elimination constant can be combined to form the effective elimination constant: λ  λ  (3.5)  which in turn gives the effective half-life: (3.6) The multi-exponential fit given in Eq. (3.3) is easily integrated to obtain the time-integrated activity: (3.7) Further details of the procedure for acquiring TACs for different source regions depends on the imaging protocol that was used. As described in 3.2.1, time-activity data has traditionally been acquired using a 2D imaging protocol and conjugate view analysis [67]. The conjugate view method estimates the source activity  using the expression: (3.8)  where ROI,  and  represent the count rates in the anterior and posterior views of the source  is the transmission factor across the patient thickness ,  for self-attenuation within the source, and  is a correction factor  is the camera calibration factor defined in  Eq.(2.20). The transmission factor can be determined from the ratio of  obtained from  counts measured in a transmission scan with ( ) and without ( ) the patient in place. Additional correction factors can be applied for overlapping source regions and subtraction of background activity. When using the hybrid planar/SPECT approach, the shape of the TAC is determined by fitting a curve through the planar data only. Then, the TAC is rescaled to pass through the  50  Chapter 3: Internal Dose Calculations  quantitative SPECT-based activity measurement. Figure 3.4 gives a graphical illustration of this procedure.  Figure 3.4 Use of the hybrid planar/SPECT technique to plot a time-activity curve (TAC). The shape of the TAC is first determined from the planar data (a) and then rescaled to pass through the quantitative SPECTbased activity measurement at the time of the SPECT acquisition (b).  For determination of organ level time-integrated activities using the hybrid planar/SPECT technique, the SPECT-based activity measurement that is used is the total activity in the organ. Alternatively, the hybrid planar/SPECT method can be used to estimate the 3D dose distribution at the voxel level, assuming that the spatial distribution of activity does not change over time and that each voxel follows the same kinetic behaviour given by the planar data. Rescaling the TAC through the planar data by each SPECT image voxel individually provides an estimate of the 3D distribution of time-integrated activities, which can then be used to form a voxelized dose estimate. A flowchart of the hybrid planar/SPECT technique is provided in Figure 3.5 and a detailed demonstration of the use of the hybrid planar/SPECT technique for patient studies is described in Chapter 8.  51  Chapter 3: Internal Dose Calculations  Figure 3.5 A flowchart depicting the general framework of the hybrid planar/SPECT approach.  When a purely SPECT-based imaging protocol is employed, this provides the option to define separate TACs for each voxel, allowing the user to account for nonuniform activity kinetics inside an organ or tumour. This requires the SPECT images to be accurately coregistered, which is challenging. Alternatively, time-integrated activities can be obtained for whole organs and tumours, or sub-regions of organs and tumours. Co-registration is not necessary for acquiring whole organ or tumour TACs. In these cases, regions can be segmented separately to determine the total activity at each time point and a TAC can be fit through the resulting time-activity data. Comparing the different imaging protocols After a pure planar, hybrid planar/SPECT or pure SPECT technique is used to obtain the time-integrated activities for each source organ or voxel, these values can be divided by the injected activity to obtain the TIACs of each source. When comparing the TIACs obtained using the different imaging protocols, the limitations of using a TAC processing method based purely on planar studies have been explored and large discrepancies between 2D and more advanced fully 3D approaches have been reported in simulations [82] and patient studies [83]. These discrepancies can be attributed to the fact that in 2D studies it is 52  Chapter 3: Internal Dose Calculations  impossible to properly compensate for organ overlap, attenuation and scatter. Furthermore, dosimetry estimates based on 2D approaches usually do not consider patient-specific organ masses, instead using reference values based on the average patient. Given the inverse dependence of absorbed dose on mass, dose calculations that consider individual patient masses can vary considerably compared to calculations based on standard organ masses [84]. The hybrid planar/SPECT method has been shown to perform much better than a purely planar approach, providing nearly the same accuracy as a purely SPECT based method for estimating TIACs at the organ level [82,85]. This accuracy makes the hybrid planar/SPECT technique an attractive option for use as part of a practical dosimetry method for clinical use. In many clinical studies, a combination of planar and SPECT/CT scans are already acquired as part of the routine diagnostic procedure. Unfortunately, when applying the hybrid planar/SPECT technique to estimate the 3D dose distribution, it is assumed that the temporal behaviour is constant over the entire volume, and this is not always valid. In this case a purely SPECT based imaging protocol may be required [86]. In summary, there is a need for careful design of the imaging protocol to keep data acquisition acceptable for the patients and to limit the clinical burden in busy nuclear medicine departments, while still acquiring all necessary spatiotemporal information. For example, for multi-exponential fitting, two to three data points are required per exponential term. Furthermore, a general recommendation is to acquire scans over a time period covering at least three effective half-lives of the radiopharmaceutical [87].  3.3 Dose Estimation Method Once the time-integrated activity coefficients for each ROI are determined, the second step in internal dose calculation is to combine this data with the physical data to find the corresponding dose distribution. The physical data includes properties of the injected radionuclide and ideally should also take into account patient-specific anatomy. Dose estimation has traditionally been performed using organ level dose calculation approaches. More advanced methods, such as the MIRD voxel S value approach [88] and different Monte Carlo based techniques provide much better accuracy, but are currently used mostly in research studies. As both voxel S values and Monte Carlo techniques use patient-specific  53  Chapter 3: Internal Dose Calculations  information in their dose calculations, they constitute an important step towards personalized dosimetry. 3.3.1  Methods for organ level dose estimation  Organ level dose calculations are most often performed using the Organ Level Internal Dose Assessment with Exponential Modeling (OLINDA/EXM) software [89]. Using OLINDA/EXM, the user inputs organ TIACs and the program calculates the resulting absorbed dose for each organ. These doses are estimated based on organ level S values calculated for standard phantoms representing the average male or female. As summarized by Eq. (3.2), the dose to  is given by summing up the  contributions from all source organs. The reference female illustrated in Figure 3.6 is an example of one of the phantoms still currently used for most organ level dose calculations (up to version 1.1 of OLINDA/EXM). Organ level S values are calculated by running Monte Carlo simulations for each of these computerized phantoms. To run these simulations, activity is assumed to be uniformly distributed in source organs and the dose is assumed to be uniformly deposited throughout each target organ. Clearly the S values based on these stylized phantoms are not patientspecific. OLINDA/EXM is able to account for patient-specific organ masses, but not for the shapes and positions of organs that can also vary from patient to patient. To correct for patient-specific organ masses for alpha and beta emissions, the dose scales linearly with the mass. For photon emissions, the absorbed fraction scales with the cube root of the mass for self-irradiation and directly with the mass for cross-irradiation [89].  54  Chapter 3: Internal Dose Calculations  Figure 3.6 Reference female used for traditional organ level dose calculation.  Recently, Stabin et al. have reported the next generation of phantoms that are based on more realistic image-based models [90]. Updated S values will be available for this new phantom series and will be used in future versions of OLINDA/EXM. Tumour doses are approximated by OLINDA/EXM using pre-calculated absorbed fractions to spheres of different sizes filled with uniform activity. These spheres are treated as isolated objects, so cross-dose to or from other objects is not accounted for. A major restriction of organ level dose estimation is that these calculations are not intended for assessing radiation risks to individuals. Organ level calculations that use standard reference phantoms may be appropriate for calculating mean doses for diagnostic applications, but voxel level calculations that fully take into account patient-specific anatomy are needed in therapeutic applications for more accurate dose estimation. 3.3.2  Voxel S values  The voxel S value approach [88,91] considers activity distributions at the voxel level and calculates the corresponding voxelized dose distribution. This method uses lookup tables of absorbed dose fractions in an array of target voxels due to radiation emitted from a single  55  Chapter 3: Internal Dose Calculations  source voxel (Figure 3.7). These voxel S value  lookup tables are  precalculated by Monte Carlo simulation [92] and are radionuclide, voxel size and tissue specific. Free databases of voxel S values for selected radionuclides and voxel sizes have also recently become available [93]. Dose calculation with voxel S values is performed by the convolution of the precalculated table of voxel doses per unit of cumulated activity in a source voxel with the patient-specific cumulated activity distribution. In a volume of N source voxels, the dose to each target voxel is calculated using: (3.9)  Figure 3.7 Voxel S values are a table of precalculated doses to an array of voxels given activity in a single source voxel.  The drawback of the voxel S value technique is that the lookup tables are calculated for a source material of uniform density and tissue inhomogeneities are not accounted for. However, the advantage of using the voxel S value approach is that it makes 3D dose calculations simple and fast. Furthermore, the assumption of uniform tissue density when voxelized dose distributions are calculated within organs or tumours is usually a reasonable approximation. 3.3.3  Monte Carlo simulation  Monte Carlo techniques use the known physics of photon and particle interactions to simulate radiation transport. Reconstructed SPECT images provide quantitative information about the activity distribution and radioactive emissions can be simulated and propagated 56  Chapter 3: Internal Dose Calculations  through a computerized patient model to determine the resulting 3D dose distribution. The computerized model can be constructed based on a CT image of the patient, thus the method is able to take into account patient-specific source and target organ geometries and tissue inhomogeneities (Figure 3.8).  Figure 3.8 Example coronal slices from a CT phantom and corresponding activity distribution used as input for patient-specific dose calculation with Monte Carlo simulation.  Monte Carlo simulation is the most robust method for dose estimation, but its use may be quite complicated and it requires very long computation times. Example Monte Carlo codes commonly used for radiotherapy and nuclear medicine applications include the electron gamma shower (EGS) code [94], MCNP [95,96], PENELOPE [97], and the GEANT code [98,99]. The EGS code is considered a gold standard of Monte Carlo codes for radiation dosimetry applications [100]. The latest EGS version, EGSnrc, is accompanied by the user code, DOSXYZnrc, which facilitates calculation of dose distributions in a rectilinear voxel phantom [101-103].  3.4 Software Tools for Internal Dose Calculations A number of quantities that are required to estimate the absorbed dose have been introduced and numerous procedures needed for obtaining the necessary quantities and calculating the absorbed dose from these quantities have been described. These procedures require the processing of a large body of data. To organize this data and to perform each of the different steps necessary for dose estimation, several software tools have been created [89,96,104-118]. However, many of these programs have limited functions and either (i) do not provide enough tools to perform all of the necessary steps from image analysis to  57  Chapter 3: Internal Dose Calculations  determination of TIACs, to calculation of final dose estimates, or (ii) contain only specific tools that do not give enough versatility to account for different dosimetry protocols. An example of the first shortcoming listed above is the popular OLINDA/EXM software [89], which uses TIACs to make an organ level dose estimate. OLINDA/EXM offers the ability to plot activity in source ROIs versus time (time-activity data) and fit this data to a multi-exponential function, which is then integrated to calculate the TIACs. However, separate software is needed to obtain the time-activity data that is used as input for the OLINDA/EXM calculation. Programs such as SPRIND [104] and ULMDOS [105,106] have been developed to provide the necessary tools to collect the time-activity data needed to estimate TIACs, which are then entered manually into OLINDA/EXM. The need to switch between programs to perform different steps of the dose calculation process is not only time consuming, but also increases the likelihood of errors, especially if data is being transferred manually between programs. The major restrictions of organ level dose calculation tools like OLINDA/EXM, SPRIND and ULMDOS were discussed in Section 3.3.1. Programs that have been designed to address some of the limitations of OLINDA type dose calculations include MABDOSE [107] and DOSE3D [108,109]. Both of these programs still use standard reference phantoms. MABDOSE features the ability to place spherical objects representing tumours within the standard reference phantoms rather than using the isolated sphere model. DOSE3D also gives the option to add spherical tumours in addition to the option to adjust organ shapes and sizes. Many dosimetry tools capable of voxelized dose calculations have been developed, including the 3D-ID code [110] now extended to the 3D-RD code [111], SIMDOS [112], RTDS [113], SCMS [96], Voxeldose [114], MrVoxel [115], RMDP [116], OEDIPE [117], and a dosimetry toolkit from Loudos et al. [118]. Of these programs, the code with the most clinical experience is 3D-ID [86,119]. Several imaging protocols used for acquiring TIACs, which include pure planar, pure SPECT and hybrid planar/SPECT imaging, were discussed in Section 3.2.1. Of the computer programs mentioned, some only offer the option to process planar data [104], while others offer tools for analyzing 3D activity distributions [114,116]. Relatively few programs  58  Chapter 3: Internal Dose Calculations  [106,113,115] have the tools to segment ROIs and collect time-activity data from both planar and SPECT images. In addition, a variety of methods that can be used to estimate the absorbed dose from the calculated TIACs have been described (Section 3.3). These methods can be divided into organ level and voxel level dose calculation techniques. Most software tools are designed to calculate organ level doses [89,104,105,107] or voxelized dose distributions [112,114116,118], but not both. Programs without the option to calculate both organ and voxel level dose distributions lack the different advantages that each of these methods has to offer. For voxelized dose calculation, the use of voxel S values (or dose voxel kernels) and Monte Carlo techniques were described in Sections 3.3.2 and 3.3.3, respectively. A third option is the use of dose-point kernels, which represent the distribution of absorbed dose surrounding a point source. Each approach has its advantages and disadvantages. Voxel S value calculations are performed quickly, but assume a uniform tissue medium. Dose point kernels can also be used to calculate the dose relatively quickly, but are not as easily implemented as voxel S values [88]. Monte Carlo simulations allow for heterogeneous tissues to be taken into account for dose calculation, but are computationally intensive. Most 3D dosimetry programs only offer the option to perform Monte Carlo simulation [96,108,112,117] or dose voxel/point kernel calculations [113,114,116,118] and therefore lack the versatility that could be gained from the option to utilize different dose estimation methods for different situations.  3.5 Patient-Specific Dose Calculation Protocol The complete protocol for planning patient-specific treatments in radionuclide therapy is summarized by the flowchart in Figure 3.9.  59  Chapter 3: Internal Dose Calculations  Figure 3.9 Flowchart outlining the necessary steps in patient-specific dose calculation for radionuclide therapies.  One of the unique features of radionuclide therapy is that a tracer amount of radiopharmaceutical can be injected into the patient before treatment. This “test dose” of activity can be used to select patients who are suitable for therapy, determine the patientspecific pharmacokinetics, and create a treatment plan by predicting the radiation dose that would be delivered if a therapeutic level of activity were injected. The pretreatment radiopharmaceutical that is injected could be the same as the radiopharmaceutical that will be used for therapy. However, for radionuclides like 90Y that do not have any gamma emissions, it is not always possible to perform an imaging study. Although bremsstrahlung radiation can be used to create images [120], these images have poor resolution. Alternatively, a gamma emitting surrogate radiotracer can be used, which should ideally mimic the therapeutic agent, but in reality only reflects the biodistribution of the therapeutic agent to a certain degree [14,121]. As part of the pretreatment study, the 4D biodistribution of radiopharmaceutical can be assessed using the techniques discussed in Section 3.2. This process involves the acquisition  60  Chapter 3: Internal Dose Calculations  of serial nuclear medicine images, segmentation of source ROIs and plotting TACs. Exponential fits through the time-activity data yield the values of  for each ROI.  At the next step, the absorbed dose delivered to the patient by a therapeutic injection can be predicted. The TIAC of the therapeutic radiopharmaceutical is identical to the TIAC of the diagnostic radiopharmaceutical if the radionuclides are the same. If the two radionuclides are not identical, then the estimated to determine the  can be combined with the therapeutic radionuclide’s  of the therapeutic agent using Eq. (3.5) and the cumulated activity  using Eq. (3.7). Finally, the absorbed dose at the organ or voxel level can be predicted using one of the methods described in Section 3.3. Use of a pretreatment plan allows for selection of a personalized therapeutic activity that is designed to maximize dose delivered to the tumour, without exceeding the maximum safe dose to healthy tissues. The delivered dose should also be estimated after injection of the therapeutic radiopharmaceutical. This dose can be compared to the prediction from the diagnostic scan so that the therapeutic dose delivered can be verified. In addition, when the absorbed dose is calculated for several treated patients, the results can be pooled to try and identify a doseresponse relationship.  3.6 Summary This chapter has given a general description of what is required to perform an internal dose calculation. Throughout this discussion some limitations of current internal dosimetry techniques have been identified. To address these limitations and towards the thesis objective of investigating and developing new methods that could be used to perform routine dose calculations in the busy clinical environment:  A GUI was developed in MATLAB to serve as a tool for all steps in the internal dose calculation process (presented in Chapter 4).  An iterative image segmentation technique was designed to improve on traditional fixed and adaptive thresholding methods (Chapter 5).  The image reconstruction method’s influence on the accuracy and reproducibility of volume and activity estimates was evaluated to determine whether the image segmentation technique introduced in Chapter 5 can be used to obtain results with  61  Chapter 3: Internal Dose Calculations  similar accuracy when doses are derived from images obtained using a typical clinical SPECT reconstruction and when using a sophisticated quantitative approach that currently is not employed in clinics (Chapter 6).  Tumour and organ dose estimates calculated using organ level, voxel S value, and Monte Carlo techniques were compared (Chapter 7). Finally, the techniques developed in this thesis were employed in two different clinical applications:  Patient-specific, organ level dose calculations were performed for the diagnostic radiotracer  99m  Tc-hydrazinonicotinamide-Tyr3-octreotide  (99mTc-HYNIC-TOC),  injected into patients with suspected neuroendocrine tumours (Chapter 8).  Voxelized dose distributions were calculated in patients treated with  188  Re  microspheres for liver cancer (Chapter 9).  62  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Chapter 4: A Graphical User Interface for Internal Dose Calculations 4.1 Introduction Accurate image-based dosimetry calculations require processing of a large body of data. To organize this data and to efficiently perform each of the steps necessary for dose estimation requires the use of software tools. As described in Section 3.4, several programs have been developed to perform specific tasks, but multiple programs are required to complete the dose calculation, often with the need for in-house software to fill in the gaps. A program that is capable of assisting the user through the entire dose calculation process would be an important contribution that could assist in the performance of routine individualized dose calculations. Towards that end, the objective of the work described in this chapter was to design an internal dosimetry toolkit that can be used for a wide variety of dose calculation approaches and that keeps all aspects of the dose calculation procedure within the framework of a single software environment. The toolkit was developed using the graphical user interface (GUI) design environment in MATLAB 7.10 and can be used for performing all steps in the dose calculation process. In particular, it provides options for considering images obtained using different acquisition protocols (only planar scans or only SPECT scans or a hybrid planar/SPECT protocol). Additionally, it allows the user to co-register a series of planar images, perform 2D and 3D image segmentation and calculate dose distributions using one of three different methods: organ level, voxel S value and Monte Carlo dose calculation.  4.2 Overview of the Dosimetry Tool Patient data collected at various stages of dose calculation is stored in MATLAB structures, which group patient data into databases. The patient data stored inside these databases can be displayed and manipulated from the main GUI. In addition to the main GUI, the toolkit consists of four sub-GUIs (called from the main GUI), which include the Planar 63  Chapter 4: A Graphical User Interface for Internal Dose Calculations  region selection, Planar image registration, Process 3D image data and Organ level dose calculation GUIs. The first three of these sub-GUIs provide tools for collecting all of the information necessary to obtain time-activity data for ROIs from planar and/or SPECT images. Back in the main GUI, curve fits are applied to this time-activity data to construct time-activity curves (TACs), which are then integrated to obtain time-integrated activity coefficients (TIACs) for ROIs. After the TIACs have been calculated, the Organ level dose calculation GUI can be used to calculate mean organ doses that are equivalent to the dose assessment performed by OLINDA/EXM. Alternatively, voxelized dose calculation options are available within the Process 3D image data GUI. Three different dosimetry protocols available in the GUI are summarized in Table 4.1. Dose estimates can be calculated using pure planar, pure SPECT and hybrid planar/SPECT imaging protocols. With pure planar imaging, only organ level dosimetry is possible. For pure SPECT, the current version of the GUI only performs organ level dosimetry. Voxel dosimetry with pure SPECT would require accurate 3D image registration, which is currently not an option in this dosimetry toolkit. The hybrid planar/SPECT technique used by this GUI can provide voxelized dose estimates using the 3D distribution of activity from a single SPECT acquisition and assuming that the kinetic behaviour is constant throughout the volume of interest and can be obtained in 2D from a series of planar images. Table 4.1 Dosimetry protocol options offered in dosimetry toolkit presented in this work.  Protocol  Image acquisition  Dose calculation  2D planar  3D SPECT  Hybrid planar/SPECT  Multiple planar  Multiple  Multiple planar,  SPECT/CT  single SPECT/CT  Organ level  Organ or voxel level  Organ level  An overview of the structure of the developed internal dosimetry toolkit is provided in Figure 4.1. The use of the main GUI and of each of the three sub-GUIs will be discussed in the following sections.  64  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Figure 4.1 Overview of workflow in the dosimetry toolkit between the 5 sub-GUIs, which are the main GUI, the Planar region selection GUI, the Planar image registration GUI, the Process 3D image data GUI and the Organ level dose calculation GUI.  4.3 Description of the Major GUI Functions 4.3.1  Main GUI  The main GUI serves as a central location where the data obtained at each step of the dose calculation procedure can be organized and viewed (Figure 4.2). Specifically, the main GUI is used to begin processing a new patient, load saved data from patients previously analyzed, provide menu options to open the other sub-GUIs where planar and SPECT data is  65  Chapter 4: A Graphical User Interface for Internal Dose Calculations  processed, view the time-activity data resulting from these analyses, and to calculate TIACs from this time-activity data.  Figure 4.2 Screen capture of the main GUI.  The top left part of the main GUI is a ‘Patient Data’ panel where general patient data relevant for dose calculation can be entered. If desired, the user can separate groups of patients into different databases to differentiate patients belonging to different studies or  66  Chapter 4: A Graphical User Interface for Internal Dose Calculations  projects. The patient’s gender and age are displayed, which are each needed if an anthropomorphic phantom is to be used for estimating organ level dosimetry. When a new patient is entered into the database, the date and time of injection should be entered as well as the radionuclide used and the amount of activity injected. After entering this information, the next step is to analyze the imaging data. Depending on the type of data, the user must open the sub-GUI(s) that will be used to load and analyze planar and/or SPECT/CT images to acquire all of the data necessary to plot TACs for ROIs. These sub-GUIs are accessed from the procedure drop-down menu in the main GUI. If planar images have been acquired, then the Planar region selection and Planar image registration sub-GUIs (described in sections II.B. and II.C., respectively) are needed to determine the count rates in ROIs from anterior and posterior views. These count rates are converted to activities using either the conjugate view or single view effective point source method [67]. If SPECT/CT image(s) have been acquired, then the Process 3D image data sub-GUI (described in section II.D.) can be used to segment 3D ROIs and to determine total organ volumes and activities at the time of each acquired SPECT scan. The results of these analyses are accessible from the main GUI. Next, the user returns to the main GUI where the time-activity data generated in the image analysis sub-GUIs can be used to calculate TIACs for each ROI. In the ‘Calculate Organ Level TIAC’ panel (middle left part of Figure 4.2), the user must first select a region from the dropdown menu, which will contain a list of organs and tumours that have been segmented in the planar and/or SPECT images. After a region is selected, results from the planar and/or SPECT image analysis corresponding to that region can be viewed from the two panels on the right hand side of the main GUI. If planar image analysis was performed, the results can be viewed in the top right ‘Planar Data’ panel, where the total number of planar scans acquired is displayed, and the times of these planar scans are listed in a table. For each region there is the possibility of storing multiple datasets corresponding to different ways the data was analyzed. These can be used for research purposes to compare different image processing techniques. For example, two separate datasets for a single ROI could be created to compare the use of two different thresholds to segment the region. The dataset that has been set as the main planar dataset is used to fill in the activity column of the planar data table (Figure 4.2).  67  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Results from the SPECT/CT analysis can be viewed in the bottom right ‘SPECT Data’ panel of the main GUI. Similar to the layout of the planar data panel, the SPECT data panel includes information about the number of SPECT scans acquired, the regions included in the SPECT analysis, and the option to view several different datasets corresponding to multiple ways the same region was analyzed. This could be used for research investigations, and also when using techniques that use different segmentations to obtain total region activity and volume estimates separately. For example, multiple datasets saved for a particular ROI could correspond to the same region segmented within SPECT images generated using different thresholds. This is necessary when using the iterative adaptive thresholding technique that will be described in Chapter 5. Once the ROI and appropriate datasets are selected, the activity and volume determined from each SPECT imaging time point are listed in a table on the ‘SPECT Data’ panel. Next, a method for acquiring a TAC from the information stored in the planar and SPECT data tables needs to be selected using the ‘Calculate Organ Level TIAC’ panel. Options for the TAC processing method include hybrid planar/SPECT, pure planar and pure SPECT techniques. The TAC can be plotted using a multi-exponential fit through the data or using trapezoidal areas. For multi-exponential analysis, fit options include selection of the number of exponential terms, starting estimates and setting of lower and upper bounds for the unknown coefficients. For trapezoidal areas, fit options include a choice of how the area under the curve is estimated after the last data point. Pressing the ‘Calculate TIAC’ button will perform the desired fit and integrate the curve to obtain the TIAC. Finally, the ‘Replace Radionuclide’ panel can be used in cases where the biodistribution information obtained from the study performed with a diagnostic radiopharmaceutical is being used to predict the dose that would be deposited from a therapeutic injection using a different isotope. The user simply needs to select a replacement radionuclide from the dropdown menu and the TIAC for the therapeutic injection is calculated and displayed. The procedures discussed above can be illustrated in the example shown in the screen capture of the main GUI displayed in Figure 4.2. In this case the data is loaded for patient 19 from the “Tektrotyd” database. This patient was injected with 810 MBq of 99mTc. Four planar scans and one SPECT/CT scan have been acquired for this patient and the current region under investigation is the left kidney. A hybrid planar/SPECT TAC processing method and  68  Chapter 4: A Graphical User Interface for Internal Dose Calculations  multi-exponential fit have been selected for plotting the TAC. The TIAC calculated from integration of the TAC is 0.1175 MBq·h/MBq. In the ‘Replace Radionuclide’ panel,  188  Re  has been selected as a therapeutic agent. Assuming that the biodistribution remains the same as the 188  99m  Tc administration, with the same biological half-life, the predicted TIAC of the  Re administration for the left kidney is 0.3133 MBq·h/MBq.  4.3.2  Planar region selection GUI  The Planar region selection GUI (Figure 4.3) is the first of two sub-GUIs used for collecting time-activity data from planar images. This GUI is used to load a time series of planar images and delineate two-dimensional ROIs on all of them.  Figure 4.3 Screen capture of the Planar region selection GUI. Anterior (left) and posterior (right) views from the first of three planar scans is displayed. A region of interest delineating the spleen and surrounding background is drawn (solid magenta line) on the posterior view and automatically copied to the anterior. In addition, a background region is drawn (solid green line) to be used for geometric background subtraction.  After the planar images are loaded, anterior and posterior scans are viewed side-by-side. Usual image display options such as zoom and windowing are available. The analysis begins by the user manually drawing ROIs on the areas surrounding organs and tumours of interest on either the anterior or posterior view. These ROIs are then copied onto the opposite view.  69  Chapter 4: A Graphical User Interface for Internal Dose Calculations  At this step these outlines do not need to accurately segment the organs and tumours; they just need to separate regions encompassing organs and surrounding background from other activity filled regions. Two-dimensional segmentation is performed at a later step inside these outlines. After each ROI is drawn, the user is prompted to draw a background ROI, which can be used later for geometric background subtraction using the conjugate view method. ROIs and corresponding background regions should be delineated at each time point. After these regions have been drawn, the Planar image registration GUI can be accessed from the Planar region selection GUI. 4.3.3  Planar image registration GUI  The Planar image registration GUI (Figure 4.4) offers tools for manually or automatically registering the ROIs drawn in the Planar region selection GUI. If an automatic registration is used, a reference image must be chosen from an ROI drawn at one time point and the method to use for segmentation inside the outlines drawn on the reference image must be specified. For example, the reference image could be chosen as the ROI surrounding the left kidney acquired from the posterior planar scan at the second time point and a 50% threshold could be chosen as the segmentation method. In this case, a reference binary mask of the left kidney is created by taking all pixels with counts greater than or equal to 50% of the maximum counts within the outline of the left kidney and surrounding background drawn in the second planar image. The ROIs drawn in the remaining planar images are then automatically registered to the reference binary mask. This is accomplished by an algorithm that translates and rotates each ROI and finds the x-y shift values and rotation angles that maximize the number of counts inside the reference binary mask. This function serves as a semi-automatic planar registration process, since it does require the user to draw approximate regions at all time points in order to separate neighbouring regions with high uptake.  70  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Figure 4.4 Screen capture of the Planar image registration GUI. The top left window displays the reference image, which in this case is a posterior view of the spleen in the first of three whole body planar scans. The bottom left window shows the image of the spleen from the second whole body planar scan registered to the reference image. This registration was performed using the semi-automatic algorithm, which shifted the image to register by one pixel in the vertical direction and rotated it by three degrees in the counter clockwise direction. The plots in the top right and bottom right windows are profiles through the reference image and the registered image, which serve to check the accuracy of the registration.  During this process, the registered ROIs can be examined to check the accuracy of alignment and to apply manual shifts or rotations if desired. A table in the ‘Update Database’ panel displays the anterior and posterior counts at each time point for the currently selected region. This data can be saved, and then becomes available to view from the main GUI. 4.3.4  Process 3D image data GUI  If patient datasets contain SPECT/CT image(s), the Process 3D image data GUI (Figure 4.5) is used to load reconstructed SPECT and CT images, segment 3D ROIs to find total region volumes and activities, and to perform voxelized dose calculations using voxel S values or Monte Carlo simulation with the EGSnrc user code, DOSXYZnrc [101,102]. 71  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Figure 4.5 Screen capture of the Process 3D image data GUI demonstrating how reconstructed SPECT images are analyzed. The user draws regions slice by slice in the top left window, with a choice of using a transaxial, coronal or sagittal view. The remaining two views are displayed in the top right and bottom right windows, which are used to check the slice position. The bottom left window is used for different purposes and in this case it shows outlines of the regions segmented from the nuclear medicine image on the corresponding CT slice.  The loaded 3D SPECT or CT image is displayed on three plots corresponding to transaxial, coronal and sagittal views. Any one of these image orientations can be chosen as the main view, which is displayed in the upper left plot. The remaining two views shown in the upper and lower right contain markers indicating the corresponding position of the slice currently being displayed in the main view. The plot in the lower left is reserved for viewing the results of image segmentation. Options related to the image display include windowing, zoom, a choice of slice view or maximum intensity projection, and post-reconstruction filtering. Image segmentation tools available in this GUI include manual region drawing in SPECT and CT images as well as fixed and adaptive thresholding of SPECT images. In general,  72  Chapter 4: A Graphical User Interface for Internal Dose Calculations  adaptive thresholding methods use a threshold that depends on the source-to-background ratio (SBR) of activity concentrations. This GUI features the iterative adaptive thresholding technique where background regions are chosen semi-automatically [122]. Alternatively, the GUI can be used to perform traditional adaptive thresholding, where background regions are drawn manually by the user. After the segmentation is completed, the user can check the segmentation results in the lower left plot slice-by-slice. Outlines of delineated regions from SPECT segmentation can be displayed on the corresponding CT image slices and vice versa. Progress is saved so that the analysis can be resumed or modified at a later time. At the next step, the Process 3D image data GUI can be used to calculate voxelized dose rate distributions from a single SPECT image using Monte Carlo or voxel S values. These calculated dose rate distributions can be converted to dose distributions using TACs from a hybrid planar/SPECT approach. Using this approach, planar time-activity data is used to determine the general shape of TACs for ROIs. These TACs provide information about the rate of uptake and washout needed to convert the dose rate maps to absolute dose distributions, under the assumption that the rate of uptake and washout does not vary between individual voxels within the 3D ROI. For voxelized dose calculation with Monte Carlo simulation, the GUI provides all of the necessary tools for creating input files and running the EGSnrc user code, DOSXYZnrc. First, source and target regions are selected, which can encompass the entire image or they can correspond to one or more segmented regions. Coordinates of the selected regions and the CT image dataset are used by the program ctcreate to generate a CT phantom in which the dose distribution is calculated by DOSXYZnrc. Since the time required by this Monte Carlo calculation depends on the size of the volume for which dose is calculated, simulations will run faster when this volume is just big enough to contain the source and target regions rather than the entire CT dataset. After the CT phantom is generated, the next step is to choose parameters for the Monte Carlo simulation, which include the number of histories, the cutoff energy for electron and photon transport and the option to turn on and off various physical effects (Figure 4.6). Finally, the Monte Carlo simulation is started and cycles through each SPECT image voxel within the selected source region, treating each voxel as a single source, which allows for dose calculation based on heterogeneous activity distributions.  73  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Figure 4.6 A sub-GUI that is called from the Process 3D image data GUI for creating DOSXYZnrc/EGSnrc input files.  Another useful application of the Monte Carlo capabilities of the GUI is the option to calculate organ level patient-specific S values for any radionuclide. These S values can be generated using regions segmented from a SPECT/CT or just a CT scan of a patient. The second option for performing voxelized dose estimation within the Process 3D image data GUI is using voxel S values. Voxel S values are stored in the program database in the form of radionuclide, voxel size and tissue specific lookup tables that represent the 3D dose distribution in a homogeneous medium given activity in a single source voxel [88]. These lookup tables can be generated by Monte Carlo simulation. As an alternative to voxel level dose calculation, the total activities and volumes of segmented ROIs determined from SPECT/CT data can be used for organ level dose calculation. In this case, ROI activities and volumes are saved to the patient database and can then be viewed from the main GUI. Within the main GUI, the total activities determined from SPECT images at multiple time points can be used to create organ level TACs based on pure SPECT data, or activity determined from SPECT at a single time point can be used in  74  Chapter 4: A Graphical User Interface for Internal Dose Calculations  combination with planar time-activity data as part of the hybrid planar/SPECT method. As stated in the description of the main GUI, the plotted TACs are integrated to calculate regional TIACs, which in turn are used for organ level dose estimation in the Organ level dose calculation GUI. 4.3.5  Organ level dose calculation GUI  Organ level TIACs estimated using this dosimetry toolkit can be used as input for dose calculation with OLINDA/EXM. To keep the procedure within a single software environment, this dose calculation option is available from the Organ level dose calculation GUI (Figure 4.7), which can be called from the main GUI.  Figure 4.7 Screen capture of the Organ level dose calculation GUI.  When the Organ level dose calculation GUI is opened, the first step is to ensure that the appropriate reference phantom and radionuclide are selected. An identical dose calculation to the calculation performed by the OLINDA/EXM software will be performed if the user has access to the S values associated with the 10 reference phantoms of Cristy and Eckerman [123] and of Stabin et al. [124], which are used by OLINDA. In addition, the user can upload his/her own set of reference phantoms and associated S values, which can for example be generated within the Process 3D image data GUI. The program adds TIAC results and 75  Chapter 4: A Graphical User Interface for Internal Dose Calculations  patient-specific organ masses from investigated organs into the organ TIACs and masses table. If necessary, the TIACs and masses from more than one investigated region can be summed. This is useful for combining TIAC and mass measurements from paired organs (such as kidneys). The dose calculation uses the same method to account for patient-specific organ masses as the OLINDA/EXM software. For alpha and beta emissions, the dose scales linearly with the mass. For photon emissions, the dose scales with the cube root of the mass for selfirradiation and directly with the mass for cross-irradiation (Section 3.3.1). The Organ level dose calculation GUI does not include the sphere model, since tumour doses can be calculated using Monte Carlo or the voxel S value approach.  4.4 Results and Discussion The developed GUI can be used to calculate patient-specific absorbed dose distributions at the organ and voxel levels. Tools are provided for every step in the dose calculation process from loading the images to dose estimation. This eliminates the need for manual data transfer between programs, which saves times and minimizes user errors. Several options are available at each step in the dose calculation process. For example, data can be processed using a pure planar, pure SPECT or hybrid planar/SPECT technique, and the final dose estimate can be obtained using Monte Carlo, voxel S values, or an organ level dose calculation technique. Since OLINDA/EXM is currently the most widely used dosimetry program, it was important for the GUI to have the option to perform an OLINDA-based dose calculation. The advantage of using this GUI is that it combines several functions into a single environment. First, it can be used to collect all of the time-activity data needed to plot and integrate TACs, and then it can be used to perform the organ level dose calculation without the need to open the OLINDA/EXM program. This calculation does require the user to have access to the organ level S values included in OLINDA/EXM. Alternatively, the ability to generate S values from any CT dataset is a useful feature. For example, this option could be used to create patient-specific S values from a diagnostic scan of a patient scheduled for radionuclide therapy. These precalculated patient-specific S values could then be applied for predicting or calculating organ level doses from a therapeutic injection.  76  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Different functions of the GUI are exemplified in this chapter using a hybrid set of planar and SPECT/CT images and dose calculation results are provided. The TIACs from an injection of 99mTc displayed in the screen capture of the Organ level dose calculation GUI in Figure 4.7 are 0.4207, 1.0625, 0.7796 MBq·h/MBq for the kidneys, liver and spleen, respectively. The patient-specific organ masses are 421.0, 2537.3, 434.6 g for the kidneys, liver and spleen, respectively. The GUI output for the organ dose calculation using these TIACs, organ masses and the adult male reference phantom used by OLINDA are listed in Table 4.2. These organ doses are identical to the output from the OLINDA/EXM software. A Monte Carlo simulation using the Process 3D image data GUI was also performed for this patient. The mean kidney, liver and spleen doses and range of doses calculated from the Monte Carlo 3D dose distribution data are listed in Table 4.3. While the organ level dose calculation is performed practically instantaneously at the press of a button, the Monte Carlo simulation of this patient dataset took approximately 30 hours on a Windows 7 (Microsoft) machine with a 2.93 GHz Intel Core i7 processor and 8 GB of RAM. This Monte Carlo calculation was performed by simulating 100,000 photon histories and 100,000 electron histories per source voxel, where source voxels only included voxels within delineated ROIs, which in this case were the kidneys, liver and spleen.  77  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Table 4.2 Example output from organ level absorbed dose calculation.  Absorbed Dose (mGy/MBq) 'Organ'  'Beta'  'Photon'  'Total'  'Adrenals'  0.00E+00  4.05E-03  4.05E-03  'Brain'  0.00E+00  4.82E-06  4.82E-06  'Breasts'  0.00E+00  4.13E-04  4.13E-04  'Gallbladder Wall'  0.00E+00  4.32E-03  4.32E-03  'LLI Wall'  0.00E+00  2.69E-04  2.69E-04  'Small Intestine'  0.00E+00  1.05E-03  1.05E-03  'Stomach Wall'  0.00E+00  3.12E-03  3.12E-03  'ULI Wall'  0.00E+00  1.34E-03  1.34E-03  'Heart Wall'  0.00E+00  1.49E-03  1.49E-03  'Kidneys'  8.66E-03  8.61E-03  1.73E-02  'Liver'  3.63E-03  6.54E-03  1.02E-02  'Lungs'  0.00E+00  1.36E-03  1.36E-03  'Muscle'  0.00E+00  7.26E-04  7.26E-04  'Ovaries'  0.00E+00  3.60E-04  3.60E-04  'Pancreas'  0.00E+00  5.82E-03  5.82E-03  'Red Marrow'  0.00E+00  8.13E-04  8.13E-04  'Osteogenic Cells'  0.00E+00  1.07E-03  1.07E-03  'Skin'  0.00E+00  2.94E-04  2.94E-04  'Spleen'  1.56E-02  1.54E-02  3.09E-02  'Testes'  0.00E+00  1.68E-05  1.68E-05  'Thymus'  0.00E+00  3.63E-04  3.63E-04  'Thyroid'  0.00E+00  5.95E-05  5.95E-05  'Urinary Bladder Wall'  0.00E+00  9.54E-05  9.54E-05  'Uterus'  0.00E+00  2.95E-04  2.95E-04  'Total Body'  2.67E-04  1.03E-03  1.29E-03  78  Chapter 4: A Graphical User Interface for Internal Dose Calculations  Table 4.3 Organ dose estimates obtained using the Organ level dose calculation GUI and Monte Carlo simulation with the Process 3D image data GUI.  Absorbed Dose (mGy) Organs  Organ level  Monte Carloa  Kidneys  15.2  14.9 (4.6 – 37.9)  Liver  9.0  9.1 (3.3 – 25.2)  Spleen  27.2  27.9 (11.1 – 49.7)  a  Values in parentheses are the range of doses from minimum to  maximum voxel dose.  4.5 Conclusion The internal dosimetry toolkit presented in this work improves on the currently available internal dosimetry software by providing a versatile set of tools for each step in the dose calculation process. It can be used to efficiently perform patient-specific internal dose calculations in a variety of clinical situations. Although, it is difficult to include every possible dosimetry method into the program, the GUI is easily customized within the MATLAB environment to meet specific user needs and interests. Availability of a program such as this in clinics is an important step towards routine individualized treatment planning and dose calculations.  79  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities in SPECT Images 5.1 Introduction In SPECT-guided radionuclide therapy, the accuracy of calculated dose strongly depends on the accuracy of both the volume and the activity estimates of the object under investigation. The method employed to obtain these estimates must not only be accurate, but should also be reproducible, robust and easy to use in a busy clinical department. Currently, most segmentation methods used in internal dose calculations do not meet these requirements (Section 3.2.2). The aim of the work presented in this chapter was to develop a practical and reproducible image segmentation method for calculations of total absorbed dose in organs and tumours. In particular, a modified iterative adaptive thresholding algorithm that is designed to accurately estimate tumour/organ volumes and activities, and which produces consistent estimation of the source-to-background ratio (SBR) of activity concentrations is described. Its two main characteristics are as follows. First, since the same segmented region cannot be used for estimating volume and activity, the proposed method uses two different thresholds – one for volume and another for activity determination. Second, manual background region drawing utilized by traditional adaptive thresholding approaches, is replaced by an iterative technique that features semi-automatic selection of the proper background region. The proposed method is intended to be used for total absorbed dose calculation in organs and tumours, such as those performed using the OLINDA/EXM software [89]. In this chapter, to introduce this novel iterative adaptive thresholding method: the setup of the calibration phantom experiment needed to relate optimal thresholds to measured SBR values is described, the algorithms used to reconstruct the SPECT acquisitions of the calibration phantom are outlined, the methods used to process the calibration phantom data to 80  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  construct threshold-SBR curves are presented, and details of the iterative adaptive thresholding method are explained. Next, the results of a validation phantom experiment designed to test the accuracy of the method are reported. The influence of the SPECT reconstruction method on the accuracy of volume and activity estimates is also evaluated. This investigation was performed to determine whether the proposed method can be used to obtain similarly accurate dose estimates when volume and activity estimates are derived from images acquired using a typical clinical SPECT reconstruction as well as using a sophisticated quantitative approach that is currently not used in clinics. The robustness of the method is further examined by analyzing the stability of the results as a function of different parameters selected to initiate the iterative process. This analysis includes evaluation of intraobserver and interobserver variability of the segmentation results.  5.2 Materials and Methods 5.2.1  Calibration phantom experiment  The first step in applying the adaptive thresholding technique is to perform a calibration phantom experiment. This experiment was done using phantoms filled with active (99mTc) water and containing objects with known volumes and activities. Twenty activity filled cylindrical bottles with volumes of 12 mL (x2), 17 mL (x2), 33 mL (x8), 76 mL (x1), 120 mL (x5) and 200 mL (x2), were placed inside four large cylindrical (Jaszczak; Biodex, Shirley, New York, USA) phantoms. These volumes were measured by finding the difference in mass between each bottle filled with water and empty and dividing by the density of water. The activity concentrations in the bottles varied from 20 to 200 kBq/mL and the background concentrations in the different phantoms varied from 6 to 12 kBq/mL, setting up a range of SBRs from approximately 2 to 20. A SPECT/CT scan of each phantom was performed with 60 camera stops (20 seconds per stop) over 360° and 128x128 projection matrices with a pixel size of 4.42 mm. All experiments were performed on an Infinia Hawkeye 4 SPECT/CT camera (GE Healthcare, Waukesha, Wisconsin, USA), equipped with a low energy high resolution (LEHR) collimator.  81  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  5.2.2  SPECT image reconstruction and activity quantification  The data from these SPECT/CT phantom acquisitions was reconstructed using three different methods, each of which used the OSEM algorithm (2.9.2) with different sets of corrections for attenuation, scatter and spatial resolution compensation (Table 5.1). The first reconstruction, referred to as the Clinical method, was conducted on a clinical Xeleris workstation (GE Healthcare) using the manufacturer’s software with its recommended settings. The other two reconstructions, referred to as the BBACRR and SCACRR methods, used in-house software with previously optimized parameters [74]. For the BBACRR method, attenuation correction was performed using a CT-based attenuation map scaled (by a factor of 0.8) to broad-beam values. As described in Section 2.11.2, broad-beam attenuation correction is a simple and commonly used approach to account for scattered photons in the measured projections. Attenuation correction and resolution recovery corrections were incorporated directly into the system matrix. The SCACRR method included attenuation correction, resolution recovery and scatter modelling. As with the BBACRR method, attenuation correction and resolution recovery were incorporated directly into the system matrix  . However, in this case, attenuation  correction was performed using CT-based attenuation maps scaled to narrow-beam values. The scatter correction component  was only incorporated in the forward step (in the  denominator) of the OSEM algorithm: (5.1) The scatter component  was calculated using the analytical photon distribution interpolated  (APDI) method [50,51], which calculates the distribution of scattered photons from an initial estimate of the 3D activity distribution. In summary, the entire procedure for reconstruction with the SCACRR method is performed in a series of three steps. First, an initial estimate of the 3D activity distribution must be made. In this case, the BBACRR reconstruction served as the initial estimate. Next, APDI is used to calculate scatter projections from the BBACRR image and the narrow-beam attenuation map. Finally, an additional OSEM reconstruction is performed using the calculated scatter projections in the forward projection step (Eq. (5.1)). The SCACRR method served as a gold standard to which the other methods were compared since it contained the most advanced set of corrections. 82  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  A planar, in air calibration experiment (as described in Section 2.12) was performed to estimate the sensitivity factor needed to convert SPECT counts to activity. Table 5.1 Summary of parameters for the different reconstruction methods.  Clinical  BBACRR  SCACRR  Yes (broad-beam)  Yes (broad-beam)  Yes (narrow-beam)  Resolution recovery  No  Yes  Yes  Scatter correction  No  No  Yes  2 (10)  6 (10)  6 (10)  Smoothing filter (cutoff/critical  Hanning (0.4  Butterworth (0.4  Butterworth (0.4  frequency, order)  cycles/pixel)  cycles/pixel, 8)  cycles/pixel, 8)  Attenuation correction  # iterations (subsets)  5.2.3  Processing calibration phantom data  In order to determine the parameters of the functions relating the two proposed thresholds to the SBR values, the calibration phantom data was analyzed as follows. First, rough 3D volumes of interest (VOI0) enclosing each bottle and some surrounding background were created by drawing a series of 2D regions in consecutive transaxial slices. Then, different thresholds, calculated as percentages of the maximum activity, were applied to this data until the optimal threshold (ThV) that returned each bottle’s true volume (corresponding to bottle volumes listed in Section 5.2.1) was found. The maximum activity in this case was defined as the average activity in N of the hottest voxels in each bottle, where N was set at 1% of the total number of voxels in the bottle (the closest integer number). Although 1% is an arbitrary number, the motivation was to make the number of voxels used to define the maximum activity dependent on the total object volume. In the same manner, the optimal threshold (ThA) needed to recover each bottle’s true activity was determined by adjusting the value of the second threshold until the true bottle activity was reached. As expected, due to PVE the volume determined by ThA was larger than that determined by ThV (Figure 5.1).  83  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  Figure 5.1 (a) Transaxial SPECT slice of the image of a bottle, showing region boundaries delineated by thresholds ThV (dashed green line) and ThA (dotted red line). The white solid line corresponds to the region manually drawn by the operator. (b) A plot of the activity concentration profile drawn through this slice with labelled positions of the true volume and activity boundaries.  In the next step, the SBR for each analyzed bottle/background combination was determined. Since the adaptive thresholding method was intended to be applied to patient data where the true SBR is unknown and can only be estimated from the acquired image, the SBRs of the phantom experiment were also derived from the image rather than being defined using the known true values. To accomplish this, the SBR for each bottle was found by first defining the source activity concentration as the average activity concentration inside the volume delineated using ThV. The background activity concentration was set equal to the average activity concentration outside the larger volume delineated using ThA, but still inside the original volume, VOI0, drawn to include the bottle and surrounding background activity. This approach allowed for exclusion of the object spill-out activity from background estimation. The optimal thresholds ThV and ThA, obtained in the previous step, were plotted versus SBR to create two datasets. These datasets were then fitted with the inverse functions: (5.2) (5.3) The data from all bottles, regardless of volume, were included in the same plot. This process was performed for images of the phantom reconstructed using the Clinical, BBACRR 84  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  and SCACRR methods so that six sets of curve fit parameters (two thresholds for each of the three reconstruction methods being investigated) were acquired. 5.2.4  Description of the iterative adaptive thresholding technique  The iterative technique proposed in this study for determination of the SBR is summarized by the flowchart in Figure 5.2. The procedure was incorporated into the dose calculation GUI described in Chapter 4, which substantially simplified the data processing. To begin, the range of slices encompassing the object of interest is determined and an initial rough 2D ROI is manually drawn around the object in one of these slices. Then, an initial volume (VOI0) is created by copying (with modifications to avoid surrounding organs, if necessary) this ROI to the remaining transaxial or coronal slices of the SPECT image. The resulting volume contains the investigated object as well as some of the surrounding background. This is a very simple and fast procedure and as will be shown later in the intraand inter-observer analysis, the initial VOI0 drawing does not influence the final segmentation results. Next, the initial estimate of the object volume (VOI1) is acquired using a threshold set at 40% of the maximum counts in VOI0. This VOI1, is used to generate an initial guess of the SBR (SBR1) by setting the source activity concentration (S1) to the average activity concentration inside VOI1 and the background activity concentration (B1) to the average activity concentration outside VOI1, but inside VOI0. This initial estimate of the SBR may be quite far from the truth because the background region defined in this way may be affected by spill-out from the object. At the next step, the value of SBR1 is used to find the corresponding values of ThV1 and ThA1 from Eqs. (5.2) and (5.3) using the parameters obtained from the calibration phantom experiment (reconstructed with exactly the same algorithm as the patient data). An updated estimate of the source region activity concentration (S2) is generated from the region delineated inside VOI0 using ThV1, which is a procedure analogous to the one that was used in the calibration phantom experiment. Similarly, an updated estimate of the background activity concentration (B2) is acquired by applying the threshold ThA1 and finding the activity outside the resulting segmented region, but inside VOI0. The values of S2 and B2 lead to the calculation of SBR2, which in turn is used to find updated thresholds ThV2 and ThA2. These updated thresholds are again used to find a value 85  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  of SBR3 using the same procedure as outlined above and the process is repeated iteratively until the percentage difference between SBRi and SBRi+1 is less than a predefined value which in this case was set at 0.01%. The adaptive threshold values of ThVf and ThAf, corresponding to the final estimate of the SBR (SBRf) are considered to be the optimal thresholds necessary for the recovery of “true” target volume and activity. Since initially the volume of the investigated object is unknown, the maximum activity is defined as the average activity in the 10 hottest voxels inside VOI0 for the 40% threshold used to start the iterative process. In subsequent steps, the same procedure as used for processing the calibration phantom data is applied, where estimates of the maximum activity (average of the N hottest voxels) are updated by setting Ni equal to 1% of the total number of voxels in VOIi.  86  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  Figure 5.2 Flowchart illustrating the iterative adaptive thresholding technique.  87  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  5.2.5  Validation phantom experiment  The accuracy of volumes and activities estimated using the iterative adaptive thresholding technique was evaluated using a second phantom experiment. This validation experiment consisted of bottles with known volumes and activities placed inside two large cylindrical phantoms (in addition to the four large phantoms used for calibration), which were filled with background activity. The validation phantoms contained a total of ten bottles with volumes of 12 mL (x2), 17 mL (x2), 33 mL (x2), 76 mL (x1), 120 mL (x2) and 200 mL (x1). SBRs of approximately 2 to 15 were generated. A SPECT/CT was acquired for each validation phantom using the same acquisition parameters as were used for the calibration experiment. Although it cannot be claimed that the bottles accurately modeled different tumours and organs, they represented a wide range of sizes and shapes. Certainly these bottles were more irregular than spheres, and thus provided more challenge for both establishing calibration parameters as well as for validation of the segmentation technique. Furthermore, the variety of investigated bottle shapes and sizes allowed for verification that, indeed, a single calibration curve can provide accurate threshold values for segmenting objects of different sizes and activities. While the calibration represented an average over the different bottle volumes, the validation experiment served to test the accuracy of this calibrated threshold-SBR relationship on individual bottles ranging in volume from 12 mL to 200 mL. A schematic of the validation phantom setup is illustrated in Figure 5.3 with the corresponding bottle volumes and activities.  88  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  Figure 5.3 Schematic diagram of validation setup showing an axial view of phantoms 1 and 2 with background concentrations of 12 kBq/mL and 9 kBq/mL, respectively. The numbers besides each container indicate its volume (mL) and activity (MBq) in parentheses.  The iterative adaptive thresholding technique was used to estimate volumes and activities of each of the 10 bottles for images reconstructed using the Clinical, BBACRR and SCACRR methods. The percentage errors of these volume and activity estimates relative to the true values were calculated. 5.2.6  Patient studies  At the next stage, the method was applied to patient studies to investigate convergence of the iterative method and to perform intraobserver and interobserver studies. Four patients with diagnosed or suspected neuroendocrine tumours were injected with 800-910 MBq of 99m  Tc-HYNIC-TOC. A SPECT/CT scan was performed for each patient using identical  acquisition parameters as used for the phantom experiments. 5.2.7  Dependence of volume and activity estimates on the initial parameters  Influence of the initial SBR estimate In order to assess the robustness of the method, three of the patient datasets were used to test how well the final estimate of the SBR (SBRf) converged to a consistent value, and to see how sensitive the iterative method was to the selection of initial parameters. For this test, the iterative process was repeated for each patient dataset using three different starting estimates of the SBR: two extreme values of 2 and 50, and the SBR estimated using a starting threshold of 40%. The agreement in SBRf’s resulting from these different starting estimates was 89  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  investigated using ten separate regions that included the liver, spleen, kidneys and tumours of the three patients. Intraobserver and interobserver variability An intraobserver and interobserver analysis was performed to assess the dependence of the final volume and activity estimates of organs and tumours on the initial VOI0 selection. The intraobserver and interobserver analysis included two observers who drew the VOI0 as required by the iterative thresholding method. The analysis was performed on two different regions, which included a tumour and a left kidney. The tumour represented a simple case with no significant sources of background activity. The left kidney represented a more difficult situation given its location next to the spleen, which contained a similar activity concentration. Each observer repeated the drawing of these VOI0’s five times in one week intervals. The average and standard deviation of volumes and activities were determined from each set of 5 measurements made by each observer. To compare the interobserver measurements, the mean values of volume and activity obtained by each observer were compared using a t test with a 95% confidence level. In addition to the t test, the interobserver agreement was assessed by comparing the overlap of the segmented regions. This was evaluated using the concordance index, which is defined as the ratio of the volume of intersection to the volume of union of the two regions being compared. A concordance index of 1 signifies equal volumes and complete overlap between the two regions and a value of 0 results if there is complete disagreement between the two regions. Calculation of the concordance index is a more stringent test of agreement than a comparison of the total volumes measured. For example, two regions might be determined to have the same total volume, but if just half of one region is overlapping the other, then the concordance index will only be 0.33. The concordance index was calculated for every possible pair of regions segmented by each observer, which resulted in 25 concordance indices calculated for the set of 5 measurements. These concordance indices were calculated to compare the regions drawn manually by each user at the beginning of the procedure and also to compare the final regions obtained with the iterative thresholding method from each of the starting points.  90  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  5.3 Results 5.3.1  Analysis of the calibration phantom experiment  The threshold-SBR data is plotted in Figure 5.4 and the curve fit parameters corresponding to Eqs. (5.2) and (5.3) are summarized in Table 5.2. Out of the three reconstruction methods considered, images reconstructed using the Clinical method required the highest threshold for recovery of true volume, while images reconstructed using the SCACRR method required the lowest threshold. In contrast, the threshold needed to recover true activity was lower when using the Clinical reconstruction compared to the SCACRR image.  Figure 5.4 Functions relating threshold values ThV and ThA to SBR obtained from the phantom data reconstructed using the Clinical (a), BBACRR (b) and SCACRR (c) algorithms. Table 5.2 Curve fit parameters for threshold-SBR data from the calibration phantom experiment reconstructed using three different algorithms.  ThV  ThA  a  b  R2  c  d  R2  Clinical  37.3  42.8  0.90  1.5  93.4  0.98  BBACRR  29.4  43.6  0.80  3.1  86.3  0.97  SCACRR  27.2  42.1  0.82  10.2  68.7  0.93  Reconstructions  Parameters a, b, c and d are defined in Eqs. (5.2) and (5.3), and R2 is the coefficient of determination.  5.3.2  Validation of the iterative adaptive thresholding technique  Results for each of the ten bottles used in the validation phantom experiment, where the accuracy of volume and activity estimates was assessed, are illustrated in Figure 5.5. The average percentage errors (± standard deviation) in the bottle volumes determined by ThV were 3.6 ± 3.2%, 3.7 ± 2.7% and 2.8 ± 2.7% for the Clinical, BBACRR and SCACRR methods, respectively. The total activity in these bottles measured by ThA was estimated 91  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  within 2.9 ± 2.0%, 2.2 ± 0.8% and 2.0 ± 1.6% of the true values for the Clinical, BBACRR and SCACRR methods, respectively. The accuracy of the volume and activity estimates did not appear to depend on the bottle volume, for the range of bottle volumes analyzed.  Figure 5.5 The ratios of the measured to true volumes (a) and activities (b) for the whole range of values investigated in the validation experiment, as determined using ThV and ThA, respectively. The data were reconstructed using the Clinical, BBACRR and SCACRR methods.  5.3.3  Influence of initial parameters on the final volume and activity estimates  Influence of the initial SBR estimate Typically 5-10 iterations were required before the convergence requirement (0.01% difference between SBRi and SBRi+1) was reached. An example convergence plot of SBRi versus the iteration number (i) for each of the three starting estimates that were used in this test is displayed in Figure 5.6. In this example, and in the majority of cases analyzed, the SBRf’s resulting from different starting estimates of the SBR agreed within 0.1%. The worst agreement observed was a 0.6% difference between the SBRf’s estimated for the spleen of patient 2 when the iterative procedure was initiated using starting SBR estimates of 2 and 50. However, even in this worst case, when these two different SBR’s were used to obtain the final volume and activity of the spleen, the two volume estimates were identical and the activity estimates only differed by 0.1%.  92  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  Figure 5.6 An example of a plot showing the convergence of the SBR estimated by the iterative adaptive thresholding technique for three different initial values of the SBR.  Intraobserver and interobserver analysis The average (± standard deviation) volumes and activities of the tumour and left kidney segmented by the two observers are summarized in Table 5.3. Using a t test with a 95% confidence interval, no significant difference was found between each pair of mean values obtained by both observers. The average concordance index of all 25 pairs of initial tumour regions drawn manually (VOI0) by the observers was 0.78. When the iterative adaptive thresholding method was applied within these regions, the average concordance index of the final segmented tumour regions was 0.99. For the left kidney, these average concordance indices were 0.69 and 0.93, respectively.  93  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  Table 5.3 Interobserver comparison of tumour and left kidney volume and activity estimates.  Measurement  Observer 1  Observer 2  Tumour volume (mL)  73.3 ± 0.2  73.4 ± 0.1  Tumour activity (MBq)  16.3 ± 0.1  16.3 ± 0.2  Left kidney volume (mL)  152.4 ± 3.0  149.8 ± 2.8  Left kidney activity (MBq)  15.5 ± 0.5  15.9 ± 0.4  Data are average ± standard deviation.  5.4 Discussion Analysis of the results from the calibration experiment reconstructed using three different algorithms (Figure 5.4 and Table 5.2) clearly demonstrated that in all cases two different thresholds were necessary for accurate determination of object volume and activity. For both thresholds, the role of the reconstruction algorithm and the influence of image resolution on their values were investigated. In general, the image with the worst spatial resolution displayed the greatest difference between ThV and ThA, because the ThA value had to be much lower than ThV in order to recover activity blurred out of the source. For example, at an SBR of 2 and using the Clinical reconstruction method, ThA had to be 10.5% lower than ThV, whereas this difference was only 5.0% and 3.7% for the BBACRR and SCACRR methods, respectively. At an SBR of 10, the difference between ThV and ThA was 30.7%, 22.0% and 14.3% for the Clinical, BBACRR and SCACRR methods, respectively. Since object activity estimates were all affected by PVE, none of the SBRs recovered from the experimental data reached the value of 20, which was the maximum SBR prepared in the phantom experiments. Additionally, the measured SBR always underestimated the true SBR because the proposed algorithm estimates the source activity concentration by using the total activity within the “true” volume of the object only. Larger errors in fitting of the threshold-SBR data for the BBACRR and SCACRR methods compared to the Clinical method can be observed in Figure 5.4 and in the R2 values listed in Table 5.2. This can likely be explained by the smoother filter applied to the Clinical reconstructions (Hanning versus Butterworth). Despite the greater error in the SCACRR curve fits, this method performed slightly better in estimation of bottle volumes and activities in the validation phantom experiment compared to the Clinical method.  94  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  A direct comparison to currently used iterative adaptive thresholding techniques was not included in this chapter because with the exception of the most recent method proposed by Pacilio et al. [81], most adaptive thresholding methods aim at accurate volume determination, but not activity estimation. As discussed in Section 3.2.2, these methods are not suitable for accurate internal dose calculations. However, the improvements achieved by our method relative to currently used adaptive thresholding techniques have been demonstrated. First, the presented method applies a separate threshold for activity segmentation, which recovers activity spilled out of the object. This threshold is obtained using pre-calculated calibration curves in the same manner as is traditionally done for volume estimation. The importance of this additional threshold was substantiated by the validation phantom experiment, which showed that the activities in regions corresponding to true volumes were underestimated by up to 60%. As expected, the severity of this spill out depended on the corrections incorporated into the reconstruction algorithm. For example, the most severe spill-out (activity underestimated by 45%, on average) was observed when the method without resolution recovery or scatter correction was used, while the least spill-out (activity underestimated by 21%, on average) was obtained using the most advanced reconstruction, SCACRR. Although some sophisticated methods of PVE correction are currently actively researched, these correction methods are quite complicated and require substantial data processing. The technique presented in this chapter is intended to be practical to implement into routine clinical use, using typical clinical SPECT reconstructions. The second important feature of the proposed method in comparison to existing techniques is the semi-automatic selection of background regions. Proper determination of the background activity concentration is a key element for adaptive thresholding methods. The properly selected background should neighbour the object, but must not be affected by spill out from the object and other nearby active sources. Manual delineation of background regions can be challenging in typical clinical situations where an object can be surrounded by a nonuniform activity distribution. In the case of relatively homogeneous background activity, Hatt et al. reported no significant difference in the results obtained by two observers using an adaptive thresholding method [125]. However, in cases with heterogeneous background activity concentrations, results from adaptive thresholding have been shown to  95  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  be very user dependent [126]. The ability of the proposed method to automatically exclude regions affected by spill out and to average background activity concentrations over all surrounding tissues may substantially contribute to the consistency of results, even in the presence of heterogeneous background. Furthermore, compared to traditional methods, the time to segment an object of interest is reduced considerably by not having to manually select background regions. The iterative adaptive thresholding method introduced in this chapter requires the same level of user interaction as fixed thresholding, which involves the manual delineation of a region of interest containing the source and surrounding background. A simple method, such as this one, is an important element of an image processing technique that would be practical to use in a busy clinical environment, leading in the future to fully personalized dosimetry for patients receiving radionuclide therapy. A disadvantage of the proposed technique is that it may not be suitable in certain situations involving heterogeneous activity distributions where volumes may be underestimated when using a threshold-based approach [125]. The underestimation can be caused by hotspots that will potentially raise the threshold based on the maximum uptake to a value that is too high for lower uptake regions in the volume of interest to be included in the segmentation. In these scenarios it may be necessary to apply a more sophisticated segmentation method. Another potential limitation is that the use of a single threshold-SBR curve, which does not require any a priori knowledge of the object volume, assumes that the optimal threshold needed to recover object volume does not depend on the object size. This approach is not appropriate for delineating very small lesions. Bramibilla et al. investigated phantoms with a series of spherical inserts and found that the threshold needed to recover object volume had greater dependence on the sphere diameter than on the SBR for lesions smaller than 0.5 mL [127]. This dependence was reversed for larger spheres ranging from 1 – 5.5 mL [127]. The same observation was made by Daisne et al. who considered spheres between 2.12 and 17.15 mL in volume [77]. An additional limitation of this study is that the phantom validation was performed using the same bottles as those used to generate the calibration curve. A more robust test of the  96  Chapter 5: An Iterative Adaptive Thresholding Method for Determination of Volumes and Activities  proposed method would be to perform a validation experiment using objects of different shapes and volumes compared to those used for calibration. Despite these limitations and the additional limitations of existing approaches that have already been mentioned, adaptive thresholding methods have previously been shown to delineate regions accurately in certain situations [128] and have been reported to be quite useful under clinical conditions [129].  5.5 Conclusion The proposed semi-automatic image segmentation technique provides clinicians with quantitative estimates of both activity and volume for tumours and organs, which are needed for patient-specific dosimetry. While numerous volume quantification methods have previously been investigated, less attention has been paid to activity quantification approaches. The concept of adaptive SBR-based thresholding incorporates properties of both the imaged object (through its SBR) and data processing methodology (through camera and reconstruction specific calibration curves). The goal of this work was to keep these advantages and to modify the technique to improve its efficiency and to allow for accurate activity estimation. The proposed iterative thresholding technique is relatively simple and can be easily implemented into clinical practice. Compared to traditional adaptive thresholding techniques, the procedure presented in this work (i) allows the user to quantify not only volume, but also the activity of the investigated object, and (ii) does not necessitate the manual selection of background regions separate from the region already drawn around the object of interest. Further investigation into the ability of the iterative adaptive thresholding technique to produce repeatable and reproducible volume, activity and dose estimates in patient studies is described in Chapter 6.  97  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates Derived From SPECT Images 6.1 Introduction The previous chapter introduced an iterative adaptive thresholding method for estimating tumour/organ volumes and activities. The accuracy of this method was evaluated using a phantom experiment with bottles of known volume and activity. This evaluation was performed using different reconstruction methods to determine whether a typical clinical reconstruction can be used to obtain similar results to those obtained using a sophisticated reconstruction algorithm that is currently not practical to implement into routine clinical use. To further explore the usefulness of this adaptive thresholding method, this chapter presents repeatability and reproducibility studies performed on patient data. For the purposes of this study, repeatability was investigated by evaluating the consistency of final volume and activity estimates as obtained from two consecutive SPECT scans of the same subject. The reproducibility of volumes and activities estimated using the iterative adaptive thresholding technique was assessed in terms of its ability to obtain consistent results using different reconstruction methods. The different reconstruction methods that were evaluated were the same as the methods outlined in Table 5.1.  6.2 Methods 6.2.1  Patient studies  Thirteen patients, including the four patients introduced in Section 5.2.6, with diagnosed or suspected neuroendocrine tumours were injected with 760-1000 MBq of  99m  Tc-HYNIC-  TOC. Two different imaging protocols were used. Group 1 consisted of five patients for whom two consecutive SPECT/CT scans were acquired within 2-5 hours after injection. This  98  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates  protocol allowed for performance of a repeatability study where volume and activity estimates from two consecutive SPECT/CT acquisitions were compared. Group 2 comprised the remaining eight patients who each received one SPECT/CT. The SPECT acquisition parameters for the patient studies were identical to those used for the phantom experiments. In addition, three whole body planar scans were acquired for all patients over a period of 24 hours after injection. The iterative adaptive thresholding technique was applied to the patient SPECT data to segment tumours and organs with significant uptake, which included the kidneys, liver, spleen and thyroid. 6.2.2  Repeatability of volume and activity estimates  The repeatability assessment involved the patients in group 1 who had received two consecutive SPECT/CT scans and a series of whole body planar scans. The decrease in radioactivity due to physical and biological washout between the two SPECT time points was corrected using the effective half-life of the radiopharmaceutical in each region of interest. The effective half-life was determined with a monoexponential fit through counts obtained inside regions of interest delineated using data from the whole body planar scans. For each region, the percentage difference between estimates (xj) from the two SPECT images was found: (6.1) This analysis was done separately for each of the different reconstruction methods. 6.2.3  Reproducibility of different reconstruction methods  The reproducibility analysis included all 13 patients enrolled in the study. For each patient, the consistency in volume and activity estimates of each region was compared using SCACRR as a gold standard. The percentage difference between volume and activity estimates (xmethod) derived from these different image reconstructions was estimated using: (6.2) (6.3)  99  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates  In addition to the percentage difference between total estimated volumes, the concordance index was calculated to compare the volume segmentations from each pair of reconstruction methods. 6.2.4  Dosimetry  Finally, to demonstrate the usefulness of the presented segmentation method for clinical dose calculations, the dose from 99mTc activity for each segmented region was estimated. For each region, the hybrid planar/SPECT technique for determining the time-integrated activity coefficient was utilized (Section 3.2.3). Time-integrated activity coefficients for each patient were entered into the OLINDA/EXM 1.1 software for dose calculation, where the option to adjust the organ masses to patient-specific values was also applied [89]. The organs investigated in this dosimetry analysis were the kidneys, liver, spleen and thyroid. The sphere model included with the OLINDA/EXM software was used to estimate doses for tumours. Three dose estimates were made for each individual region, one for each reconstruction method assessed in this study.  6.3 Results 6.3.1  Repeatability of volume and activity estimates  The volumes of segmented regions from the patients in groups 1 and 2 are summarized in Table 6.1. Organs that were not entirely visible in the SPECT field of view were excluded from the study. The sample size indicates the number of patients analyzed for each organ. Table 6.1 Summary of region volumes and the number of samples segmented for each region.  Sample size Mean volume (mL) ± SD Range of volumes  Left  Right  Kidney  Kidney  Liver  Spleen  Thyroid  Tumour  9  10  7  10  4  4  155 ± 50  166 ± 50  1681 ± 497  282 ± 122  14.8 ± 2.9  76.4 ± 38.2  109 – 243  99 – 221  1244 – 2537  132 – 450  12.5 – 19  22.8 – 110.6  100  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates  Results of the repeatability test performed using consecutive scans of the five patients in group 1 are presented in Table 6.2. This repeatability analysis was performed using a total of 13 organs including the kidneys, liver and spleen. No tumours were found in any of the patients in group 1. When volume and activity estimates of each organ in a single patient were compared between the two SPECT acquisitions using Eq. (6.1), the repeatability of these measurements was similar for all reconstruction methods being investigated (Table 6.2). In all cases, the average percentage difference in volume and activity estimates between the two time points was 5% or better. Although the SCACRR method performed slightly better than the other two methods, a t test performed on these data did not show any statistically significant difference between the mean percentage differences calculated. Table 6.2 Repeatability analysis showing percentage differences in volume and activity estimates for kidneys, liver and spleen derived from two consecutive SPECT images of patients in group 1.  Reconstructions  % Volume Difference  % Activity Difference  Clinical  4.1 ± 3.5 (0.0 – 12.4)  3.6 ± 2.2 (0.4 – 7.7)  BBACRR  4.5 ± 2.6 (0.7 – 10.0)  3.8 ± 2.3 (0.8 – 7.7)  SCACRR  2.8 ± 2.2 (0.1 – 7.2)  2.8 ± 2.7 (0.0 – 7.7)  Data are average ± standard deviation, followed by range in parentheses.  6.3.2  Reproducibility of volume, activity and dose estimates  The average percentage differences between volume, activity and dose estimates obtained using the different reconstruction methods for patients in groups 1 and 2 are summarized in Table 6.3. These reproducibility results are reported separately for organs and tumours. A total of 40 organs and 4 tumours were included in this analysis. On average, organ volumes and activities estimated with the Clinical and BBACRR methods differed from the SCACRR image estimates by about 5-6%. The average percentage difference between organ doses in the Clinical versus SCACRR and the BBACRR versus SCACRR comparisons were approximately 4%. Tumour volumes, activities and doses estimated with the Clinical and BBACRR methods differed from the SCACRR estimates by 5-10%.  101  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates  Table 6.3 Reproducibility analysis showing percentage differences in volume, activity and dose estimates for organs (kidneys, liver, spleen and thyroid) and tumours derived from SPECT images of patients in groups 1 and 2 reconstructed using different methods.  Reconstructions  % Volume Difference  % Activity Difference  % Dose Difference  Clinical vs. SCACRR  5.4 ± 4.0 (0.5 – 16.1)  6.0 ± 4.3 (0.2 – 17.5)  3.6 ± 3.2 (0.0 – 10.7)  BBACRR vs. SCACRR  4.8 ± 3.3 (0.5 – 13.1)  6.3 ± 4.9 (0.1 – 17.9)  3.8 ± 2.9 (0.0 – 11.9)  Clinical vs. SCACRR  10.0 ± 6.9 (2.6 – 16.7)  5.1 ± 4.6 (1.3 – 11.7)  8.5 ± 5.5 (0.6 – 13.3)  BBACRR vs. SCACRR  4.8 ± 2.7 (1.3 – 7.7)  4.3 ± 2.4 (1.7 – 7.2)  5.3 ± 3.5 (1.8 – 8.5)  Organs (n = 40)  Tumours (n = 4)  Data are average ± standard deviation, followed by range in parentheses.  An example of the segmentation of organs and a tumour performed using the iterative adaptive thresholding technique on images reconstructed using the SCACRR and Clinical methods is presented in Figure 6.1. This figure demonstrates that similar regions were delineated in the SCACRR and Clinical images when using ThV, whereas much larger regions were required to recover the total activity in the Clinical image, which had relatively poor spatial resolution compared to the SCACRR image.  Figure 6.1 Example coronal slice from images reconstructed using SCACRR (a) and Clinical (b) methods. Segmented regions include the liver, left and right kidneys, spleen and a tumour, which are each delineated using ThV (solid magenta line) and ThA (white dashed line).  The average concordance index of volumes segmented using the Clinical and SCACRR methods was 0.88 ± 0.04. Comparing the BBACRR and SCACRR methods, the average concordance index was 0.93 ± 0.03. 102  Chapter 6: Repeatability and Reproducibility of Volume, Activity and Dose Estimates  6.4 Discussion The robustness of the iterative adaptive thresholding method proposed in Chapter 5 has been demonstrated in Chapters 5 and 6 by its ability to provide reproducible results in a wide variety of situations. The situations in which this reproducibility has been demonstrated include: (1) the use of different starting estimates of the SBR; (2) an intraobserver and interobserver analysis, which tested the reproducibility of results when different initial volumes (VOI0) were used; (3) a test of the consistency of volumes and activities estimated from two consecutive SPECT/CT scans of the same patient; (4) images reconstructed with various corrections incorporated into the reconstruction algorithm. The average percentage difference in organ dose (Table 6.3) was slightly lower than the percentage difference in volume and activity estimates because in most cases where there was a large overestimation in the organ activity (>10%), the volume would also be overestimated relative to the SCACRR reconstruction. Similarly, when organ activity was underestimated, volume tended to be underestimated as well. When performing dosimetry calculations where both activity and volume are under or overestimated at the same time, they compensate for each other, such that the percentage difference in dose is roughly the difference between percentage differences in volume and activity. This observation was not made for tumour doses, although this could be due to the small sample size of tumours analyzed.  6.5 Conclusion The iterative adaptive thresholding technique presented in Chapter 5 has been found to accurately determine volumes and activities of objects in phantom studies. Furthermore, volume and activity estimates were reproducibly estimated in both phantom and patient studies. The reproducibility of this technique indicates that it could be successfully used by clinicians to perform dose calculations from images reconstructed with clinical software and provide similar results as those derived from more sophisticated reconstruction methods.  103  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  Chapter 7: Comparison of Dose Estimates Obtained Using Organ Level, Voxel S Value and Monte Carlo Techniques 7.1 Introduction Three dose estimation methods have been presented, namely the organ level dose calculations with OLINDA/EXM, the use of voxel S values and Monte Carlo simulation. These methods vary in complexity and in the accuracy of the dose estimates that they produce. OLINDA/EXM dose calculations are performed very quickly by using precalculated organ level S values based on reference phantoms used to represent the standard patient. Monte Carlo calculations are the most rigorous, but in turn are very computer intensive. In this chapter, the doses calculated using these three methods are compared. These comparisons were performed in order to further investigate the advantages and disadvantages of each method, to study the accuracy of organ level dose calculations based on reference phantoms and to investigate which dose estimation method is appropriate in any particular situation. These goals ultimately fit within the objective of this thesis, which is to investigate a dose estimation approach that can be used to efficiently perform accurate patient-specific dose calculations in the clinical environment. Throughout this chapter, several aspects of OLINDA/EXM dose calculations are compared to the results from Monte Carlo simulation, which is considered as a gold standard for the purposes of this work. The study includes a comparison of patient-specific S values calculated by Monte Carlo to the S values used by the OLINDA/EXM software, mean organ doses calculated by the two dose estimation methods, and an assessment of the sphere model used to estimate dose to tumours by OLINDA/EXM. Finally, a comparison of 3D dose distributions calculated using Monte Carlo and voxel S values was performed. This chapter  104  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  also highlights the use of the GUI presented in Chapter 4 not only for patient-specific dose calculations, but also as a useful tool for research investigations.  7.2 Methods 7.2.1  Patient studies  A total of 6 patients (3 males and 3 females) injected with 800-1000 MBq of  99m  Tc-  HYNIC-TOC were included in this study. For each patient, a series of 3-4 whole body planar scans were acquired over a period of 24 hours following injection. In addition, a single SPECT/CT scan was acquired approximately 3 hours after injection. SPECT reconstruction was performed using the SCACRR method described in Section 5.2.2. 7.2.2  Calculation of TIACs  A hybrid planar/SPECT approach was used to plot and integrate TACs for all tumours and each organ with significant uptake, which were the kidneys, liver and spleen. This approach utilized all sub-GUIs from the internal dosimetry toolkit. The Planar region selection and Planar image registration sub-GUIs were used to draw and register ROIs in the series of planar images. The sub-GUI for processing 3D image data was used in the analysis of the SPECT images to segment each ROI to find the 3D activity distribution as well as the total volume and absolute activity for each region. Back in the main GUI, a monoexponential fit was applied to the planar data, which was then scaled to pass through the activity determined from the SPECT image (  ) at the time of the SPECT acquisition (  ).  The TACs obtained from these exponential fits were integrated to find the cumulated activity of the 99mTc in each source region: (7.1) where  is the effective elimination constant obtained from the exponential fit.  The  99m  Tc TIACs needed for organ level dose calculation were also determined in the  main GUI by dividing the cumulated activities by the injected activity  . The  replacement radionuclide feature in the main GUI was used to find predicted TIACs of and  177  131  I  Lu assuming a pharmaceutical labelled with these radionuclides would follow the  same biodistribution as the 99mTc labelled tracer. 105  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  7.2.3  Absorbed dose calculation  Mean organ doses to each segmented organ from the organ doses from  131  I and  177  99m  Tc injection and the predicted  Lu injection were calculated using the Organ level dose  calculation GUI. The mean dose calculated for each organ included contributions from the self-dose as well as the cross-dose from other segmented regions. The OLINDA/EXM sphere model was used to calculate the dose to tumours. The voxel S value method and Monte Carlo simulation were used to calculate 3D dose rate distributions for tumours and organs using the Process 3D image data GUI. For voxel S value dose calculation, the voxel S values were precalculated using the EGSnrc user-code DOSXYZnrc. The relevant parameter settings for creation of the voxel S values were 109 histories originating from a single source voxel with dimensions of 4.42 mm on edge at the center of a 215x215x215 voxel grid inside a soft tissue equivalent medium. The size of this voxel array was large enough so that cross-dose could be calculated between all ROIs using the voxel S value method. For Monte Carlo dose calculation, the total number of histories simulated per source voxel was 100,000 photons and 100,000 electrons sampled from each radionuclide’s emission spectrum. The total number of voxels simulated from the liver, kidneys and spleen of each patient was on the order of 104 voxels, so that an approximate total of 109 histories were simulated for each patient. The EGSnrc input parameter, ECUT, which is the electron cutoff energy (sum of rest mass and kinetic energy) for which a history is terminated and energy is deposited in the current voxel, was set at 0.561 MeV for the  99m  Tc simulation,  0.591 MeV for the 177Lu simulation and 0.611 MeV for the 131I simulation. Selection of these ECUT values was based on a separate study (not discussed in this thesis) that investigated the choice of Monte Carlo input parameters that reduced the simulation time without a significant loss of accuracy in the resulting dose distribution.1 The 3D dose rate distributions calculated by Monte Carlo simulation and the voxel S value technique were multiplied by the factor  for conversion to 3D  1  This study compared dose distributions resulting from a single source voxel of activity in a uniform tissue medium, simulated using different Monte Carlo input parameters. Different values of ECUT were compared to the use of an ECUT value of 0.521 MeV. When corresponding voxels in different dose distributions were compared, the chosen ECUT values (0.561 MeV, 0.591 MeV and 0.611 MeV for 99mTc, 177Lu and 131I, respectively) were found to produce dose distributions with source voxel doses that agreed within 1% and neighbouring voxels that agreed within 5% of the doses calculated using an ECUT value of 0.521 MeV. 106  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  dose distributions. Similar to the organ level calculations, these dose distributions included self-dose as well as the cross-doses from all organs with significant uptake. Thus, it was important to only include patients in this study if the kidneys, liver and spleen were all visible in a single SPECT field of view. For each source organ, a separate dose distribution in the volume enclosing all target regions was calculated so that a set of S values could be determined. These S values were found by dividing the mean dose to each target region the cumulated activity in source region  by  : (7.2)  For example, the cross organ S value for activity in the spleen irradiating the liver, , was calculated as the mean dose to the liver, calculated by Monte Carlo (or voxel S values) using the spleen as a source organ, divided by the cumulated activity in the spleen. The total dose to each target was also determined by summing the dose distributions calculated for each source organ. After the Monte Carlo and voxel S value calculations were performed for the  99m  Tc activity distribution, these calculations were repeated using sources  of 131I and 177Lu. 7.2.4 The  Evaluation of dose estimation methods 99m  Tc,  131  I and  177  Lu dose estimates obtained using each of the three methods were  evaluated by (i) comparing the organ level S values corresponding to each method, (ii) comparing the total tumour and organ doses calculated by each method, (iii) investigating the difference in right and left kidney doses from the Monte Carlo simulation, and (iv) comparing the Monte Carlo and voxel S value dose distributions voxel-by-voxel. In each assessment, the Monte Carlo dose distributions were assumed to be the gold standard that the two other methods were compared to. Comparison of reference and patient-specific S values at the organ level The values of  based on the reference phantoms used by OLINDA/EXM were  compared to the S values calculated by Monte Carlo simulation using patient-specific CT images. This assessment was done by finding the percentage difference between the two S values for each source and target region pair. The OLINDA/EXM S values were mass  107  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  corrected using patient-specific organ masses as described in Section 3.3.1. The percentage difference  for each S value array element was calculated as follows: (7.3)  This yielded a 3x3 array of percentage differences for each patient, which were then used to find the average, standard deviation, minimum and maximum percentage difference for each array element over the entire patient population. Similarly, the organ level S values calculated from the voxel S value method were compared to the Monte Carlo determined S values to find: (7.4)  Total dose assessment The mean organ and tumour doses calculated using Monte Carlo were compared to the organ and tumour doses calculated using OLINDA/EXM and voxel S values by finding the percentage difference between doses estimated by each of these methods. In all cases the total doses included contributions from self and cross organ irradiation, except for the tumour doses calculated by OLINDA/EXM, where its sphere model was used to calculate self doses only. The Monte Carlo results were analyzed further to determine the percentage of total dose that was due to cross organ irradiation. In addition, the assumption by OLINDA/EXM that electrons are fully absorbed by source organs was tested by assessing the percentage of the total dose that was due to cross organ irradiation from electron emissions. Paired organ analysis The Monte Carlo dose calculation was used to assess the dose to the right and left kidney separately in order to investigate the accuracy of the assumption made by OLINDA/EXM that paired organs each receive equal dose. The percentage difference between right and left kidney doses  was found using:  .  (7.5)  108  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  Monte Carlo and voxel S value voxel-by-voxel comparison The 3D dose distributions calculated by Monte Carlo simulation and the voxel S value method were compared visually by plotting cumulative dose volume histograms (DVHs). In addition, a voxel-by-voxel analysis was carried out to determine the average (± standard deviation) percentage difference between corresponding voxels in the two 3D dose distributions calculated for each patient.  7.3 Results Time-integrated activity concentrations of  99m  Tc for the investigated organs and tumours  are summarized in Table 7.1. Table 7.1 Time-integrated activity concentrations for investigated regions in patients 1 to 6.  Time-integrated activity concentrations (MBq∙h/mL) Region  1  2  3  4  5  6  Left kidney  0.62  1.33  0.86  0.63  1.26  1.02  Right kidney  1.07  1.34  0.91  0.63  1.17  0.86  Liver  0.37  0.41  0.55  0.42  0.53  0.32  Spleen  1.58  1.08  2.42  1.21  2.12  1.39  Tumour in lung  -  0.71  -  -  -  -  Tumour in pancreas  -  -  1.40  -  -  -  Tumour in small bowel  -  -  -  2.84  -  -  7.3.1  Comparison of organ level S values  When the S values used by OLINDA/EXM for each source and target region pair and the corresponding S values calculated using Monte Carlo simulation were compared using Eq. (7.3), there was generally a good agreement in S values for self-irradiation, but a very poor agreement in S values for cross-irradiation (Table 7.2, Table 7.3, and Table 7.4 for 99mTc, 131I and 177Lu, respectively). The S values for self-irradiation all agreed within 2.3%, on average, regardless of the radionuclide used. Comparison of the Monte Carlo generated cross organ S values with those used by OLINDA/EXM gave percentage differences that ranged between -83% for in patient 4 (with  in patient 3 (with  177  Lu) and 105% for  99m  Tc).  109  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  In the comparison of S values calculated by the voxel S value method and Monte Carlo simulation, there was again good agreement in the S values for self-irradiation (Table 7.2, Table 7.3, and Table 7.4). Unlike the Monte Carlo comparison to OLINDA/EXM S values, there was reasonable agreement with S values calculated using the voxel S value method for cross-irradiation between the kidneys and liver and between the kidneys and spleen, where the average  was found to be 11% or less. There was not good agreement  between the liver and spleen where the average values of  and  were found to range between 14% with 131I and 31% with 99mTc. Table 7.2 Summary of percentage differences between the  99m  Tc patient-specific S values calculated using  Monte Carlo and the corresponding S values used by OLINDA/EXM and those calculated using voxel S values for each source and target region pair.  Source Target  Kidneys  Liver  Spleen  0.2 ± 1.3 (-1.9 – 2.1)  -21 ± 24 (-51 – 10)  25 ± 49 (-38 – 105)  0.5 ± 1.1 (-1.5 – 1.8)  11 ± 5 (4.7 – 17)  10 ± 5 (4.6 – 17)  -23 ± 23 (-43 – 14)  2.3 ± 2.2 (-0.4 – 4.7)  -31 ± 27 (-72 – 10)  11 ± 5 (3.3 – 16)  3.8 ± 1.3 (1.6 – 4.9)  28 ± 11 (9.5 – 37)  -0.9 ± 34 (-47 – 38)  -45 ± 23 (-68 – -7)  1.1 ± 2.1 (-2.5 – 3.8)  11 ± 5 (4.6 – 17)  31 ± 13 (11 – 43)  0.7 ± 1.8 (-2.2 – 2.5)  Kidneys  Liver  Spleen  Data are mean ± standard deviation; the range is provided in parentheses.  110  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  Table 7.3 Summary of percentage differences between the 131I patient-specific S values calculated using Monte Carlo and the corresponding S values used by OLINDA/EXM and those calculated using voxel S values for each source and target region pair.  Source Target  Kidneys  Liver  Spleen  -1.3 ± 1.3 (-3.7 – -0.3)  -25 ± 26 (-57 – 12)  19 ± 50 (-46 – 97)  -0.2 ± 1.5 (-2.9 – 1.4)  5.4 ± 2.3 (1.9 – 7.7)  3.4 ± 3.1 (-0.2 – 7.1)  -30 ± 28 (-55 – 12)  -0.3 ± 1.6 (-2.3 – 1.3)  -32 ± 25 (-72 – 3.2)  4.6 ± 2.1 (1.8 – 7.4)  1.6 ± 1.3 (-0.3 – 2.5)  14 ± 5 (4.6 – 18)  1.0 ± 40 (-58 – 37)  -47 ± 20 (-70 – -15)  -1.8 ± 2.2 (-5.6 – 0.3)  4.7 ± 2.1 (1.7 – 7.8)  15 ± 6 (4.3 – 19)  -0.6 ± 2.3 (-4.5 – 1.8)  Kidneys  Liver  Spleen  Data are mean ± standard deviation; the range is provided in parentheses. Table 7.4 Summary of percentage differences between the  177  Lu patient-specific S values calculated using  Monte Carlo and the corresponding S values used by OLINDA/EXM and those calculated using voxel S values for each source and target region pair.  Source Target  Kidneys  Liver  Spleen  -1.3 ± 1.4 (-3.9 – -0.2)  -30 ± 31 (-76 – 9.5)  16 ± 54 (-55 – 100)  -0.3 ± 1.6 (-3.2 – 1.5)  9.0 ± 4.2 (3.8 – 14)  7.1 ± 5.8 (0.5 – 14)  -40 ± 35 (-83 – 10)  0.1 ± 1.7 (-2.2 – 1.6)  -33 ± 27 (-73 – 7.1)  6.8 ± 4.6 (2.5 – 12)  1.3 ± 1.6 (-1.0 – 2.5)  24 ± 10 (8.7 – 32)  -1.9 ± 42 (-62 – 35)  -46 ± 22 (-70 – -11)  -2.1 ± 2.4 (-6.1 – -0.2)  9.0 ± 3.9 (3.4 – 14)  28 ± 11 (9.7 – 38)  -0.8 ± 2.7 (-5.3 – 2.0)  Kidneys  Liver  Spleen  Data are mean ± standard deviation; the range is provided in parentheses.  Differences in patient specific-anatomy leading to the large discrepancies in OLINDA/EXM and Monte Carlo generated S values can be explained by visualizing the 111  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  anatomy of two patients that are displayed in Figure 7.1. The corresponding S value comparisons are listed in Table 7.5, Table 7.6 and Table 7.7 for  99m  Tc,  131  177  I and  Lu,  respectively.  Figure 7.1 Coronal view of maximum intensity projections for patients 2 (a) and patient 4 (b) illustrating differences in patient-specific anatomy and relative uptakes of 99mTc-HYNIC-TOC in the kidneys, liver and spleen. Table 7.5 Percentage differences between 99mTc patient-specific S values calculated by Monte Carlo simulation and the corresponding reference S values used by OLINDA/EXM and those calculated using voxel S values for patients 2 and 4. 99m  Tc  Patient 2  Patient 4  Kidneys  Liver  Spleen  Kidneys  Liver  Spleen  -0.04  -41.9  -38.1  -0.3  -0.9  105  1.8  16.9  7.0  0.3  13.6  16.8  -43.0  4.7  -71.7  14.0  1.0  10.3  15.6  4.8  29.2  12.5  3.1  37.0  -32.5  -67.0  0.6  37.5  -35.3  2.3  11.7  34.5  -0.01  17.5  40.9  2.3  Kidneys  Liver  Spleen  112  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  Table 7.6 Percentage differences between  131  I patient-specific S values calculated by Monte Carlo simulation  and the corresponding reference S values used by OLINDA/EXM and those calculated using voxel S values for patients 2 and 4. 131  I  Patient 2  Patient 4  Kidneys  Liver  Spleen  Kidneys  Liver  Spleen  -0.6  -46.1  -46.1  -1.4  -3.0  97.2  1.4  7.7  -0.2  -0.4  6.2  7.1  -48.1  0.8  -71.7  11.5  -2.2  3.2  5.1  2.3  14.2  6.1  0.06  17.5  -37.8  -68.0  -2.4  33.8  -40.6  -0.3  4.8  15.5  -1.5  7.8  19.0  0.9  Kidneys  Liver  Spleen  Table 7.7 Percentage differences between 177Lu patient-specific S values calculated by Monte Carlo simulation and the corresponding reference S values used by OLINDA/EXM and those calculated using voxel S values for patients 2 and 4. 177  Lu  Patient 2  Patient 4  Kidneys  Liver  Spleen  Kidneys  Liver  Spleen  -0.3  -47.5  -55.0  -1.3  -3.3  99.8  1.5  14.5  0.5  -0.6  11.3  14.0  -53.9  0.9  -73.0  10.2  -1.8  7.1  12.0  2.0  25.5  10.1  -0.5  32.5  -44.9  -68.5  -2.9  35.0  -37.7  -0.2  10.1  30.4  -1.8  14.3  35.8  0.9  Kidneys  Liver  Spleen  7.3.2  Total organ and tumour dose assessment  There was good agreement between the total organ doses calculated by OLINDA/EXM and the mean doses calculated by Monte Carlo simulation (Figure 7.2), regardless of the  113  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  radionuclide used. The worst agreement was found in the spleen of patient 1 (using  177  Lu),  where the percentage difference between OLINDA/EXM and Monte Carlo doses was -6.2%. Similarly, total doses calculated using the voxel S value method were in good agreement with the Monte Carlo results. The worst agreement was found in the liver of patient 6 (using 99m  Tc), where the percentage difference between the voxel S value and Monte Carlo doses  was 7.4%. For tumour dose calculation, the average percentage differences between the Monte Carlo calculation and the sphere model used by OLINDA/EXM (voxel S value method) were -8.8 ± 11.1% (1.1 ± 1.9%), -6.0 ± 5.4% (-1.1 ± 4.0%), and -3.8 ± 5.2% (-1.5 ± 4.6%) for 131  I and  177  99m  Tc,  Lu, respectively.  114  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  Figure 7.2 Total organ and tumour doses calculated by OLINDA/EXM and the voxel S value (VSV) method compared to mean doses from Monte Carlo simulation for  99m  Tc (a) and (b),  131  I (c) and (d), and  177  Lu (e) and  (f). One data point not visible in part (a) is the OLINDA/EXM underestimation of 22% in the tumour dose of patient 2. Similarly in part (b), tumour dose in patient 2 was underestimated by 12% and is not displayed.  Analysis of the cross organ irradiation data from Monte Carlo simulation revealed that cross organ doses made the greatest contribution to total dose calculated for distributions and the least contribution for cross organ irradiation with  177  99m  Tc  Lu distributions (Figure 7.3). For example,  99m  Tc contributed as much as 18% to the total organ dose (in the  kidneys of patient 3), whereas the largest contribution to total organ dose from  177  Lu cross  organ irradiation was 1.8% (also in the kidneys of patient 3). Furthermore, electron emissions 115  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  were found to contribute less than 0.05% to cross organ doses in most cases. The largest beta contribution to cross dose was in the range 0.4-0.6%, which was observed for  131  I in  situations where two organs were in contact (e.g. the left kidney and the spleen or the right kidney and the liver). For tumours, cross organ irradiation contributed as little as 0.2% (with 177Lu in patient 4) and as much as 16.2% (with 99mTc in patient 2) to the total tumour dose.  Figure 7.3 Average percent contributions of self and cross doses to the total organ and tumour doses for simulation with 99mTc (a), 131I (b), and 177Lu (c). Note that the x-axis scale begins at 50%.  7.3.3  Paired organs  Results from the comparison of right and left kidney doses, calculated by Monte Carlo simulation, are reported in Table 7.8. For  99m  Tc, the doses deposited in the right and left  kidneys agreed within 7.3% in four out of six patients. Two extreme cases where there was a relatively large difference between individual kidney doses were observed for patients 1 and 6 (Figure 7.4). There was generally a poor agreement between kidney doses calculated with 177  131  I and  Lu activities. This was due to a combination of the long physical half-lives of these  radionuclides and differences in the washout rates between the two kidneys (demonstrated by  116  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  the differences in biological half-lives listed in Table 7.8 for the right and left kidneys). The average percentage differences between individual kidney doses were 5.9 ± 21.1%, 15.1 ± 39.7%, and 14.7 ± 38.4% for simulation with 99mTc, 131I and 177Lu, respectively. Table 7.8 Percentage difference between right and left kidney doses  calculated by Monte Carlo for three  radionuclides with corresponding biological half-lives in each patient.  (h) Patient  99m  131  Tc  177  I  Lu  R. Kidney  L. Kidney  1  44.6  37.2  37.8  30.8  37.0  2  2.3  10.1  10.4  38.0  33.3  3  7.3  13.7  14.2  31.2  28.4  4  2.6  3.5  3.2  36.0  35.0  5  -2.1  73.0  68.9  194.7  47.9  6  -19.3  -47.2  -46.5  35.0  56.3  5.9 ± 21.1  15.1 ± 39.7  14.7 ± 38.4  61.0 ± 65.6  39.7 ± 10.4  Mean ± SD  Figure 7.4 Coronal maximum intensity projections of patients 1 (a) and patient 6 (b) who had the largest differences between right and left kidney doses with 99mTc.  7.3.4  Monte Carlo and voxel S value comparison  The dose distributions calculated with the voxel S value method closely matched the dose distributions obtained with Monte Carlo simulation. To illustrate the similarity between the calculated dose distributions, a comparison of the cumulative DVHs determined by the two methods is presented in Figure 7.5. On average, the D90’s agreed within 2.8 ± 2.9%, -1.4 ± 2.1%, and -2.3 ± 2.4% for  99m  Tc,  131  I and  177  Lu, respectively. When the dose distributions  117  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  were analyzed voxel-by-voxel, the largest percentage difference between corresponding voxels averaged over all voxels in the target organs of a single patient was 6.1 ± 6.6% in patient 6 with 99mTc.  Figure 7.5 Cumulative dose volume histograms based on 3D dose distributions calculated by Monte Carlo (dashed lines) and voxel S values (solid lines) for organs and a tumour analyzed in patient 3, using 131I. For each region, the dose volume histograms determined using each dose estimation method are nearly overlapping, demonstrating the similarity between the Monte Carlo and voxel S value dose distributions.  7.4 Discussion Analysis of the results presented in this chapter clearly indicates that patient anatomy had a large impact on cross organ S values. For example, the value of was -38% for patient 2 and 105% for patient 4 when using  99m  Tc (Table 7.5). As  illustrated in Figure 7.1, the spleen is found closer to the left kidney in patient 2 then it is in patient 4 leading to the relative differences observed in this S value comparison. Similar results were reported by Divoli et al., who compared 131I S values used by OLINDA/EXM to patient-specific values calculated by the MCNPX2.5.0 Monte Carlo code [130]. They reported percentage differences between OLINDA/EXM and patient-specific S values ranging from -51% to 84%.  118  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  The cross organ S values calculated using voxel S values and Monte Carlo simulation were in good agreement between the liver and kidneys and between the spleen and kidneys since the volume enclosing each of these pairs of regions is a relatively uniform tissue medium. However, there was poor agreement in the cross organ S values between the liver and spleen, where the voxel S value technique overestimated the Monte Carlo calculation by about 30% on average for  99m  Tc and  177  Lu, and by about 15% on average for  131  I. This  overestimation was caused by radiation partially absorbed in the spine lying partly between the spleen and liver, which was properly handled by Monte Carlo simulation, but was not accounted for by the voxel S value calculation where a uniform soft tissue medium was assumed. The overestimation was not as severe for  131  I due to the higher energy gammas  (364 keV) it emits compared to the other two radionuclides. Despite the disagreements in cross organ S values, the total mean organ doses calculated by each technique remained in reasonable agreement (Figure 7.2). This was due to the fact that the self-organ S values were in good agreement (Table 7.2) and the self-dose accounted for the majority of the dose to each organ containing activity (Figure 7.3). The errors in organ cross-dose calculated by OLINDA/EXM may become more relevant in organs with no specific uptake of the radiopharmaceutical; however, this was not investigated as a part of this study. Notably, in the comparison of Monte Carlo and OLINDA/EXM, there was greater variance in the percentage differences calculated for  99m  Tc than with the other two  radionuclides (Figure 7.2). This resulted from the large errors in cross organ S values (Table 7.2) and the fact that cross organ doses were most significant with 99mTc activity distributions (Figure 7.3). Cross organ irradiation contributed more to total organ doses with compared to  131  I and  99m  Tc  177  Lu because the latter two radionuclides are beta emitters, and beta  particles deposit the majority of their energy locally. Furthermore, results of this study confirmed that bremsstrahlung contributions to cross organ doses are negligible. Cross organ contributions were greater with keV gamma emitted by gammas emitted by  177  131  131  I than with  177  Lu, again because of the higher energy 364  I (82% abundance), compared to the 113 keV and 208 keV  Lu with 6.4% and 11% abundances, respectively. In general, lower  energy photons have a higher probability of depositing energy in target regions that are in  119  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  close proximity to the source, while higher energy photons have a higher probability of reaching targets further away from the source. The small cross-dose contributions observed with  131  I and  177  Lu led to good agreement  between tumour doses calculated by Monte Carlo and with the OLINDA/EXM sphere model. With  99m  Tc, the inability to account for cross-dose using the sphere model resulted in  underestimation of tumour dose by 9 ± 11%, on average. In the study by Divoli et al., the good agreement between the sphere model and the results from Monte Carlo were attributed to the small sizes of the tumours (7 g) studied . However, in the current study, tumours up to 95 g were investigated. The biggest discrepancy between the sphere model and Monte Carlo occurred in a patient with a tumour in the lung where there was a greater chance of receiving cross-irradiation due to the lower attenuation coefficient of the surrounding lung tissue. In the paired organ analysis, it is interesting to note the influence of the different physical half-lives of the investigated radionuclides. For the shorter lived radionuclide,  99m  Tc,  differences in right and left kidney doses were caused by differences in the total activity uptake (Figure 7.4a) and proximity to hot organs (Figure 7.4b), whereas for the longer lived radionuclides,  131  I and  177  Lu, differences in left and right kidney doses were also heavily  influenced by differences in the rates of washout from these organs. For example, in patient 2, the right and left kidney doses agreed within 2.3% for 99mTc, but a difference of 4.7 hours in the biological half-lives of each kidney raised the percent difference to 10% for 177  131  I and  Lu. Finally, 3D dose distributions calculated by the voxel S value method in approximately  one hour were found to produce very similar dose distributions to those calculated by Monte Carlo simulation, which took 30 hours or more. The use of the fast Hartley transform could be used to significantly decrease the computation time of the voxel S value calculation [131]. The DVHs presented in Figure 7.5 correspond to dose calculations performed in the abdominal region where tissue heterogeneities are minimal. Larger differences between voxel S value and Monte Carlo dose calculations would be expected in the thoracic region.  7.5 Conclusion Several aspects of OLINDA/EXM dose calculations were tested by comparing them with Monte Carlo dose estimates. Although anatomy can differ significantly between patients,  120  Chapter 7: Comparison of Dose Estimates Obtained Using Different Dose Estimation Techniques  organ doses calculated by OLINDA/EXM were found to be in good agreement with Monte Carlo mean dose estimates. Furthermore, the sphere model used by OLINDA/EXM agreed reasonably well with Monte Carlo dose estimates with therapeutic agents (131I and  177  Lu)  even with cross organ irradiation. The treatment of paired organs by OLINDA/EXM was found to be inaccurate in several cases where the dose to the right and left kidneys differed significantly. This is an important issue, especially in treatments where the kidneys are the dose limiting organ, such as in peptide receptor radionuclide therapy (PRRT). Furthermore, the lack of voxelized dose information is a major limitation of organ level calculations since nonuniform dose distributions may have important clinical implications. Although Monte Carlo simulation may not currently be feasible for routine patient dose calculations, voxel S values have been shown to produce nearly equivalent 3D dose distributions.  121  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNICTyr3-Octreotide in Neuroendocrine Tumours 8.1 Introduction In this chapter, the methods that have been described in Chapters 4-7 are employed to perform a 3D imaging based dosimetry and biodistribution analysis of commercially available  99m  Tc-HYNIC-TOC (99mTc-Tektrotyd, IAE POLATOM, Poland), in a group of  patients with diagnosed or suspected neuroendocrine tumours (NETs). This analysis was performed using the hybrid planar/SPECT technique introduced in Sections 3.2.1 and 3.2.3. Although  99m  Tc-HYNIC-TOC has been increasingly gaining acceptance as a new  radiopharmaceutical for the diagnosis of pathological lesions overexpressing somatostatin receptors (SSTRs), little information has been published regarding the radiation dosimetry of this agent. The work presented in this chapter was published in The Journal of Nuclear Medicine in 2011 as the first paper reporting a 3D dosimetry analysis of this radiopharmaceutical [132].  8.2 Background to Neuroendocrine Tumour Imaging NETs represent a group of human tumours known to express SSTRs to a varying degree [133]. Accordingly, SSTR imaging with a radiolabelled somatostatin analog is an important diagnostic procedure that is routinely used for localization of primary and metastatic sites of disease, staging, qualification of patients for treatment with cold or radiolabelled somatostatin analog and therapy follow-up. Until recently, the radiopharmaceutical of choice for NET imaging was  111  drawbacks to imaging with  In-DTPA-octreotide (Octreoscan) [134]. However, there are  111  In given its high cost and emission of medium energy photons,  which result in suboptimal image quality and elevated radiation dose to the patient and staff. As an alternative to  111  In,  99m  Tc labelled somatostatin analog,  99m  Tc-HYNIC-TOC, has  been utilized by a number of groups for early diagnosis and staging of tumours expressing SSTRs [135-137] and has been demonstrated to provide improved NET lesion detection 122  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  compared to  111  In-DTPA-octreotide [138]. PET tracers like 68Ga labelled DOTA-derivatives  of somatostatin have also become very popular in recent years [139-141]. However, application of these tracers is limited to PET centers, while SPECT still remains the most prevalent nuclear medicine imaging modality worldwide. Given the promising applications of  99m  Tc-HYNIC-TOC, there are situations where  accurate activity quantification and dose estimation are desirable. In particular, accurate assessment of tumour and organ uptake is indicative of the intensity of tumour SSTR expression and potential organ toxicities, which can influence treatment planning decisions. However, very little has been published regarding the radiation dosimetry of  99m  Tc-HYNIC-  TOC. Prior to the work presented in this chapter, the only paper addressing this issue was published by González-Vázquez et al., who employed a 2D dosimetry protocol based on a series of conjugate view whole body scans for dose estimation [142].  8.3 Methods 8.3.1  Patient studies  Twenty-eight patients (14 males, 14 females) with diagnosed or suspected NETs and a median age of 54 (range 23-78) were included in this study. All scans were acquired using a dual head Infinia Hawkeye 4 camera (GE Healthcare). For each patient, multiple (3-4) whole body planar scans and one SPECT/CT scan were obtained over a period of 1-24 hours following injection of 750-1020 MBq of  99m  Tc-Tektrotyd. The whole body anterior and  posterior views were acquired into 256x1024 matrices with a pixel size of 2.21 mm and a scan speed of 20 cm/min. For quality control, in all planar studies a small  99m  Tc point source  was placed beside the patient within the field of view of the camera. For the SPECT scans, 60 projections were acquired over 360° with a non-circular orbit. Each SPECT scan used 128x128 projection matrices with a pixel size of 4.42 mm and 20 seconds per stop. The low dose CT (Hawkeye) was used to create attenuation maps. Additionally, for nineteen patients, a 30 minute dynamic planar scan was obtained in order to investigate the temporal characteristics of the early uptake phase. Each dynamic scan began immediately following the injection and recorded 60 frames for 30 seconds per frame into 128x128 matrices for anterior and posterior views of the suspected tumour location.  123  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  In all patients, this  99m  Tc-Tektrotyd study was performed as a part of the diagnostic  routine for primary and metastatic lesion localization, staging or follow-up. The final diagnosis of these patients was based on histopathological examination. 8.3.2  Determination of the time-activity curves and effective half-lives  The first step in data processing for the hybrid planar/SPECT procedure is determination of the shape of the TACs for tumours and normal organs from the whole body images (Section 3.2.3). To achieve this, the first image from the series of whole body scans of each patient was selected and regions were manually drawn around every tumour and organ with significant uptake, which included the kidneys, liver, spleen and to a lesser extent, the thyroid. Each region contained the tumour or organ and the surrounding background while care was taken to avoid other nearby areas with high uptake. In patients with liver metastases, regions were only drawn around the whole liver and the individual metastases were not included in this analysis. ROIs were automatically created by applying a threshold of 50% of the maximum pixel counts. These 2D ROIs were then registered using the Planar image registration GUI (Section 4.3.3). In addition, regions were manually drawn around the whole body. For each ROI segmented from the planar data, adjacent background regions were drawn in order to subtract background activity [67] using the correction factor, : (8.1) Where  and  represent the counts in the anterior and posterior views of the source ROI,  is the mean counts per pixel in the background region multiplied by the number of pixels in the source ROI, and  is the fraction of the patient thickness corresponding to  background tissue at the location of the ROI. Patient and source region thicknesses were estimated from the SPECT image. For the majority of cases, the product of the background correction factor and the geometric mean of counts,  , was plotted versus time. In  some instances, tumours were found near the surface and only appeared in one planar view, in which case only counts from that single view were used. No attenuation correction was performed for the planar studies since the attenuation correction factor remains constant at each time point and these data were only used for determining the shape of the TACs. For each source region, a monoexponential fit through the planar data was used to find the  124  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  effective elimination constant  . Using the quantitative SPECT images, these  monoexponential fits were subsequently normalized to pass through the data point corresponding to the reconstructed SPECT activity ( acquisition (  ) at the time of the SPECT  ). The resulting activity as a function of time for each source region, could  then be expressed as: (8.2) Additionally, to explore the error in time-integrated activity estimates associated with this monoexponential fit to the data, biexponential functions were fit to the spleen for 20 patients, where uptake was observed to be slower than in the other organs. The fitted values for the decay constants were used to determine the effective half-life in each source ROI using Eq. (3.6). To verify the procedure, ROIs were also drawn around the 99m  Tc markers placed outside the patient. The calculated half-life obtained from the  exponential fit through the geometric mean of the counts in these ROIs plotted versus time was investigated to confirm that it matched the physical half-life of 99mTc (6.01 h). 8.3.3  Activity quantification and dose estimation  All SPECT reconstructions were performed using the SCACRR method (Section 5.2.2), which includes resolution recovery, CT-based attenuation and analytical scatter corrections. A calibration scan, using a planar point source in air, was performed to determine the sensitivity of the camera (in cpm/kBq) so that counts in the reconstructed images could be translated into absolute activity. Volumes and absolute activities of tumours and other organs were obtained from the reconstructed SPECT images using an adaptive thresholding technique. Two separate adaptive thresholds, ThV and ThA (Section 5.2.3) were used. Regions segmented by use of ThV provided volume estimates and thus patient-specific organ masses. Tumour/organ absolute activities that were determined from the SPECT images using ThA provided the normalization factors based on which the TACs for each patient were rescaled. Source region TIACs were then computed by integrating the TACs and dividing the obtained values by the injected activities. In addition, the TIAC for the urinary bladder contents was calculated using the voiding bladder model. Inputs for the voiding bladder model, which include the fractions and biologic half-lives for activity excreted in the urine were obtained from the  125  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  whole body clearance curves. The TIAC for the remainder of the body was determined by subtraction of the TIACs calculated for all segmented organs from that of the whole body. Individual tumour and organ doses were then calculated with the OLINDA/EXM 1.1 software, using the determined TIACs as input. The organ doses were corrected using the option in OLINDA/EXM to adjust the organ masses to the patient-specific values. Tumour doses were estimated using the sphere model provided by OLINDA/EXM. Volume and activity estimates determined using the adaptive thresholding technique were compared to those obtained using a 40% threshold. The choice of a fixed threshold set at 40% was based on the fact that this value is commonly used in clinical practice [72].  8.4 Results 8.4.1  Time-activity curves  Tumours were revealed in 12 of the patients examined in this study. Example whole body images from three of these cases, as well as the image from one patient with no visible lesions are displayed in Figure 8.1.  126  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  Figure 8.1 Anterior (a, b and c) and posterior (d) whole body planar images acquired at approximately 1.5 hours after injection with pathological uptake indicated by arrows. (a) Neuroendocrine tumour in the right maxilla with metastasis to the left mastoid, both thyroid lobes, left subdiaphragmatic and lower intra-abdominal region in patient 1. (b) Neuroendocrine tumour in the small bowel of patient 24. (c) Neuroendocrine tumour in the small bowel with metastasis to the liver in patient 28. (d). Patient 5 has no visible lesions. In this case the posterior view is displayed to show the kidney uptake.  Figure 8.2 shows typical organ TACs for the left kidney, liver and spleen in 5 patients. Median values for the effective and biologic half-lives, determined from the exponential fits through the time activity data for both tumours and organs are summarized in Table 8.1. In several instances, most notably for tumours and the spleen of some patients, the time-activity data was more appropriately fit by a biexponential function than by a monoexponential. The reason for this behaviour may be seen when analysing the dynamic data acquired over the first 30 minutes after injection.  127  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  Figure 8.2 Examples of kidney and liver decay corrected time-activity data fitted with a monoexponential for five patients. Spleen time-activity data is plotted for three of these patients, also showing the use of a biexponential in addition to the monoexponential fit. Table 8.1 Effective and biologic half-lives determined from monoexponential fits through tumours and normal organs.  Region  Teff (h)  Tbiol (h)  Kidneys  5.4 (4.5 – 5.9)  51.5 (18.6 – 511.5)  Liver  5.4 (4.6 – 5.8)  51.0 (20.2 – 136.0)  Spleen  5.3 (4.9 – 5.9)  47.6 (27.8 – 333.5)  Thyroid  4.6 (4.0 – 4.9)  19.7 (11.9 – 26.0)  Tumours  5.3 (4.8 – 6.0)  47.9 (24.1 – 1800)  Data are median, followed by range in parentheses.  The typical uptake behaviour in a patient with no visible lesions, as determined by the 30 minute dynamic planar scans is depicted in Figure 8.3a. In general, the kidneys were found to reach maximum uptake after 5-10 minutes, followed by the washout phase. In the liver, maximum uptake occurred in less than 5 minutes, followed by a rapid decrease and then a more gradual washout phase. Uptake in the spleen was slower and usually extended beyond the length of the 30 minute scan, which explains why the spleen TAC is better fit by a biexponential function. Figure 8.3b shows an example of the typical dynamic behaviour for a patient including normal organ and tumour data. Similar to the spleen, uptake in the tumour increased slowly and did not reach maximum uptake until after the 30 minute scan.  128  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  Figure 8.3 Dynamic planar plots for a patient with no visible lesions (a) and a patient with pathological uptake (b).  In the comparison of spleen TIACs calculated from TACs fitted by monoexponential and biexponential functions in 20 patients, the monoexponential fit was found to overestimate the biexponential derived TIACs by 6.6% ± 2.9% on average. 8.4.2  Organ segmentation  Delineation of organs in the SPECT images using ThV, evaluated from measured SBRs in each patient and parameters from the phantom study, led to the determination of patientspecific organ masses (Table 8.2). Table 8.2 Normal organ masses for males, females and corresponding reference phantom values.  Organ Mass (g)  Males  Mean +/- SD Range Reference Phantom  Females Mean +/- SD Range Reference Phantom a  a  L. Kidney  R. Kidney  Liver  Spleen  177 +/- 22  169 +/- 30  1693 +/- 411  276 +/- 97  141 – 209  126 – 212  1158 – 2553  82 – 422  150  150  1910  183  127 +/- 29  138 +/- 31  1549 +/- 408  228 +/- 85  81 – 187  86 – 194  1072 – 2603  123 – 362  138  138  1400  150  The reference phantom values come from the phantom series of Cristy and Eckerman [123] and from Stabin  et al. [124].  129  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  Regions defined using a fixed threshold of 40% underestimated organ volumes obtained using ThV by 22.6% for the kidneys, 11.7% for the liver and 16.1% for the spleen. A visual check of the SPECT derived organ contours drawn on the CT slices confirmed that organ segmentation using ThV represented organ boundaries more accurately than the 40% threshold (Figure 8.4). The total activity inside volumes delineated using ThV were lower than the total activity inside volumes delineated using ThA by 17%, on average.  Figure 8.4 Segmentation of the left kidney on transaxial SPECT (a) and CT (b), and coronal SPECT (c) and CT (d) slices comparing the use of a fixed 40% threshold (white dashed line), ThV (solid magenta line), and ThA (dotted yellow line).  8.4.3  Dose calculation  Table 8.3 lists the average values of the TIACs used as input for dose calculation with the OLINDA/EXM software. The average relative absorbed doses calculated for normal organs are summarized in Table 8.4. The spleen typically received the highest relative absorbed dose 130  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  with an average over all patients of 0.030 ± 0.012 mGy/MBq from the injection of  99m  Tc-  Tektrotyd. The kidneys followed with an average dose of 0.021 ± 0.007 mGy/MBq, while the average doses to the urinary bladder wall, liver and thyroid were found to be 0.014 ± 0.004 mGy/MBq, 0.012 ± 0.005 mGy/MBq and 0.004 ± 0.001 mGy/MBq, respectively. The calculated effective doses after  99m  Tc-Tektrotyd injection ranged from 2.7 to 6.4 mSv.  Separate values of the average normal organ doses were also calculated based on the division of patients into two groups, representing those with and without observed pathological uptake. A t test with a confidence level of 95% showed no statistical difference between the mean doses in any one organ in the two groups of patients. Table 8.3 Organ time-integrated activity coefficients.  Organ  TIAC (Bq·h/Bq)  Kidneys  0.35 ± 0.10 (0.19 – 0.54)  Liver  0.75 ± 0.31 (0.20 – 1.72)  Spleen  0.43 ± 0.20 (0.04 – 0.89)  Thyroid  0.01 ± 0.004 (0.01 – 0.02)  Urinary Bladder  0.23 ± 0.09 (0.12 – 0.53)  Remainder  4.34 ± 0.74 (3.00 – 6.10)  Data are average ± standard deviation, followed by range in parentheses. Table 8.4 Relative absorbed doses.  Organ  Dose (mGy/MBq)  Kidneys  0.021 ± 0.007 (0.011 – 0.039)  Liver  0.012 ± 0.005 (0.005 – 0.028)  Spleen  0.030 ± 0.012 (0.005 – 0.057)  Thyroid  0.004 ± 0.001 (0.003 – 0.005)  Urinary Bladder Wall  0.014 ± 0.004 (0.008 – 0.024)  Effective Dose (mSv)  4.6 ± 1.1 (2.7 – 6.4)  Data are average ± standard deviation, followed by range in parentheses.  The estimated tumour masses and doses along with the normal organ doses for patients with pathological uptake are summarized in Table 8.5. There was a wide spread in tumour  131  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  relative absorbed dose estimates ranging from 0.003 mGy/MBq to 0.053 mGy/MBq, with the majority of tumour doses on the low end of this range. The median tumour dose was 0.007 mGy/MBq. Table 8.5 Tumour masses and doses as well as normal organ doses for patients with pathological uptake.  # of tumours  Dose (mGy/MBq)  analyzed  Tumour mass (g) a  Tumour(s) b  Kidneys  Liver  Spleen  1  3  139 (8 – 123)  0.041 (0.031 – 0.047)  0.024  0.014  0.030  3  7  85 (8 – 25)  0.004 (0.003 – 0.005)  0.012  0.011  0.015  10  1  17  0.005  0.023  0.015  0.039  17  2  1455 (600 – 855)  0.004 (0.003 – 0.005)  0.023  0.014  0.037  19  3  249 (47 – 141)  0.024 (0.022 – 0.028)  0.023  0.012  0.034  21  1  23  0.013  0.033  0.014  0.025  22  1  122  0.018  ND  ND  ND  23  1  86  0.029  0.026  0.018  0.055  24  1  76  0.053  0.018  0.015  0.029  25  1  44  0.033  0.019  0.010  0.040  27  2  246 (70 – 176)  0.006 (0.004 – 0.007)  ND  ND  ND  28  1  7  0.016  0.013  0.010  0.024  Pt.  a  When more than 1 lesion was analyzed, total mass is listed, followed by range in parenthesis.  b  When more than 1 lesions was analyzed, average dose is listed, followed by range in parentheses.  ND, not determined in cases where region fell outside of SPECT field of view.  8.5 Discussion The image-based dose calculation method presented in this chapter was designed to provide accurate patient-specific dose estimates while working within the limitations of a busy clinical environment. To this end, a hybrid planar/SPECT technique was employed to obtain source region cumulated activities, with image segmentation performed using adaptive thresholds. The average normal organ relative absorbed doses estimated in this study were 0.021 mGy/MBq, 0.030 mGy/MBq, 0.014 ± 0.004 mGy/MBq and 0.012 mGy/MBq for the kidneys, spleen, urinary bladder and liver, respectively (Table 8.4). No statistical differences were noted in average organ absorbed doses when considering patients with and without  132  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  pathological uptake separately. The 99mTc-HYNIC-TOC doses reported here are significantly lower than previously reported  111  In-DTPA-octreotide (Octreoscan) dosimetry estimates of  0.41 mGy/MBq, 0.57 mGy/MBq, 0.20 mGy/MBq and 0.10 mGy/MBq for the kidneys, spleen, bladder and liver respectively [143]. Since four to five times lower than  111  In injected activities are usually about  99m  Tc activities, this translates to approximately four times  higher radiation dose to the kidneys with Octreoscan. The lower radiation dose associated with  99m  Tc-HYNIC-TOC imaging is particularly desirable for patients undergoing repeated  scans, as well as for younger patients. In the only other published dosimetry study of 99mTc-HYNIC-TOC by González-Vázquez et al., the reported doses for the kidneys, spleen and liver were 0.029 ± 0.005 mGy/MBq, 0.033 ± 0.005 mGy/MBq and 0.007 ± 0.001 mGy/MBq, respectively [142]. While the dose estimates in the two studies agree within the provided uncertainties, the mean doses estimated in the current study are approximately 30% less in the kidneys and 50% greater in the liver compared to the González-Vázquez et al. estimates. These differences could be due to the use of 2D imaging by González-Vázquez et al. versus the 3D imaging employed in this work. It is widely recognized that there can be a large discrepancy in dose estimation when using either a 2D or 3D dosimetry protocol (Section 3.2.3). Significant errors in activity quantification based on a 2D processing method can be attributed to the inability to correctly account for attenuation, scatter, and overlapping source regions in planar images. Using the patient-specific 3D dosimetry protocol outlined in this work, a large interpatient dose variation in normal organs and tumours was observed. The interpatient kidney dose ranged from 0.011 mGy/MBq to 0.039 mGy/MBq (Table 8.4), while the range of tumour doses varied from 0.003 mGy/MBq to 0.053 mGy/MBq (Table 8.5). Furthermore, the ratio of tumour-to-kidney dose ranged from 0.13 (patient 17) to 2.9 (patient 24). These different values reveal the importance of determining patient-specific dose estimates. The large variation in the tumour-to-kidney dose ratio among patients demonstrates how  99m  Tc-  HYNIC-TOC could be used to characterize therapy candidates and thus choose which cases are the more suitable for peptide receptor radionuclide therapy (PRRT). Notably, the therapeutic agents of choice in PRRT are commonly DOTATATE, for which  90  Y-DOTATOC and  177  Lu-  99m  Tc-HYNIC-TOC might not be an ideal candidate for accurate  dose prediction, given the slightly different peptide structure and relatively short half-life of  133  Chapter 8: Patient-Specific Dosimetry of 99mTc-HYNIC-TOC in Neuroendocrine Tumours  99m  Tc. Nonetheless, the information gained from a diagnostic scan with  99m  Tc-HYNIC-TOC  can still be useful for identifying patients with an increased SSTR subtype 2 expression, and for recognizing those at increased risk of renal toxicity if treated with PRRT.  8.6 Conclusion This work outlines how hybrid planar/SPECT  99m  Tc-HYNIC-TOC studies can provide  clinicians not only with diagnostic (qualitative or semi-quantitative) information, but also patient-specific quantitative parameters such as TACs, absolute activity and absorbed doses for tumours and normal organs. The large dose variations observed in this study demonstrate the significant impact that patient-specific considerations can have on treatment planning decisions. In the future, similar comprehensive analyses may serve as a supplement to the standard clinical examination of lesions overexpressing SSTR and could aid in the selection of patients for therapy and therapy planning through the accurate quantification of tumour and normal organ uptake of radiolabelled somatostatin analogs.  134  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres 9.1 Introduction The final stage of this thesis was to perform patient-specific dose calculations in order to investigate a potential dose-response relationship. For this purpose, ten patients treated by radioembolization with  188  Re-human serum albumin (HSA) microspheres were analyzed.  Building on the work described in previous chapters, dose calculations were performed for these patients using the dose calculation GUI described in Chapter 4 and using the voxel S value method, which was described in Section 3.3.2. This approach was chosen based on the results of Chapter 7, which demonstrated that voxel S values provide a method for quickly calculating dose distributions that are nearly equivalent to the results from Monte Carlo simulation for self-organ irradiation. Additionally, the liver is a relatively uniform soft tissue medium, which further justifies the use of voxel S values for this application.  9.2 Radioembolization Overview Radioembolization is a form of radionuclide therapy used to treat advanced cancers of the liver. It involves the injection of radiolabelled microspheres directly into the hepatic artery. The effectiveness of this therapy for the treatment of hepatocellular carcinomas is based on the fact that tumours greater than 2 cm in diameter draw more than 80% of their blood supply from the hepatic artery, whereas normal liver parenchyma draws more than 80% of its blood supply from the portal vein [144]. The most commonly used radionuclide for radioembolization is  90  Y [145,146]. In a  randomized phase II clinical trial reported by Van Hazel et al., radioembolization in combination with chemotherapy was found to result in a substantial increase in the time to progression of disease (18.6 months, compared to a time of 3.6 months in patients receiving chemotherapy alone) [147]. Although effective, the use of 90Y also comes with a number of disadvantages. Besides its high cost,  90  Y is a pure beta emitter, which makes post-therapy 135  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  imaging challenging. As a result, accurate dose assessment is difficult. Solutions for imaging 90  of  Y activity distributions, which include bremsstrahlung imaging [148] and  90  Y PET  imaging [149], generally produce images of poor quality that may only potentially be improved through sophisticated software and hardware [150-152]. Alternatively, the use of  188  Re for radioembolization has also been explored [153,154].  The maximum beta emission energy of 188Re is 2.12 MeV, which is similar to the 2.28 MeV maximum beta energy of  90  Y (Table 2.1). Furthermore,  188  Re emits gammas with an energy  of 155 keV and an abundance of 15.1%. This gamma can be used for SPECT imaging, providing the opportunity to perform post-therapy dose estimation. Additional advantages include the fact that 188Re is a low-cost alternative to 90Y and can be produced on site using a 188  W/188Re generator. Three-dimensional dose distributions are rarely calculated during radioembolization  procedures [155]. Instead, radiation dose estimates in the treatment of hepatic tumours with microspheres have traditionally been performed assuming uniform activity throughout the entire liver and tumour volume [156]. This approach leads to an overestimation of liver dose and an underestimation of tumour dose. As an alternative, the partition model was developed in the 1990s as a method for calculating the activity to inject to deliver a prescribed tumour dose [157]. Using the partition model, mean liver doses and mean tumour doses are estimated based on the ratio of activity concentrations of these two regions and their corresponding masses. As summarized in Section 3.3, options for estimating voxelized 3D dose distributions include the use of voxel S values or Monte Carlo simulation. Voxel S values are an attractive option, especially since the majority of the injected activity is constrained to the liver, where the homogeneous medium assumed by the voxel S value approach serves as a reasonable approximation.  9.3 Methods 9.3.1  Patient studies  The ten patients included in this work were part of a phase II study approved by the Clinical Ethics Committee at the Department of Radiology and Diagnostic Imaging of the Hospital of Ministry of Internal Affairs & Administration (Warsaw, Poland). The selected patients had advanced primary or secondary liver cancers and had already been 136  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  unsuccessfully treated with standard therapies. These patients underwent radioembolization with 188Re-HSA microspheres through catheterization of the left or right hepatic artery. The amount of  188  Re administered was based on the ratio of tumour to liver volume  determined by segmentation of these regions from an MRI scan. For tumour to liver volume ratios over 50%, between 25% and 50%, and under 25%, the total administered activity was over 9 GBq, between 6 GBq and 9 GBq, and less than 3 GBq, respectively. A hybrid SPECT/CT camera was not available, so a CT scan was performed before radioembolization for each patient. A SPECT scan was acquired using a Siemens e.cam with a medium energy low penetration (MELP) collimator between 18 and 24 hours after administration of 188Re-HSA. For this SPECT scan, 128 projections were acquired over 360º at 15 seconds per stop. The projection matrices were 128 x 128 in size with a pixel dimension of 4.8 mm. The data was collected using two energy windows: one centered on the main photopeak at 155 keV ± 10% and a second set at 190 keV ± 7.5% to measure bremsstrahlung contamination. An MRI was also obtained for each patient and used to determine the liver and tumour volume with the commercial software syngo.via (Siemens).2 9.3.2  Image reconstruction  SPECT reconstruction of  188  Re activity distributions is complicated by the high energy  beta emissions, which result in bremsstrahlung radiation that contaminates the energy window centered on the  188  Re photopeak. Thus, image reconstruction was performed similar  to the SCACRR method described in Section 5.2.2 with an additional step to compensate for the bremsstrahlung contamination.3 In this approach, an initial activity distribution estimate was obtained with attenuation correction and resolution recovery incorporated into the system matrix and the bremsstrahlung component  incorporated into the forward step of the  OSEM algorithm: (9.1) The bremsstrahlung component was calculated using a window based method. Specifically,  2 3  was set equal to the number of counts recorded in the upper energy window  The patient studies described in this section were performed by collaborators in Warsaw, Poland. The image reconstruction described in this chapter was performed by Dr. Sergey Shcherbinin. 137  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  (centered on 190 keV) multiplied by a normalization coefficient. The normalization coefficient was set equal to the ratio of counts in a background region in the main energy window over the counts in the same background region in the upper energy window. This normalization took into account differences in the energy window widths set at 155 keV ± 10% and 190 keV ± 7.5%. Next, the image estimated with attenuation correction, resolution recovery and bremsstrahlung compensation was used to calculate the scatter component  using the  analytical photon distribution interpolated (APDI) method (Section 5.2.2). A final OSEM reconstruction was performed with the scatter component added into the forward projection step: (9.2) The resulting image included corrections for attenuation, resolution loss, bremsstrahlung contamination and scatter. This reconstruction approach, which incorporated corrections for attenuation and resolution recovery into the system matrix, and that included model-based scatter correction and window-based correction for bremsstrahlung into the forward step of the OSEM algorithm has been previously discussed by Shcherbinin et al. [158]. 9.3.3  Activity quantification  As described in Section 2.12, absolute quantification requires a calibration factor  in  order to convert reconstructed counts into an activity distribution in units of becquerels. For this study, a planar sensitivity measurement was performed using a  188  Re point source  scanned in air. However, a modification to Eq. (2.20) was needed to account for bremsstrahlung contamination  in the sensitivity measurement. (9.3)  Thus, the sensitivity measurement was performed using the same two energy windows used for the patient studies. As before,  represents the bremsstrahlung contribution in the main  window and was calculated using the number of counts in the upper window multiplied by the normalization coefficient (Section 9.3.2).  138  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  The last consideration needed to achieve absolute quantification was the fact that therapeutic levels of activity (several GBq) can result in significant dead time of the camera (Section 2.10.5). With the Siemens e.cam, the percent dead time can be recorded using the manufacturer’s software. However, the dead time was not recorded during the patient scans. Thus, a series of phantom experiments were performed in order to characterize the dead time of the system. These experiments comprised nine planar scans of sources of 188Re ranging in activity from 191 MBq to 6148 MBq. For each scan, the percent dead time provided by the manufacturer’s software was recorded and plotted versus the observed number of counts. Analysis of this data showed that a paralyzable model described the behaviour of the system, with a maximum observable count rate in the main energy window of 2200 kilocounts per minute. The dependence of dead time on the number of counts in the main energy window (  ) was modeled using the exponential function  parameters  ,  and  , where the  were found using the trust-region-reflective algorithm with the  MATLAB optimization toolbox. This procedure provided the exponential function that could be used to determine the percent dead time as a function of the observed number of counts for each patient scan. 9.3.4  Dose calculation  Patient-specific internal dose calculations were performed using the quantitative 3D images of the biodistribution of 188Re-HSA microspheres reconstructed from the post-therapy SPECT acquisitions. The 3D distribution of time-integrated activities needed for dose calculation were determined from the single SPECT time point using physical decay only, since the microspheres were assumed to remain in the liver permanently. The voxelized distribution of absorbed dose in the liver was calculated from this 3D distribution of timeintegrated activities using the voxel S value approach (Section 3.3.2). 9.3.5  Data analysis  Tumours were segmented in SPECT images by finding the threshold value that recovered the tumour volume determined from the MRI scan of each patient. These segmented regions were used to determine the average and maximum tumour doses, and the D90 for each tumour volume. Finally, to investigate a potential dose-response relationship, patient survival was plotted as a function of mean tumour dose, maximum tumour dose and as a function of  139  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  D90. Spearman’s rank correlation coefficient was used to evaluate whether there was any association between the various dose parameters (mean, maximum, D90) and patient overall survival. The cut-off level for statistical significance of the correlation coefficient was taken at a P value of 0.05 [159].  9.4 Results 9.4.1  Dose estimates  The voxel S value calculation used to determine the 3D dose distributions was finished in less than 10 seconds for each patient. A summary of the patient’s overall survival and corresponding dose estimates is listed in Table 9.1. The maximum tumour dose ranged from 31 Gy (patient 3) to 226 Gy (patient 2), while the average tumour dose ranged from 10.5 Gy (patient 9) to 81.6 Gy (patient 2). The tumour dose volume histograms (DVHs) for all ten patients are plotted in Figure 9.1. Table 9.1 Radioembolization patient information and dose parameters.  OS  Tumour  Liver  Mean tumour  Maximum tumour  D90  (months)  Volume (ml)  volume (ml)  dose (Gy)  dose (Gy)  (Gy)  1  6  303  1432  41.7  161.3  21  2  10  145  1432  81.6  226  35.5  3  7  1815  3147  11.9  31.1  6.8  4  22  357  1866  42.8  115.6  24.1  5  12  451  2642  46.8  75.3  35.9  6  7  290  2477  28.8  38.8  25.3  7  4  550  2360  44  176.1  11.6  8  3  580  2347  41.2  143.9  20.9  9  7  2100  3140  10.5  35.4  3.1  10  10  920  2880  21.3  86.5  9.1  Pt.  OS, Overall Survival  140  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  Figure 9.1 Tumour dose volume histograms for ten radioembolization patients treated with  188  Re-HSA  microspheres.  9.4.2  Dose-response  Plots investigating the relationship between overall survival and the absorbed dose are included in Figure 9.2. No significant correlation was observed between overall survival and the dose parameters plotted in Figure 9.2(a-c). To remove potential influence of tumour burden on overall survival, results from a subset of patients with similar tumour volumes were plotted in Figure 9.2(d-f). The patients excluded from this analysis were patients 3, 9 and 10, who had tumour volumes of 1815 mL, 2100 mL and 920 mL, respectively. The remaining patients all had tumour volumes between 145 mL and 580 mL. The correlation between overall survival and D90 plotted for the subset of patients in Figure 9.2(f) was 0.75, which was statistically significant (P < 0.05). No other statistically significant correlation was found between overall survival and the other measured dose parameters (P > 0.2).  141  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  Figure 9.2 Overall survival of all ten patients is plotted versus average tumour dose (a), maximum tumour dose (b), and D90. In the bottom row, overall survival data from a subset of seven patients with similar tumour volumes is plotted against average tumour dose (d), maximum tumour dose (e), and D90 (f). Of these data, only the plot of overall survival in the subset of patients versus D90 was found to have a statistically significant correlation of 0.75 (P < 0.05).  9.5 Discussion In this study, it has been demonstrated that maximum tumour doses for patients treated with  188  Re microspheres can reach levels well over 100 Gy. However, there was significant  interpatient variability with maximum tumour doses ranging from 31 Gy to 226 Gy. Similarly large ranges of absorbed doses have been observed in other studies treating hepatocellular carcinoma with intra-arterial injection of radionuclides, where the target dose is usually between 110 and 150 Gy [160]. Large interpatient variability was also observed in the shape of the DVHs plotted in Figure 9.1. This was due to differences in the heterogeneity of uptake in different patients. Patients with more uniform uptake throughout the tumour volume had sharper DVHs, whereas patients with more heterogeneous uptake had flatter DVHs. Some preliminary observations can be made based on the small population of patients included in this study. First, the only significant correlation between dose and patient response was observed when focusing on a subgroup of patients with similar tumour 142  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  volumes. Although no significant correlation was observed between tumour volume and patient survival, this finding could be related to the observation by Kwekkeboom et al., who reported that patients with greater tumour load were less likely to respond to treatment [161]. Second, there was no correlation between overall survival and average tumour dose or between overall survival and the maximum tumour dose, but there was a significant correlation between D90 and overall survival. These results could be explained by the fact that average and maximum tumour dose do not provide information about dose uniformity and possibly under-dosed regions of the tumour, which would be expected to have a large influence on patient response. The D90 is the only one of the three estimated parameters that contains information about the minimum tumour dose. The dose results listed in Table 9.1 and illustrated in the DVHs in Figure 9.1 demonstrate that the dose distributions in patients with the highest maximum tumour doses were not as uniform as the dose distributions in some of the other patients. For example, patient 5 had the highest D90 at 35.9 Gy, but a maximum tumour dose of only 75 Gy. On the other hand, of the patients with the four highest maximum tumour doses (all above 144 Gy), only one (patient 2) had a D90 close to that of patient 5. Patients 2 and 5 ended up with the second and third longest overall survival periods compared to all other patients. The other three patients with high maximum tumour doses (patients 1, 7 and 8) all had D90’s of 21 Gy or less. This could help explain why patients 1, 7 and 8 ended up with the three shortest overall survival periods of the entire group even though they had high maximum tumour doses. Patient 4 was an outlier with the longest survival time of 22 months, but dose parameters that ranked in the middle of the group. It should be noted that all patients in this study had advanced disease and had already failed multiple previous treatments. Better response rates could be expected in the treatment of patients with less advanced disease.  9.6 Conclusion Radioembolization using  188  Re microspheres is a promising treatment option. This study  provides evidence that detailed 3D dose estimation, which is needed to calculate the D90, is required in order to predict patient response. However, an additional study with a larger  143  Chapter 9: Patient-Specific Dosimetry for Radioembolization with 188Re Microspheres  population of patients would be required to further investigate a possible dose-response relationship.  144  Chapter 10: Conclusions and Future Work  Chapter 10: Conclusions and Future Work 10.1 Conclusions The objective of this thesis was to develop and investigate techniques for internal dose calculation that would be practical to implement into a busy clinical environment. The first step towards meeting this objective was the development of a dosimetry tool that could be used for a variety of dose calculation methodologies and that kept the procedure within the framework of a single software environment. The next step was the design of a segmentation method that is simple enough to be used by any nuclear medicine department, but robust enough to meet the challenge of producing accurate and consistent results independent of the user and image processing method. The repeatability and reproducibility analysis performed in Chapter 6 validated the robustness of the method under these conditions, which was a further step towards the goal of developing a method that could potentially be put into widespread clinical use. The accurate volume and activity estimates provided by the iterative adaptive thresholding technique demonstrate the usefulness of this method for organ level dose calculations. The comparison of a variety of dose estimation methods in Chapter 7 provided insight into the impact of using stylized phantoms representing the average patient, which is the method used in the majority of currently performed dose calculation procedures. Looking at the total dose distributions, it was observed that the mean organ doses calculated by OLINDA/EXM usually agreed well with mean organ doses calculated by Monte Carlo simulation. This was in spite of the fact that patient-specific cross organ S values differed greatly compared to the reference S values used by OLINDA/EXM. Given the similar results obtained by the two methods, the current use of OLINDA/EXM at least for diagnostic procedures where the dose is relatively low appears to be appropriate. However, OLIDNA/EXM has important limitations for therapeutic applications. In these situations, the calculation of 3D dose distributions, the assessment of individual kidney doses and the accurate determination of tumour doses (rather than using the sphere model) requires more  145  Chapter 10: Conclusions and Future Work  sophisticated dosimetry software. The internal dosimetry toolkit developed in this thesis is an example of a dosimetry program that can meet this need. The Monte Carlo results in Chapter 7 were also compared to voxel S value calculations. Although the cross-organ doses calculated by Monte Carlo and voxel S values did not always agree, the close agreement between these two methods within source organs indicates that fast and accurate 3D dose distributions can be calculated using the voxel S value technique. Next, in Chapter 8, OLINDA/EXM calculations were used to calculate the mean doses to source organs of interest in the diagnostic study of patients with suspected NETs. This study revealed a wide interpatient variability in absorbed doses to tumours and healthy organs. This variability provides motivation for performing patient-specific internal dose calculations. Personalized dose calculations allow for selection of patients for therapy, patient-specific treatment planning and monitoring of therapeutic doses. Furthermore, as more accurate internal dose calculations are performed, knowledge of the dose-response relationship will be enhanced. Relatively few patients were included in the  188  Re radioembolization study conducted in  Chapter 9 making it difficult to identify a significant relationship between dose and response. However, in the small number of patients investigated, it was observed that patient survival tended to increase with an increase in D90. Furthermore, there did not appear to be a relationship between patient survival and mean tumour dose. This finding strengthened the opinion that organ level calculations, which calculate the mean organ dose, are inadequate for therapeutic procedures. A 3D dose estimation approach using the voxel S value technique (as used in Chapter 9) or Monte Carlo simulation should be considered mandatory for suitable therapeutic dose assessment in order to predict response.  10.2 Future Work There are several possibilities for expanding on the work presented in this thesis. First of all, there will always be new and useful functions to add to the dose calculation GUI. Of particular importance, it would be valuable to give the user the option to coregister multiple SPECT images and to calculate time-integrated activities at the voxel level. Furthermore, it would be very useful to add the in-house SPECT image reconstruction software that was used throughout this work. This would allow one to start from the projection data and  146  Chapter 10: Conclusions and Future Work  perform a SPECT-based dosimetry calculation all within a single software environment. In addition, it would also be of interest to add the capability to calculate radiobiological quantities such as the biological effective dose. Finally, various dosimetry models could be added to the GUI including the urinary bladder model, gastrointestinal tract model and a method for calculating the dose to the bone marrow. There are also many possibilities for further investigation of the iterative adaptive thresholding method developed in this work. This method was only validated for distributions of 131  99m  Tc. It would be interesting to test the method using radioisotopes such as  I that lead to images with suboptimal image quality compared to  99m  Tc. Furthermore, the  accuracy of this segmentation method was only assessed using a phantom experiment. Although this may be the best option for investigating the accuracy of activity estimates, it would be useful to assess the accuracy of volume estimates in patient studies by comparing tumour and organ volumes obtained using the iterative adaptive thresholding method to volume estimates obtained from anatomical imaging modalities (CT or MRI). Since the volume in which the radionuclide localizes can be different from the anatomical volume, in some cases it may not be possible to get good agreement between the SPECT-derived and CT-derived boundaries. The method should also be tested under more conditions where it might be expected to fail, such as in the investigation of small volumes and regions with heterogeneous uptake. In particular, volumes lower than 12 mL, which was the smallest bottle volume considered in this work, should be experimented with to demonstrate the lower limit for the proposed method. Future studies should include repeated acquisitions of the test phantom so that bias and variance in volume and activity estimates can be evaluated. The comparison of Monte Carlo simulation to OLINDA/EXM dose estimation in Chapter 7 was performed using the most up-to-date version of the OLIDNA/EXM software (version 1.1), which uses S values calculated for computational phantoms based on stylized models. It would be interesting to repeat the comparison of patient-specific S values calculated using Monte Carlo to the new reference phantom S values based on the more realistic image-based models, which will be available in future versions of OLINDA/EXM [90]. Additionally, the total organ doses were only compared in organs with specific uptake of the radiopharmaceutical. Organs without specific uptake, where dose contributions only come from cross-organ irradiation and the remainder of the body, should also be investigated. It is  147  Chapter 10: Conclusions and Future Work  expected that in these organs, the comparison of total doses calculated by Monte Carlo and OLINDA/EXM will be in greater disagreement than the source organs analyzed in this work. Finally, one of the major issues that needs to be explored in more detail is further examination into the relationship between absorbed dose and patient response. It would be informative to use response parameters other than patient survival in the dose-response evaluation. For example, a correlation between tumour reduction and dose could have been examined. Although tumour reduction data was not available from the patients in this work, this could be investigated in future work. A clear dose-response relationship will need to be presented before patient-specific internal dose calculations are accepted universally as a crucial requirement in radionuclide therapy procedures.  148  Bibliography 1.  Winau F, Westphal O, Winau R. Paul Ehrlich--in search of the magic bullet. Microbes Infect 2004;6:786-9.  2.  Kogelnik HD. The history and evolution of radiotherapy and radiation oncology in Austria. Int J Radiat Oncol Biol Phys 1996;35:219-26.  3.  Proescher F, Almquest BR. Contribution on the biological and pathological action of soluble radium salts - with special reference to its therapeutic value in pernicious anaemia and leukemia. Radium 1914;3:65-71.  4.  Patton DD. The birth of nuclear medicine instrumentation: Blumgart and Yens, 1925. J Nucl Med 2003;44:1362-5.  5.  McCready VR. Milestones in nuclear medicine. Eur J Nucl Med 2000;27:S49-79.  6.  Thwaites DI, Tuohy JB. Back to the future: the history and development of the clinical linear accelerator. Phys Med Biol 2006;51:R343-62.  7.  Seidlin SM, Marinelli LD, Oshry E. Radioactive iodine therapy; effect on functioning metastases of adenocarcinoma of the thyroid. J Am Med Assoc 1946;132:838-47.  8.  Marinelli LD, Quimby EH, Hine GJ. Dosage determination with radioactive isotopes; practical considerations in therapy and protection. Am J Roentgenol Radium Ther 1948;59:260-81.  9.  Greig WR, McDougall IR, Halnan KE. Treatment with unsealed radioisotopes. Br Med Bull 1973;29:63-8.  10.  van Essen M, Krenning EP, Kam BL, de Jong M, Valkema R, Kwekkeboom DJ. Peptide-receptor radionuclide therapy for endocrine tumors. Nat Rev Endocrinol 2009;5:382-93.  11.  Cremonesi M, Botta F, Di Dia A, Ferrari M, Bodei L, De Cicco C, Rossi A, Bartolomei M, Mei R, Severi S, Salvatori M, Pedroli G, Paganelli G. Dosimetry for treatment with radiolabelled somatostatin analogues. A review. Q J Nucl Med Mol Imaging 2010;54:37-51.  12.  Intensity Modulated Radiation Therapy Collaborative Working Group. Intensitymodulated radiotherapy: current status and issues of interest. Int J Radiat Oncol Biol Phys 2001;51:880-914.  149  13.  Shen S, Meredith RF, Duan J, Macey DJ, Khazaeli MB, Robert F, LoBuglio AF. Improved prediction of myelotoxicity using a patient-specific imaging dose estimate for non-marrow-targeting (90)Y-antibody therapy. J Nucl Med 2002;43:1245-53.  14.  Pauwels S, Barone R, Walrand S, Borson-Chazot F, Valkema R, Kvols LK, Krenning EP, Jamar F. Practical dosimetry of peptide receptor radionuclide therapy with (90)Ylabeled somatostatin analogs. J Nucl Med 2005;46 [Suppl 1]:92S-8S.  15.  Barone R, Borson-Chazot F, Valkema R, Walrand S, Chauvin F, Gogou L, Kvols LK, Krenning EP, Jamar F, Pauwels S. Patient-specific dosimetry in predicting renal toxicity with (90)Y-DOTATOC: relevance of kidney volume and dose rate in finding a dose-effect relationship. J Nucl Med 2005;46 Suppl 1:99S-106S.  16.  Siegel JA, Stabin MG, Brill AB. The importance of patient-specific radiation dose calculations for the administration of radionuclides in therapy. Cell Mol Biol (Noisyle-grand) 2002;48:451.  17.  Thierens HM, Monsieurs MA, Bacher K. Patient dosimetry in radionuclide therapy: the whys and the wherefores. Nucl Med Commun 2005;26:593-9.  18.  Flux G, Bardies M, Chiesa C, Monsieurs M, Savolainen S, Strand SE, Lassmann M. Letter to the editor: Clinical radionuclide therapy dosimetry: the quest for the "Holy Gray". Eur J Nucl Med Mol Imaging 2007;34:1699-700.  19.  Stabin MG. The Case for Patient-Specific Dosimetry in Radionuclide Therapy. Cancer Biother Radiopharm 2008;23:273-83.  20.  Fenyes T. Chart of the nuclides. In: Vertes A, Nagy S, Klenscar Z, Lovas RG, Rosch F, editors. Handbook of nuclear chemistry. New York: Springer; 2011. pp. 61-62.  21.  Mulford DA, Scheinberg DA, Jurcic JG. The promise of targeted {alpha}-particle therapy. J Nucl Med 2005;46 Suppl 1:199S-204S.  22.  Saha GB. Production of radionuclides. In: Saha GB, editor. Physics and radiobiology of nuclear medicine. New York, NY: Springer; 2006. pp. 44-55.  23.  Wessels BW, Meares CF. Physical and chemical properties of radionuclide therapy. Semin Radiat Oncol 2000;10:115-22.  24.  Behr TM, Behe M, Stabin MG, Wehrmann E, Apostolidis C, Molinet R, Strutz F, Fayyazi A, Wieland E, Gratz S, Koch L, Goldenberg DM, Becker W. High-linear energy transfer (LET) alpha versus low-LET beta emitters in radioimmunotherapy of  150  solid tumors: therapeutic efficacy and dose-limiting toxicity of 213Bi- versus 90Ylabeled CO17-1A Fab' fragments in a human colonic cancer model. Cancer Res 1999;59:2635-43. 25.  Cremonesi M, Ferrari M, Bodei L, Tosi G, Paganelli G. Dosimetry in Peptide radionuclide receptor therapy: a review. J Nucl Med 2006;47:1467-75.  26.  Madsen MT. Recent advances in SPECT imaging. J Nucl Med 2007;48:661-73.  27.  Peterson TE, Furenlid LR. SPECT detectors: the Anger Camera and beyond. Phys Med Biol 2011;56:R145-82.  28.  Moore SC, Kouris K, Cullum I. Collimator design for single photon emission tomography. Eur J Nucl Med 1992;19:138-50.  29.  Knoll GF. Scintillation detector principles. In: Knoll GF, editor. Radiation detection and measurement. Danvers, MA: John Wiley & Sons, Inc.; 2010. pp. 223.  30.  Wilkinson III F. Scintillation conversion efficiency. In: Wernick MN, Aarsvold JN, editors. Emission tomography. San Diego, California: Elsevier; 2004. pp. 236-237.  31.  Cherry SR, Sorenson JA, Phelps ME. Photomultiplier tubes. In: Cherry SR, Sorenson JA, Phelps ME, editors. Physics in nuclear medicine. Philadelphia, PA: Saunders; 2003. pp. 101-103.  32.  Khalil MM. Elements of gamma camera and SPECT systems. In: Khalil MM, editor. Basic sciences of nuclear medicine. Berlin Heidelberg: Springer-Verlag; 2011. pp. 166-167.  33.  Bruyant PP. Analytic and iterative reconstruction algorithms in SPECT. J Nucl Med 2002;43:1343-58.  34.  Chang LT. A method for attenuation correction in radionuclide computed tomography. IEEE Trans Nucl Sci 1978;25:638-43.  35.  Vandenberghe S, D'Asseler Y, Van de Walle R, Kauppinen T, Koole M, Bouwens L, Van Laere K, Lemahieu I, Dierckx RA. Iterative reconstruction algorithms in nuclear medicine. Comput Med Imaging Graph 2001;25:105-11.  36.  Shepp LA, Vardi Y. Maximum likelihood reconstruction for emission tomography. IEEE Trans Med Imaging 1982;1:113-22.  37.  Hudson HM, Larkin RS. Accelerated image reconstruction using ordered subsets of projection data. IEEE Trans Med Imaging 1994;13:601-9.  151  38.  Frey EC, Tsui BMW. Collimator-detector response compensation in SPECT. In: Zaidi H, editor. Quantitative analysis in nuclear medicine imaging. New York, NY: Springer; 2006. pp. 141-166.  39.  Lyra M, Ploussi A. Filtering in SPECT Image Reconstruction. Int J Biomed Imaging 2011;2011:693795.  40.  Dewaraja YK, Frey EC, Sgouros G, Brill AB, Roberson P, Zanzonico PB, Ljungberg M. MIRD pamphlet No. 23: quantitative SPECT for patient-specific 3-dimensional dosimetry in internal radionuclide therapy. J Nucl Med 2012;53:1310-25.  41.  Soret M, Bacharach SL, Buvat I. Partial-Volume Effect in PET Tumor Imaging. J Nucl Med 2007;48:932-45.  42.  Cherry SR, Sorenson JA, Phelps ME. Dead time. In: Cherry SR, Sorenson JA, Phelps ME, editors. Physics in nuclear medicine. Philadelphia, PA: Saunders; 2003. pp. 178182.  43.  Blankespoor SC, Wu X, Kalki K, Brown JK, Tang HR, Cann CE, Hasegawa BH. Attenuation Correction of SPECT Using X-Ray CT 011 ai1 Emission-Transmission CT System: Myocardial Perfusion Assessinelit. IEEE Trans Nucl Sci 1996;43:226374.  44.  King MA. Compensation methods for correction of nonuniform attenuation. In: Wernick MN, Aarsvold JN, editors. Emission tomography. San Diego, California: Elsevier; 2004. pp. 480-483.  45.  Jaszczak RJ, Greer KL, Floyd CE, Harris CC, Coleman RE. Improved SPECT Quantification Using Compensation for Scattered Photons. J Nucl Med 1984;25:893900.  46.  Koral KF, Swailem FM, Buchbinder S, Clinthorne NH, Rogers WL, Tsui BM. SPECT dual-energy-window Compton correction: scatter multiplier required for quantification. J Nucl Med 1990;31:90-8.  47.  Ichihara T, Ogawa K, Motomura N, Kubo A, Hashimoto S. Compton scatter compensation using the triple-energy window method for single- and dual-isotope SPECT. J Nucl Med 1993;34:2216-21.  48.  Dewaraja YK, Li J, Koral KF. Quantitative  131  I SPECT with triple energy window  compton scatter correction. IEEE Trans Nucl Sci 1998;45:3109-14.  152  49.  Dewaraja YK, Ljungberg M, Fessler JA. 3-D Monte Carlo-Based Scatter Compensation in Quantitative I-131 SPECT Reconstruction. IEEE Trans Nucl Sci 2006;53:181.  50.  Wells RG, Celler A, Harrop R. Analytical calculation of photon distributions in SPECT projections. IEEE Trans Nucl Sci 1998;45:3202-14.  51.  Vandervoort E, Celler A, Wells G, Blinder S, Dixon K, Pang Y. Implementation of an analytically based scatter correction in SPECT reconstructions. IEEE Trans Nucl Sci 2005;52:645-53.  52.  King MA, Schwinger RB, Doherty PW, Penney BC. Two-dimensional filtering of SPECT images using the Metz and Wiener filters. J Nucl Med 1984;25:1234-40.  53.  Tsui BMW, Hu HB, Gilland DR, Gullberg GT. Implementation of simultaneous attenuation and detector response correction in SPECT. IEEE Trans Nucl Sci 1988;35:778-83.  54.  Koral KF, Dewaraja Y. I-131 SPECT activity recovery coefficients with implicit or triple-energy-window scatter correction. Nucl Instrum Meth A 1999;422:688-92.  55.  Shcherbinin S, Celler A. Assessment of the severity of partial volume effects and the performance of two template-based correction methods in a SPECT/CT phantom experiment . Phys Med Biol 2011;56:5371.  56.  Delpon G, Ferrer L, Lisbona A, Bardies M. Correction of count losses due to deadtime on a DST-XLi (SmVi-GE) camera during dosimetric studies in patients injected with iodine-131. Phys Med Biol 2002;47:N79-90.  57.  Koral KF, Zasadny KR, Ackermann RJ, Ficaro EP. Deadtime correction for two multihead Anger cameras in 131I dual-energy-window-acquisition mode. Med Phys 1998;25:85-91.  58.  Willowson K, Bailey DL, Baldock C. Quantitative SPECT reconstruction using CTderived corrections. Phys Med Biol 2008;53:3099-112.  59.  Zeintl J, Vija AH, Yahil A, Hornegger J, Kuwert T. Quantitative accuracy of clinical 99mTc SPECT/CT using ordered-subset expectation maximization with 3dimensional resolution recovery, attenuation, and scatter correction. J Nucl Med 2010;51:921-8.  153  60.  Bolch WE, Eckerman KF, Sgouros G, Thomas SR. MIRD Pamphlet No. 21: A Generalized Schema for Radiopharmaceutical Dosimetry--Standardization of Nomenclature. J Nucl Med 2009;50:477-84.  61.  Bouchet LG, Bolch WE, Blanco HP, Wessels BW, Siegel JA, Rajon DA, Clairand I, Sgouros G. MIRD Pamphlet No 19: absorbed fractions and radionuclide S values for six age-dependent multiregion models of the kidney. J Nucl Med 2003;44:1113-47.  62.  Yorke ED, Williams LE, Demidecki AJ, Heidorn DB, Roberson PL, Wessels BW. Multicellular  dosimetry  for  beta-emitting  radionuclides:  autoradiography,  thermoluminescent dosimetry and three-dimensional dose calculations. Med Phys 1993;20:543-50. 63.  Gardin I, Faraggi M, Huc E, Bok BD. Modelling of the relationship between cell dimensions and mean electron dose delivered to the cell nucleus: application to five radionuclides used in nuclear medicine. Phys Med Biol 1995;40:1001-14.  64.  Matthay KK, Panina C, Huberty J, Price D, Glidden DV, Tang HR, Hawkins RA, Veatch J, Hasegawa B. Correlation of Tumor and Whole-Body Dosimetry with Tumor Response and Toxicity in Refractory Neuroblastoma Treated with 131IMIBG. J Nucl Med 2001;42:1713-21.  65.  Wiseman GA, Kornmehl E, Leigh B, Erwin WD, Podoloff DA, Spies S, Sparks RB, Stabin MG, Witzig T, White CA. Radiation Dosimetry Results and Safety Correlations from 90Y-Ibritumomab Tiuxetan Radioimmunotherapy for Relapsed or Refractory Non-Hodgkin's Lymphoma: Combined Data from 4 Clinical Trials. J Nucl Med 2003;44:465-74.  66.  Fisher DR, Shen S, Meredith RF. MIRD Dose Estimate Report No. 20: Radiation Absorbed-Dose Estimates for 111In- and 90Y-Ibritumomab Tiuxetan. J Nucl Med 2009;50:644-52.  67.  Siegel JA, Thomas SR, Stubbs JB, Stabin MG, Hays MT, Koral KF, Robertson JS, Howell RW, Wessels BW, Fisher DR, Weber DA, Brill AB. MIRD Pamphlet No. 16: Techniques for Quantitative Radiopharmaceutical Biodistribution Data Acquisition and Analysis for Use in Human Radiation Dose Estimates. J Nucl Med 1999;40:37S61.  154  68.  Koral KF, Zasadny KR, Kessler ML, Luo J, Buchbinder SF, Kaminski MS, Francis I, Wahl RL. CT-SPECT Fusion Plus Conjugate Views for Determining Dosimetry in Iodine-131-Monoclonal Antibody Therapy of Lymphoma Patients. J Nucl Med 1994;35:1714-20.  69.  Koral KF, Dewaraja Y, Li J, Barrett CL, Regan DD, Zasadny KR, Rommelfanger SG, Francis IR, Kaminski MS, Wahl RL. Initial Results for Hybrid SPECT-ConjugateView Tumor Dosimetry in 131I-Anti-B1 Antibody Therapy of Previously Untreated Patients with Lymphoma. J Nucl Med 2000;41:1579-86.  70.  Shcherbinin S, Piwowarska-Bilska H, Grimes J, Celler A, Listewnik MH, Zorga P, Birkenfeld B. Quantitative SPECT reconstructions for combined Lu-177/Y-90 radionuclide therapy: Phantom experiments [abstract]. J Nucl Med 2011;52 [Suppl 1]:1745.  71.  Beauregard JM, Hofman MS, Pereira JM, Eu P, Hicks RJ. Quantitative (177)Lu SPECT (QSPECT) imaging using a commercially available SPECT/CT system. Cancer Imaging 2011;11:56-66.  72.  Erdi YE, Wessels BW, Loew MH, Erdi AK. Threshold Estimation in Single Photon Emission  Computed  Tomography  and  Planar  Imaging  for  Clinical  Radioimmunotherapy. Cancer Res 1995;55:5823s-5826. 73.  Biehl KJ, Kong F, Dehdashti F, Jin J, Mutic S, El Naqa I, Siegel BA, Bradley JD. 18F-FDG PET Definition of Gross Tumor Volume for Radiotherapy of Non-Small Cell Lung Cancer: Is a Single Standardized Uptake Value Threshold Approach Appropriate?. J Nucl Med 2006;47:1808-12.  74.  Shcherbinin S, Celler A, Belhocine T, Vanderwerf R, Driedger A. Accuracy of quantitative reconstructions in SPECT/CT imaging. Phys Med Biol 2008;53:4595604.  75.  Mut F, Glickman S, Marciano D, Hawkins RA. Optimum Processing Protocols for Volume Determination of the Liver and Spleen from SPECT Imaging with Technetium-99m Sulfur Colloid. J Nucl Med 1988;29:1768-75.  76.  Fleming JS, Alaamer AS. A rule based method for context sensitive threshold segmentation in SPECT using simulation. Phys Med Biol Med 1998;43:2309.  155  77.  Daisne J, Sibomana M, Bol A, Doumont T, Lonneux M, Grégoire V. Tri-dimensional automatic segmentation of PET volumes based on measured source-to-background ratios: influence of reconstruction algorithms. Radiother Oncol 2003;69:247-50.  78.  Erdi YE, Mawlawi O, Larson SM, Imbriaco M, Yeung H, Finn R, Humm JL. Segmentation of lung lesion volume by adaptive positron emission tomography image thresholding. Cancer 1997;80:2505-9.  79.  Jentzen W, Freudenberg L, Eising EG, Heinze M, Brandau W, Bockisch A. Segmentation of PET Volumes by Iterative Image Thresholding. J Nucl Med 2007;48:108-14.  80.  van Dalen JA, Hoffmann AL, Dicken V, Vogel WV, Wiering B, Ruers TJ, Karssemeijer N, Oyen WJ. A novel iterative method for lesion delineation and volumetric quantification with FDG PET. Nucl Med Commun 2007;28:485-93.  81.  Pacilio M, Basile C, Shcherbinin S, Caselli F, Ventroni G, Aragno D, Mango L, Santini E. An innovative iterative thresholding algorithm for tumor segmentation and volumetric quantification on SPECT images: Monte Carlo-based methodology and validation. Med Phys 2011;38:3050-61.  82.  He B, Wahl RL, Du Y, Sgouros G, Jacene H, Flinn I, Frey EC. Comparison of Residence Time Estimation Methods for Radioimmunotherapy Dosimetry and Treatment Planning—Monte Carlo Simulation Studies. IEEE Trans Med Imaging 2008;27:521-30.  83.  Assie K, Dieudonne A, Gardin I, Buvat I, Tilly H, Vera P. Comparison between 2D and 3D dosimetry protocols in 90Y-ibritumomab tiuxetan radioimmunotherapy of patients with non-Hodgkin's lymphoma. Cancer Biother Radiopharm 2008;23:53-64.  84.  Rajendran JG, Fisher DR, Gopal AK, Durack LD, Press OW, Eary JF. High-Dose 131I-Tositumomab  (Anti-CD20)  Radioimmunotherapy  for  Non-Hodgkin's  Lymphoma: Adjusting Radiation Absorbed Dose to Actual Organ Volumes. J Nucl Med 2004;45:1059-64. 85.  He B, Wahl RL, Sgouros G, Du Y, Jacene H, Kasecamp WR. Comparison of organ residence time estimation methods for radioimmunotherapy dosimetry and treatment planning-patient studies. Med Phys 2009;36:1595-601.  156  86.  Sgouros G, Kolbert KS, Sheikh A, Pentlow KS, Mun EF, Barth A, Robbins RJ, Larson SM. Patient-specific dosimetry for 131I thyroid cancer therapy using 124I PET and 3-dimensional-internal dosimetry (3D-ID) software. J Nucl Med 2004;45:1366-72.  87.  Stabin MG. Developments in the internal dosimetry of radiopharmaceuticals. Radiat Prot Dosimetry 2003;105:575-80.  88.  Bolch WE, Bouchet LG, Robertson JS, Wessels BW, Siegel JA, Howell RW, Erdi AK, Aydogan B, Costes S, Watson EE, In collaboration with the MIRD Committee, Society of Nuclear Medicine: E.E. Watson (Chair), J.S. Robertson (Task Group Leader), W.E. Bolch, A.B. Brill, N.D. Charkes, D.R. Fisher, M.T. Hays, R.W. Howell, J.A. Siegel,S.R.Thomas and B.W.Wessels,. MIRD Pamphlet No. 17: The Dosimetry of Nonuniform Activity Distributions--Radionuclide S Values at the Voxel Level. J Nucl Med 1999;40:11S-36.  89.  Stabin MG, Sparks RB, Crowe E. OLINDA/EXM: The Second-Generation Personal Computer Software for Internal Dose Assessment in Nuclear Medicine. J Nucl Med 2005;46:1023-7.  90.  Stabin MG, Xu XG, Emmons MA, Segars WP, Shi C, Fernald MJ. RADAR Reference Adult, Pediatric, and Pregnant Female Phantom Series for Internal and External Dosimetry. J Nucl Med 2012.  91.  Dieudonne A, Hobbs RF, Bolch WE, Sgouros G, Gardin I. Fine-resolution voxel S values for constructing absorbed dose distributions at variable voxel size. J Nucl Med 2010;51:1600-7.  92.  Pacilio M, Lanconelli N, Lo Meo S, Betti M, Montani L, Torres Aroche LA, Coca Perez MA. Differences among Monte Carlo codes in the calculations of voxel S values for radionuclide targeted therapy and analysis of their impact on absorbed dose evaluations. Med Phys 2009;36:1543-52.  93.  Lanconelli N, Pacilio M, Lo Meo S, Botta F, Di Dia A, Aroche AT, Perez MA, Cremonesi M. A free database of radionuclide voxel S values for the dosimetry of nonuniform activity distributions. Phys Med Biol 2012;57:517-33.  94.  Rogers DWO. Low energy electron transport with EGS. Nucl Instrum Methods 1984;227:535-48.  157  95.  Hendricks JS, Briesmeister JF. Recent MCNP developments. IEEE Trans Nucl Sci 1992;39:1035-40.  96.  Yoriyaz H, Stabin MG, dos Santos A. Monte Carlo MCNP-4B-based absorbed dose distribution estimates for patient-specific dosimetry. J Nucl Med 2001;42:662-9.  97.  Sempau J, Acosta E, Baro J, Fernandez-Varea JM, Salvat F. An algorithm for Monte Carlo simulation of coupled electron-photon transport. Nucl Instrum Meth B 1997;132:377-90.  98.  Pia MG. The Geant4 Toolkit: simulation capabilities and application results. Nucl Phys B 2003;125:60-8.  99.  Ferrer L, Chouin N, Bitar A, Lisbona A, Bardies M. Implementing dosimetry in GATE: dose-point kernel validation with GEANT4 4.8.1. Cancer Biother Radiopharm 2007;22:125-9.  100.  Stabin MG, Zaidi H. Monte carlo codes for use in therapeutic nuclear medicine. In: Zaidi H, Sgouros G, editors. Therapeutic applications of Monte Carlo calculations in nuclear medicine. Bristol: Institute of Physics Publishing; 2003. pp. 133-157.  101.  Kawrakow I. Accurate condensed history Monte Carlo simulation of electron transport. I. EGSnrc, the new EGS4 version. Med Phys 2000;27:485-98.  102.  Walters B, Kawrakow I. DOSXYZnrc users manual 2005; Techn. Rep. No. PIRS794, RevB.  103.  Strigari L, Menghi E, D'Andrea M, Benassi M. Monte Carlo dose voxel kernel calculations of beta-emitting and Auger-emitting radionuclides for internal dosimetry: A comparison between EGSnrcMP and EGS4. Med Phys 2006;33:3383-9.  104.  Visser E, Postema E, Boerman O, Visschers J, Oyen W, Corstens F. Software package for integrated data processing for internal dose assessment in nuclear medicine (SPRIND). Eur J Nucl Med Mol Imaging 2007;34:413-21.  105.  Glatting G, Landmann M, Kull T, Wunderlich A, Blumstein NM, Buck AK, Reske SN. Internal radionuclide therapy: the ULMDOS software for treatment planning. Med Phys 2005;32:2399-405.  106.  Glatting G, Landmann M, Wunderlich A, Kull T, Mottaghy FM, Reske SN. Internal radionuclide therapy: software for treatment planning using tomographic data. Nuklearmedizin 2006;45:269-72.  158  107.  Johnson TK, McClure D, McCourt S. MABDOSE. I: Characterization of a general purpose dose estimation code. Med Phys 1999;26:1389-95.  108.  Clairand I, Ricard M, Gouriou J, Di Paola M, Aubert B. DOSE3D: EGS4 Monte Carlo code-based software for internal radionuclide dosimetry. J Nucl Med 1999;40:1517-23.  109.  Coulot J, Ricard M, Aubert B. Validation of the EGS usercode DOSE3D for internal beta dose calculation at the cellular and tissue levels. Phys Med Biol 2003;48:2591602.  110.  Kolbert KS, Sgouros G, Scott AM, Bronstein JE, Malane RA, Zhang J, Kalaigian H, McNamara S, Schwartz L, Larson SM. Implementation and evaluation of patientspecific three-dimensional internal dosimetry. J Nucl Med 1997;38:301-8.  111.  Prideaux AR, Song H, Hobbs RF, He B, Frey EC, Ladenson PW, Wahl RL, Sgouros G. Three-dimensional radiobiologic dosimetry: application of radiobiologic modeling to patient-specific 3-dimensional imaging-based internal dosimetry. J Nucl Med 2007;48:1008-16.  112.  Tagesson M, Ljungberg M, Strand SE. A Monte-Carlo program converting activity distributions to absorbed dose distributions in a radionuclide treatment planning system. Acta Oncol 1996;35:367-72.  113.  Liu A, Williams LE, Lopatin G, Yamauchi DM, Wong JY, Raubitschek AA. A radionuclide therapy treatment planning and dose estimation system. J Nucl Med 1999;40:1151-3.  114.  Gardin I, Bouchet LG, Assie K, Caron J, Lisbona A, Ferrer L, Bolch WE, Vera P. Voxeldose: a computer program for 3-D dose calculation in therapeutic nuclear medicine. Cancer Biother Radiopharm 2003;18:109-15.  115.  McKay E. A software tool for specifying voxel models for dosimetry estimation. Cancer Biother Radiopharm 2003;18:379-92.  116.  Guy MJ, Flux GD, Papavasileiou P, Flower MA, Ott RJ. RMDP: a dedicated package for 131I SPECT quantification, registration and patient-specific dosimetry. Cancer Biother Radiopharm 2003;18:61-9.  117.  Chiavassa S, Aubineau-Laniece I, Bitar A, Lisbona A, Barbet J, Franck D, Jourdain JR, Bardies M. Validation of a personalized dosimetric evaluation tool (Oedipe) for  159  targeted radiotherapy based on the Monte Carlo MCNPX code. Phys Med Biol 2006;51:601-16. 118.  Loudos G, Tsougos I, Boukis S, Karakatsanis N, Georgoulias P, Theodorou K, Nikita K, Kappas C. A radionuclide dosimetry toolkit based on material-specific Monte Carlo dose kernels. Nucl Med Commun 2009;30:504-12.  119.  Sgouros G, Squeri S, Ballangrud AM, Kolbert KS, Teitcher JB, Panageas KS, Finn RD, Divgi CR, Larson SM, Zelenetz AD. Patient-specific, 3-dimensional dosimetry in non-Hodgkin's lymphoma patients treated with 131I-anti-B1 antibody: assessment of tumor dose-response. J Nucl Med 2003;44:260-8.  120.  Minarik D, Sjogreen-Gleisner K, Linden O, Wingardh K, Tennvall J, Strand SE, Ljungberg M. 90Y Bremsstrahlung imaging for absorbed-dose assessment in highdose radioimmunotherapy. J Nucl Med 2010;51:1974-8.  121.  Brans B, Bodei L, Giammarile F, Linden O, Luster M, Oyen WJ, Tennvall J. Clinical radionuclide therapy dosimetry: the quest for the "Holy Gray". Eur J Nucl Med Mol Imaging 2007;34:772-86.  122.  Grimes J, Celler A, Shcherbinin S, Piwowarska-Bilska H, Birkenfeld B. The accuracy and reproducibility of SPECT target volumes and activities estimated using an iterative adaptive thresholding technique. Nucl Med Commun 2012;33:1254-66.  123.  Cristy M, Eckerman K. Specific Absorbed Fractions of Energy at Various Ages from Internal Photons Sources 1987; V1-V7 ORNL/TM-8381.  124.  Stabin M, Watson E, Cristy M. Mathematical Models and Specific Absorbed Fractions of Photon Energy in the Nonpregnant Adult Female and at the End of Each Trimester of Pregnancy 1995; ORNL report ORNL/TM-12907.  125.  Hatt M, Cheze-le Rest C, van Baardwijk A, Lambin P, Pradier O, Visvikis D. Impact of Tumor Size and Tracer Uptake Heterogeneity in 18F-FDG PET and CT Non-Small Cell Lung Cancer Tumor Delineation. J Nucl Med 2011;52:1690-7.  126.  Hatt M, Visvikis D, Albarghach NM, Tixier F, Pradier O, Cheze-le Rest C. Prognostic value of 18F-FDG PET image-based parameters in oesophageal cancer and impact of tumour delineation methodology. Eur J Nucl Med Mol Imaging 2011;38:1191-202.  160  127.  Brambilla M, Matheoud R, Secco C, Loi G, Krengli M, Inglese E. Threshold segmentation for PET target volume delineation in radiation treatment planning: The role of target-to-background ratio and target size. Med Phys 2008;35:1207-13.  128.  van Baardwijk A, Bosmans G, Boersma L, Buijsen J, Wanders S, Hochstenbag M, van Suylen RJ, Dekker A, Dehing-Oberije C, Houben R, Bentzen SM, van Kroonenburgh M, Lambin P, De Ruysscher D. PET-CT-based auto-contouring in non-small-cell lung cancer correlates with pathology and reduces interobserver variability in the delineation of the primary tumor and involved nodal volumes. Int J Radiat Oncol Biol Phys 2007;68:771-8.  129.  Nestle U, Weber W, Hentschel M, Grosu AL. Biological imaging in radiation therapy: role of positron emission tomography. Phys Med Biol 2009;54:R1-25.  130.  Divoli A, Chiavassa S, Ferrer L, Barbet J, Flux GD, Bardies M. Effect of patient morphology on dosimetric calculations for internal irradiation as assessed by comparisons of Monte Carlo versus conventional methodologies. J Nucl Med 2009;50:316-23.  131.  Erdi AK, Yorke ED, Loew MH, Erdi YE, Sarfaraz M, Wessels BW. Use of the fast Hartley transform for three-dimensional dose calculation in radionuclide therapy. Med Phys 1998;25:2226-33.  132.  Grimes J, Celler A, Birkenfeld B, Shcherbinin S, Listewnik MH, Piwowarska-Bilska H, Mikołajczak R, Zorga P. Patient-specific radiation dosimetry of 99mTc-HYNICTyr3-octreotide in neuroendocrine tumors. J Nucl Med 2011;52:1474-81.  133.  Reubi J, Waser B, Schaer J, Laissue J. Somatostatin receptor sst1-sst5 expression in normal and neoplastic human tissues using receptor autoradiography with subtypeselective ligands. Eur J Nucl Med Mol Imaging 2001; 28:836-846.  134.  Krenning EP, Kwekkeboom DJ, Bakker WH, Breeman WAP, Kooij PPM, Oei HY, et al. Somatostatin receptor scintigraphy with [111In-DTPA-d-Phe1]- and [123I-Tyr3]octreotide: the Rotterdam experience with more than 1000 patients. Eur J Nucl Med Mol Imaging 1993; 20:716-731.  135.  Decristoforo C, Mather SJ, Cholewinski W, Donnemiller E, Riccabona G, Moncayo R.  99m  Tc-EDDA/HYNIC-TOC: a new  99m  Tc-labelled radiopharmaceutical for  imaging somatostatin receptor-positive tumours: first clinical results and intra-patient  161  comparison with  111  In-labelled octreotide derivatives. Eur J Nucl Med Mol Imaging  2000; 27:1318-1325. 136.  Płachcińska A, Mikołajczak R, Maecke H, Młodkowska E, Kunert-Radek J, Michalski A, et al. Clinical usefulness of  99m  Tc-EDDA/HYNIC-TOC scintigraphy in  oncological diagnostics: a preliminary communication. Eur J Nucl Med Mol Imaging 2003; 30:1402-1406. 137.  Czepczyński R, Parisella M, Kosowicz J, Mikołajczak R, Ziemnicka K, Gryczyńska M, et al. Somatostatin receptor scintigraphy using  99m  Tc-EDDA/HYNIC-TOC in  patients with medullary thyroid carcinoma. Eur J Nucl Med Mol Imaging 2007; 34:1635-1645. 138.  Gabriel M, Decristoforo C, Donnemiller E, Ulmer H, Watfah Rychlinski C, Mather SJ, Moncayo R. An Intrapatient Comparison of 99mTc-EDDA/HYNIC-TOC with 111In-DTPA-Octreotide for Diagnosis of Somatostatin Receptor-Expressing Tumors. J Nucl Med 2003;44:708-16.  139.  Maecke HR, Hofmann M, Haberkorn U. 68Ga-Labeled Peptides in Tumor Imaging. J Nucl Med 2005;46:172S-178.  140.  Buchmann I, Henze M, Engelbrecht S, Eisenhut M, Runz A, Schäfer M, et al. Comparison of  68  Ga-DOTATOC PET and  111  In-DTPAOC (Octreoscan) SPECT in  patients with neuroendocrine tumours. Eur J Nucl Med Mol Imaging 2007; 34:16171626. 141.  Frilling A, Sotiropoulos GC, Radtke A, Malago M, Bockisch A, Kuehl H, Li J, Broelsch  CE.  The  impact  of  68Ga-DOTATOC  Positron  Emission  Tomography/Computed Tomography on the multimodal management of patients with neuroendocrine tumors. Ann Surg 2010;252:850-6. 142.  González-Vázquez A, Ferro-Flores G, Arteaga de Murphy C, Gutiérrez-García Z. Biokinetics and dosimetry in patients of 99mTc-EDDA/HYNIC-Tyr3-octreotide prepared from lyophilized kits. Appl Radiat Isotopes 2006;64:792-7.  143.  ICRP Publication 80. Radiation dose to patients from radiopharmaceuticals. Ann ICRP 1998;28.  144.  Giammarile F, Bodei L, Chiesa C, Flux G, Forrer F, Kraeber-Bodere F, Brans B, Lambert B, Konijnenberg M, Borson-Chazot F, Tennvall J, Luster M, Therapy,  162  Oncology and Dosimetry Committees. EANM procedure guideline for the treatment of liver cancer and liver metastases with intra-arterial radioactive compounds. Eur J Nucl Med Mol Imaging 2011;38:1393-406. 145.  Gray BN, Anderson JE, Burton MA, van Hazel G, Codde J, Morgan C, Klemp P. Regression of liver metastases following treatment with yttrium-90 microspheres. Aust N Z J Surg 1992;62:105-10.  146.  Gray B, Van Hazel G, Hope M, Burton M, Moroz P, Anderson J, Gebski V. Randomised trial of SIR-Spheres plus chemotherapy vs. chemotherapy alone for treating patients with liver metastases from primary large bowel cancer. Ann Oncol 2001;12:1711-20.  147.  Van Hazel G, Blackwell A, Anderson J, Price D, Moroz P, Bower G, Cardaci G, Gray B. Randomised phase 2 trial of SIR-Spheres plus fluorouracil/leucovorin chemotherapy versus fluorouracil/leucovorin chemotherapy alone in advanced colorectal cancer. J Surg Oncol 2004;88:78-85.  148.  Shen S, DeNardo GL, DeNardo SJ. Quantitative bremsstrahlung imaging of yttrium90 using a Wiener filter. Med Phys 1994;21:1409-17.  149.  Nickles RJ, Roberts AD, Nye AK, Converse AK, Barnhart TE, Avila-Rodriguez MA, Sundaresan R, Dick DW, Hammas RJ, Thomadsen BR. Assaying and PET imaging of yttrium-90: 1>>34ppm>0. IEEE Nucl Sci Symp Rec 2004;6:3412-4.  150.  Minarik D, Sjogreen Gleisner K, Ljungberg M. Evaluation of quantitative (90)Y SPECT based on experimental phantom studies. Phys Med Biol 2008;53:5689-703.  151.  Walrand S, Jamar F, van Elmbt L, Lhommel R, Bekonde EB, Pauwels S. 4-Step renal dosimetry dependent on cortex geometry applied to 90Y peptide receptor radiotherapy: evaluation using a fillable kidney phantom imaged by 90Y PET. J Nucl Med 2010;51:1969-73.  152.  van Elmbt L, Vandenberghe S, Walrand S, Pauwels S, Jamar F. Comparison of yttrium-90 quantitative imaging by TOF and non-TOF PET in a phantom of liver selective internal radiotherapy. Phys Med Biol 2011;56:6759-77.  153.  Wunderlich G, Pinkert J, Andreeff M, Stintz M, Knapp FF,Jr, Kropp J, Franke WG. Preparation and biodistribution of rhenium-188 labeled albumin microspheres B 20: a promising new agent for radiotherapy. Appl Radiat Isot 2000;52:63-8.  163  154.  Liepe K, Brogsitter C, Leonhard J, Wunderlich G, Hliscs R, Pinkert J, Folprecht G, Kotzerke J. Feasibility of high activity rhenium-188-microsphere in hepatic radioembolization. Jpn J Clin Oncol 2007;37:942-50.  155.  Dieudonne A, Garin E, Laffont S, Rolland Y, Lebtahi R, Leguludec D, Gardin I. Clinical feasibility of fast 3-dimensional dosimetry of the liver for treatment planning of hepatocellular carcinoma with 90Y-microspheres. J Nucl Med 2011;52:1930-7.  156.  Andrews JC, Walker SC, Ackermann RJ, Cotton LA, Ensminger WD, Shapiro B. Hepatic  radioembolization  with  yttrium-90  containing  glass  microspheres:  preliminary results and clinical follow-up. J Nucl Med 1994;35:1637-44. 157.  Ho S, Lau WY, Leung TW, Chan M, Ngar YK, Johnson PJ, Li AK. Partition model for estimating radiation doses from yttrium-90 microspheres in treating hepatic tumours. Eur J Nucl Med 1996;23:947-52.  158.  Shcherbinin S, Piwowarska-Bilska H, Celler A, Birkenfeld B. Quantitative SPECT/CT reconstruction for 177Lu and 177Lu/90Y targeted radionuclide therapies. Phys Med Biol 2012;57:5733-47.  159.  Altman DG. Practical statistics for medical research. United States of America: Chapman & Hall; 1991.  160.  Raoul JL, Boucher E, Rolland Y, Garin E. Treatment of hepatocellular carcinoma with intra-arterial injection of radionuclides. Nat Rev Gastroenterol Hepatol 2010;7:41-9.  161.  Kwekkeboom DJ, Teunissen JJ, Bakker WH, Kooij PP, de Herder WW, Feelders RA, van Eijck CH, Esser JP, Kam BL, Krenning EP. Radiolabeled somatostatin analog [177Lu-DOTA0,Tyr3]octreotate in patients with endocrine gastroenteropancreatic tumors. J Clin Oncol 2005;23:2754-62.  164  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0073590/manifest

Comment

Related Items