THE UNIVERSITY OF OKLAHOMA HEALTH SCIENCES CENTER GRADUATE COLLEGE MONTE CARLO SIMULATION FOR THE MEVION S250 PROTON THERAPY SYSTEM: A TOPAS STUDY

Transcription

1 THE UNIVERSITY OF OKLAHOMA HEALTH SCIENCES CENTER GRADUATE COLLEGE MONTE CARLO SIMULATION FOR THE MEVION S250 PROTON THERAPY SYSTEM: A TOPAS STUDY A DISSERTATION SUBMITTED TO THE GRADUATE FACULTY in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY BY MICHAEL THOMAS PRUSATOR Oklahoma City, Oklahoma 2018

2 MONTE CARLO SIMULATION FOR THE MEVION S250 PROTON THERAPY SYSTEM: A TOPAS STUDY APPROVED BY: Yong Chen, Ph.D., Chair Salahuddin Ahmad, Ph.D. Hosang Jin, Ph.D. Daniel Johnson, Ph.D. Hong Liu, Ph.D. Jagadeesh Sonnad, Ph.D. DISSERTATION COMMITTEE 2

3 COPYRIGHT by Michael Thomas Prusator May 11, 201 3

4 ACKNOWLEDGMENTS Throughout my entire graduate experience, I have had the privilege to work with and learn from a whole host of incredibly bright and talented people. The list of people who have helped me along the way is endless, and it would nearly be impossible to be able to thank everyone who has played a role in my degree. I would first like to thank my advisor, Dr. Yong Chen. Words truly can t express my gratitude. It has been an absolute privilege to work with you for the past six years. You have always been there for me, pushing and helping me to be the best Medical Physicist I can be. I would not be where I am today if I had not had you as an advisor. You are a wonderful mentor in not only the Medical Physics field, but in all aspects of navigating the difficulties of graduate school. Our time together, and your friendship, is something that is invaluable to me. I would also like to thank Dr. Salahuddin Ahmad. From the beginning you have always treated me and all other students with respect and as colleagues. Your wisdom in the academic field, and your leadership in the clinic is remarkable. Learning from you during my time as a student has been an experience that I will take with me throughout my professional career. My gratitude also goes out to Dr. Jagadeesh Sonnad. From the first time I stepped on campus, I was impressed with how student oriented you are. Thank you for always putting the students first, and having their best interest at heart. It has been a true pleasure to be in your Radiological Sciences program at the University of Oklahoma Health Sciences Center. 4

5 To my dissertation committee, Dr. Yong Chen, Dr. Salahuddin Ahmad, Dr. Hosang Jin, Dr. Daniel Johnson, Dr. Hong Liu and Dr. Jagadeesh Sonnad, thank you all for agreeing to be a part of my Ph.D. journey. It has been a wonderful experience to work with each of you and an honor to have your signature on this work. I would also like to offer my sincerest thanks to all of the faculty and staff of the Radiological Sciences department. For my duration in this program I have always been treated with respect and kindness by each of you. The people of this program have become a second family to me, and given me memories and experiences to last a lifetime. There are two more people that have played an integral role in my experience here, and deserve more gratitude than I could ever give. To George MacDurmon and Casey Schmitz, you have both become dear friends to me. Words cannot truly capture how much I have enjoyed my time in the Radiation Safety Office and working with each of you. I have learned so much as a Health Physicist in your office, and I want to thank you for giving me that opportunity. I could not have successfully accomplished this dissertation without the loving support of my family. Mom and Dad, I can never begin to thank you enough for all that you have done for me not only during graduate school but also throughout my life. I would not be the man I am today if it wasn t for both of you. Thank you for always pushing me and encouraging me to never give up, and being the perfect role models to me on how to navigate the challenges of being a professional person, a husband and a father. Ashley, you are a wonderful sister to me. You are always in my corner and you always have my back. I can never thank you enough for that. I can only hope that my children have as strong a sibling relationship as you and I have. I also would like to thank my uncle, Duane 5

6 Myers, who is the reason that I am in the field of Medical Physics. Thank you for all you have done for me through graduate school. You have always given me sound advice and made sure that I had every resource I needed to be successful in getting a start in this competitive field. To the rest of my family, grandparents, aunts and uncles, thank you for always taking an interest in everything I do, you each have had such a profound impact on my life. I also would like to thank my parents-in-law, Sheri and Larry, and my brothers and sisters-in law, Austin, Autumn, Cerik, and Destiny. Each of you have welcomed me into your family as one of your own, and have shared in this journey with me every step of the way. Lastly, to my son Bear and my wife Dawn. Bear, as I am finishing this you are about to turn one year old. You don t know it yet but you have already had a profound impact on my life. Coming home and seeing your smiling face is the highlight of my days. Thank you for being a wonderful son. Dawn, you deserve this Ph.D. as much as I do. You have been there for me every step of the way, through the highs and the lows. You have proofread CVs, personal statements, and papers, helped me through my general exam and listened to me practice presentations all with a smile on your face and while being a wonderful wife and mother to our child. You continue to amaze me every day with your wonderful gifts. I could never thank you enough for all you do, and I don t know how I could have ever done this without you. I love you dearly. 6

7 TABLE OF CONTENTS LIST OF FIGURES LIST OF TABLES ABSTRACT CHAPTER I History of Proton Therapy Benefit of Proton Therapy Basic Proton Interactions Stopping Interactions Scattering Interactions Nuclear Interactions Proton Treatment Delivery Systems Passive Double Scattering Proton Beam Delivery The First Scattering Foil The Range Modulator Wheel Second Scatterer Apertures and Compensators Pencil Beam Dose Calculation Algorithm Machine Output Calculations The Benefit of Monte Carlo Simulation TOPAS MC Code Significance of this Work Statement of Hypothesis and Specific Aims

8 TABLE OF CONTENTS CONTINUED CHAPTER II Abstract Introduction Methods Pristine Bragg Peak Simulations Spread Out Bragg Peak Simulations Lateral Beam Profile Matching Spread Out Bragg Peak Modulation Width Adjustments Results and Discussion Pristine Bragg Peaks Spread Out Bragg Peak Matching Lateral Beam Profile Matching Spread Out Bragg Peak Modulation Width Adjustments Conclusion Disclosure of Conflicts of Interest Acknowledgments CHAPTER III Abstract Introduction Methods Shielding Design Evaluation

9 TABLE OF CONTENTS CONTINUED In Room Neutron Study Measurement Monte Carlo Simulation Results and Discussion Shielding Design Evaluation In Room Neutron Study Measurement Monte Carlo Simulation Conclusion Disclosure of Conflicts of Interest Acknowledgements CHAPTER IV Abstract Introduction Methods Developing the Patient Treatment Simulation Workflow PSF and CSF Study PSF and CSF using the PBA PSF and CSF calculated in MC Phantom Study Evaluation of PSF Evaluation of CSF

10 TABLE OF CONTENTS CONTINUED 4.4 Results and Discussion Patient Treatment Simulation Procedure PBA and MC Comparison PSF and CSF PBA Results PSF Evaluation CSF Evaluation Conclusions Disclosure of Conflicts of Interest Acknowledgements CHAPTER V Summary and Conclusion LIST OF REFERENCES APPENDIX A APPENDIX B APPENDIX C

11 LIST OF FIGURES Figure 1. A depth-dose curve for a mono-energetic proton beam Figure 2. A) The process of production of the forward peaked hadronic cascade from a high energy collision and B) the production of evaporation nucleons from a compound nucleus formed from a low energy collision Figure 3. A representation of a cyclotron. The proton is accelerated in a spiral pattern via an oscillating electric field until the desired energy is reached. The beam is then extracted through a beam extraction foil Figure 4. A representation of a synchrotron. The initial proton is injected into the device and accelerated around the ring until the desired energy is reached. The beam is then extracted through the beam extraction nozzle Figure 5. Schematic of a multi-room traditional proton delivery system. Beam is delivered to a treatment room sequentially from a cyclotron located in a separate vault. (Courtesy of Varian Medical Systems, Inc) Figure 6. Schematic of a single-room compact proton therapy system where the cyclotron is located in room and mounted on a gantry for rotation around the patient (Courtesy of Mevion Medical Systems) Figure 7. Sequence of beamline components in a passive double scattering proton therapy delivery system (18) Figure 8. Mevion S250 first scattering foils. One scattering foil is loaded into the beamline prior to treatment via rotation of the pictured component (Courtesy of Mevion Medical Systems) Figure 9. A composite spread out Bragg peak created from energy modulated Bragg Peaks Figure 10. A RMW from the Mevion S250 proton therapy system. Rotation of this component during irradiation creates the SOBP Figure 11. A rotating RMW creates a succession of Bragg peaks to create an SOBP (Courtesy of Mevion Medical Systems) Figure 12. A common second scatterer. (24) The low-z material is in blue and the high-z material is in green. This component will spread and flatten the proton beam Figure 13. Typical patient specific brass aperture and Lucite range compensator used in passive scattering proton therapy

12 LIST OF FIGURES CONTINUED Figure 14. An overview of the pencil beam algorithm Figure 15. The configuration of beam line components for (A) the large group of configurations, (B) the deep group of configurations and (C) the small group of configurations. Where FS is the first Scattering foil, RMW is the range modulator wheel, AB is an energy degrader, SS is the second scattering foil, PA is the post absorber, IC are the ion chambers, SN is the snout and AP is the aperture Figure 16. Individual peaks created from the method described in the text to be summed to create the SOBP. Each peak was assigned a specific weighting factor that, when summed together with the other peaks, created a flat SOBP shown by the dotted green curve Figure 17. Measured data from commissioning (circles) and simulated (solid lines) normalized pristine Bragg peaks from (A) a large configuration with a range of 22.5 cm, (B) deep configuration with a range of 29.5 cm and (C) a small configuration with a range of 15.3 cm Figure 18. Measured data from commissioning (circles) and simulated (solid lines) normalized spread out Bragg peaks for (A) large configuration (configuration 1) with a range of 25 cm, (B) deep configuration (configuration13) with a range of 32 cm and a (C) small configuration (configuration 20) with a range of 15.3 cm Figure 19. Measured data from commissioning (circles) and simulated (solid lines) normalized lateral beam profiles from a (A) large configuration (configuration 1) at a depth of 10 cm, (B) deep configuration (configuration 13) at a depth of 20 cm and (C) small configuration (configuration 20) at a depth of 10 cm Figure 20. A) The stop pulse curve for configuration 21. The curve will give the appropriate stop pulse to use when adjusting the SOBP width for simulations. Linear interpolation is used to determine the fraction of the last stop pulse to be applied when the desired modulation width is between pulses. This curve was used to provide the stop pulse to simulate the reference beam at our institution. B) The measured data from commissioning (circles) and simulated (solid lines) normalized spread out Bragg peak for the reference SOBP Figure 21. The locations and summary of the eight critical point locations where neutron doses were calculated and measured Figure 22. Representation of the points of measurement around the phantom. d represents the distance of the detector from the proximal phantom surface. Each tic mark represents a point where a measurement was taken

13 LIST OF FIGURES CONTINUED Figure 23. Comparison of measured H/D versus distance at an angle of 90 degrees between the large configuration with a 20x20 cm 2 field, and the small and deep configurations with a 10x10 cm 2 field Figure 24. Comparison of H/D versus angle at a distance of 100 cm between the large configuration with a 20x20 cm 2 field, and the small and deep configurations with a 10x10 cm 2 field Figure 25. Measured H/D as a function of field size for the three configurations at a distance from isocenter of 100 cm Figure 26. Comparison of simulated and measured H/D values at an angle of 90 degrees around the water phantom for the large option Figure 27. Comparison of simulated and measured H/D values at an angle of 90 degrees around the water phantom for the small option Figure 28. Comparison of simulated and measured H/D values at an angle of 90 degrees around the water phantom for the deep option Figure 29. Proposed workflow for conducting patient treatment Monte Carlo simulations Figure 30. The conversion curve used by TOPAS to calculate material density for each HU value in a CT data set Figure 31. Look-up table of density to atomic composition developed by Schnieder et al. used by this study to convert HU values to atomic composition (74) Figure 32. The sequence of steps for STL file creation of A) compensator and B) Aperture (76) Figure 33. An example of the PSF and CSF calculation points using the treatment planning system verification plan tool for a brain case at a WED of 5 cm. A) Shows the calculation point in the patient, B) shows the calculation in the phantom via the verification plan tool with the compensator in place and C) shows the same calculation as B without the compensator Figure 34. A) Side by side comparison of the TPS (left) and TOPAS simulated (right) dose distributions for the IROC lung phantom. B) displays the gamma analysis with points in green taken to be passing. (76)

14 LIST OF FIGURES CONTINUED Figure 35. Visualization and dose comparison of a A) prostate, B) pancreas, C) lung and D) brain site between MC and TPS. The top dose distributions are from the TPS and the middle dose distribution are those calculated from MC. The lowest row is the gamma analysis where the points in yellow and red are the failed points. (79) Figure 36. A histogram of CT values in the treatment field for Patient S Figure 37. The relationship between the PBA calculated PSF and the percentage of air in the treatment field for the lung cases Figure 38. A histogram of CT values in the treatment field for Patient K Figure 39. The relationship between the PBA calculated PSF and the percentage of dense bone in the treatment field for the prostate cases Figure 40. An example of a beam crossing an air pocket caused by a gap between a patient immobilization device and the surface of the patient Figure 41. The relationship between the IQR of the compensator thickness and the calculated CSF

15 LIST OF TABLES Table 1. Summarization of the three beam configuration groups used in the Mevion S Table 2. Range and modulation widths of the three SOBP beams chosen for the simulation Table 3. Range and depths of the three configurations chosen for lateral beam profile calculation Table 4. Absolute differences between measured and simulated percent depth dose data for the large group Table 5. Absolute differences between measured and simulated percent depth dose data for the deep group Table 6. Absolute differences between measured and simulated percent depth dose data for the small group Table 7. Comparison of measured and simulated SOBP for configurations 1, 13 and Table 8. Comparison of penumbra, flatness, symmetry and full width at half maximum for measured and simulated cross profiles for two configurations in each group Table 9. Beam characteristics of the three configurations chosen for in room measurements Table 10. Annual neutron equivalent doses calculated from the shielding design and measured during evaluation for each of the 8 critical points Table 11. HU values of each electron density phantom plug and their corresponding material density Table 12. Gamma passing rates of the comparisons between MC, TPS and film measurement for the axial, coronal and sagittal planes of the IROC lung phantom Table 13. A comparison of the CSF and PSF factors calculated in both MC and the PBA in the TPS Table 14. The PBA calculated CSF and PSF factors for each patient plan

16 ABSTRACT Proton therapy is becoming increasingly popular as a cancer treatment modality. This modality of treatment has the potential to achieve greater healthy tissue sparing and higher target conformity than traditional photon treatments. The University of Oklahoma Health Sciences Center operates the Mevion S250 compact proton therapy unit. This passive double scattering system is the first of its kind, offering a single-room compact design where the cyclotron is mounted on a gantry and located in the treatment room with the patient. The goal of this study is to conduct a full characterization of the Mevion system including general dosimetry, in-room neutron studies and patient specific treatment dosimetry studies using the Monte Carlo simulation toolkit TOPAS (TOolkit for PArticle Simulation). The initial step in this work was to accurately build the Mevion system in its entirety in the simulation space. Each of the 24 options, including 14 range modulator wheels, 3 second scattereres, 8 first scattering foils, 24 range absorbers and 24 post absorbers, were modeled according to manufacturer provided blueprints. To ensure that the Mevion simulation could accurately characterize the radiation field exiting the beam nozzle, a benchmark study was conducted comparing simulated data to real beam data taken during commissioning. Pristine Bragg peaks, spread out Bragg peaks and lateral profiles were simulated for the beam options and compared to the analogous commissioning data. There were excellent agreements between the two data sets, where the distal depth of 90% all matched within 1 mm, the distal 80%-20% matched within 2 mm for each beam option, spread out Bragg Peak widths all matched within 3 mm, and the flatness and symmetry of the lateral profiles agreed within 1% of the beam commissioning data. 16

17 Using the benchmarked simulation model of the Mevion system, an in-room neutron study was conducted. The presence of the cyclotron in the treatment vault creates a potential for elevated neutron contamination dose to the patient during treatment. This potential was investigated by irradiating a water phantom in simulation using the deepest, largest and smallest beam option, and scoring neutron energy fluences produced at distances of 20, 40, 60, 80, 100 and 150 cm from isocenter, and at angles of 0, 45, 90 and 135 degrees relative to the beam path around the water phantom. The influence of field size on neutron production was also studied, where at the same locations neutron dose equivalents were scored for field sizes ranging from 0 x 0 to 26.5 x 26.5 cm 2. The acquired neutron energy fluences were combined with ICRP report number 74 fluence to ambient dose equivalent conversion factors to calculate the final neutron dose equivalent per Gy of proton delivered to isocenter. The calculated neutron dose equivalents were also compared to measured neutron dose equivalents using the SWENDI-II neutron detector. Calculated neutron dose equivalents ranged from 8 msv/gy to less than 1 msv/gy. Measured neutron doses were lower, in some cases by an order of magnitude. The data shows the neutron dose equivalent generally decreases as the distance from isocenter increases. The neutron dose equivalent additionally increases along with an increased angle around the water phantom. The field size also influences neutron production, as smaller field sizes tend to increase the amount of contamination neutron dose. One of the more prominent roles Monte Carlo simulations play in radiation therapy is in patient specific dosimetry. In this study, a clear workflow was developed to simulate a patient treatment and compare the 3D dose distributions to that calculated in the treatment planning system. Beginning from loading the specific patient beam parameters and patient 17

18 CT data set into the simulation space, to a final visualization and comparison of the 2D doses calculated from simulation and the treatment planning system, this work successfully demonstrates the ability to conduct patient specific applications. Using this workflow, a study of the effects of patient anatomy and range compensator on the dose per monitor unit calibration was conducted. The patient scatter factor (PSF), defined as the ratio of dose in patient to that in a homogeneous phantom, and compensator scatter factor (CSF), defined as the ratio of dose in phantom both with and without the compensator in place, are two terms largely ignored in monitor unit calibrations due to the difficulty of physical measurement. PSFs and CSFs were calculated for 8 brain, 8 lung, 11 pancreas and 11 prostate fields using the pencil beam algorithm treatment planning system. PSFs and CSFs from a sample of 3 fields from each site were also simulated using the developed workflow and compared to their treatment planning system counterpart. For the prostate and lung cases, the tissue type in the treatment field was quantitatively analyzed based on their HU values within the 50% isodose contour for each field and then plotted against the corresponding PSF value. The interquartile range (IQR) of compensator thickness values within a small region of interest along the central beam axis was calculated to evaluate the impact of compensator shape on the CSF. The range of calculated PSF and CSF values for all cases were found to be and , respectively. The average differences in CSFs between simulated fields and treatment planning system calculated fields were below 0.5%, and the average differences in PSFs for the brain, pancreas, prostate and lung were 1.6, 1.0, 0.6 and 1.9%, respectively. The largest deviations were in cases where large heterogeneities are present in the beam path. A linear relation between the quantity of air or bone and the 18

19 corresponding PSF for the treatment field was found for lung and prostate cases. Likewise, there is a proportional relationship between the IQR value of the compensator and the CSF, where the wider ranges of thickness values indicate an irregular compensator shape, thus increasing the value of the CSF. This study is the first to model the Mevion S250 accurately and in full. Further significance of this work comes from the first characterization of the in-room neutron signature of the treatment system in a clinical setting. Furthermore, the development of a simulated treatment workflow and investigation into the PSF and CSF using the Mevion unit in this study will help to deliver the desired dose to the patient with greater accuracy during proton therapy treatments involving a passive scattering system. 19

20 CHAPTER I INTRODUCTION 1.1 History of Proton Therapy Since being first proposed as a modality of radiation therapy in 1946 by Robert Wilson (1), proton therapy has steadily increased in popularity. The first human treatment utilizing proton therapy occurred at the University of California, Berkley, in Treating mainly pituitary tumors, thirty patients were treated from (2) Over the next several decades, ten more facilities converted from physics laboratories would begin using protons to treat cancer. In 1990, the first hospital based proton therapy site was built at Loma Linda, California. (2) There are currently sixty-three total proton therapy facilities worldwide, with twenty-five of them in the United States alone. Increasing from only twenty-eight sites in 2006, proton therapy is increasing in prominence and will likely continue to grow, with forty additional sites currently under construction. (3) 1.2 Benefit of Proton Therapy The nature of dose deposition from a proton beam differs from conventional photon beams in that as the proton traverses a medium, there is little energy deposited until the particle reaches the end of its track. Here, the proton rapidly deposits its remaining energy, causing a sharp increase in the dose delivered. (4) This rapid dose deposition is called a Bragg peak, shown in Figure 1. The Bragg peak enables healthy tissue sparing via a low dose region proximal to the target, and negligible dose distal to the target. This high conformity acquired through proton therapy can lead to lower instances of side effects, and a decrease 20

21 in the risk of secondary cancers, especially in pediatric patients when compared to photon (5, 6) treatments. Figure 1. A depth-dose curve for a mono-energetic proton beam. 1.3 Basic Proton Interactions Charged particles such as protons can undergo three different types of interactions when traversing medium. These three different categories are stopping, scattering and nuclear interactions. (4) The factors dictating which interaction takes place are the coulombic forces of the projectile proton, the atomic electrons of the medium, the atomic nuclei of the medium and the distance at which the projectile proton passes the atom Stopping Interactions Proton stopping interactions result from the coulombic interaction between the projectile proton and the atomic electrons. Stopping interactions are the main component of dose 21

22 deposition in medium from a beam of protons. These interactions can be grouped into two categories, soft collisions and hard collisions. (4) Soft collisions occur between the incoming proton and an atomic electron when an appreciable distance is between the proton and the atom. The influence of the proton s columbic field excites the electron to a higher energy level, resulting in an energy transfer on the order of a few electron volts (ev). Soft collisions are the most probable charged particle interaction, due to the large spacing of atoms within a target medium. Nearly half of the total energy delivered to medium occurs from soft collisions. (4) Hard collisions occur between the incoming proton and an atomic electron when the distance between the proton and atom is on the order of the atomic radius of the target atom. The proton s coulombic field interacts strongly with the coulombic field of the atomic electron, transferring enough energy to eject the electron from the atom, resulting in ionization of the target atom. The ejected electron, deemed a delta ray, acquires enough kinetic energy to undergo additional coulombic interactions of its own. Though hard collisions are more infrequent than soft collisions, the amount of the projectile proton s energy spent between the two processes is nearly equal. (4) The rate of energy loss of the incoming proton as a function of path length (stopping power) due to stopping interactions follows the Bethe-Bloch formula, shown in Equation 1. (7) de dx = Kz2 Z A c 2 v 2 [1 ln 2m ec 2 β 2 γ 2 T max 2 I 2 β 2 δ(βγ) 2 ], Eq. 1 Where de is the stopping power, z is the charge of the proton, Z is the atomic number dx A divided by mass number of the target atom, T max is the maximum energy transfer to the 22

23 electron, I is the mean excitation energy of the target atom, δ(βγ) is a shell correction factor and v is the velocity of the proton. The inverse proportionality of stopping power to v 2 of the incoming proton is responsible for the shape of the Bragg peak Scattering Interactions When an incoming proton passes near a target atom s nucleus, the positively charged coulombic fields of the proton and nucleus interact, causing a deflection of the proton from its original path. In most cases, scattering interactions result in many small angle deflections of the incoming proton. (8) Mathematically, the angle of deflection is described by the General Highland Formula, shown in Equation 2. (9) θ = 14.1 MeV pv L [1 + 1 L log ( )], L R 9 L R Eq. 2 Where p is the proton momentum, v is the proton velocity, L is the target thickness and L R is the target radiation length. According to the General Highland Formula, the total angle of deflection from the original path that a proton undergoes will increase as the energy of the particle decreases, causing larger scattering at the end of the proton s track length Nuclear Interactions When an incoming proton has sufficient kinetic energy ( 100 MeV) it may interact inelastically with the target atom s nucleus. (4) For higher energy protons, the collision with the atomic nucleus results in an interaction with individual nucleons. This process results in a spray of forward peaked fragments of the former nucleus, called a hadronic cascade. (10) 23

24 Hadronic cascades produce high energy protons and neutrons, which then further propagate the fragmenting of other atomic nuclei. For interacting protons of relatively lower energy, the proton enters the target atomic nuclei, forming a compound nucleus. Compound nuclei are unstable and will decay rapidly, isotropically emitting secondary evaporation particles including gamma rays, neutrons and protons. The evaporation nucleons emitted are relativley lower in energy than those emitted through the hadronic cascade process. (4) Both processes result in the production of secondary particles that will have an effect on the dose distribution delivered from a proton beam. Hadronic cascade and evaporation nucleon production are depticted in Figure 2. 24

25 A) B) Figure 2. A) The process of production of the forward peaked hadronic cascade from a high energy collision and B) the production of evaporation nucleons from a compound nucleus formed from a low energy collision. 1.4 Proton Treatment Delivery Systems Typical nominal treatment energies in proton therapy can reach up to 250 MeV. To achieve this energy, treatment delivery systems utilize one of two different types of accelerator: A cyclotron or synchrotron. A cyclotron is a cylindrical accelerator divided into two sections called dees. Perpendicular to the dees is a static magnetic field. The dees are electrically polarized by 25

26 a square radiofrequency wave. The frequency of the wave matches the exact transit time it takes for the proton to travel around the dee. The oscillating electric field between the dees accelerates the proton, while the perpendicular magnetic field directs the particle in a circular path traveling between dees. After the proton achieves the nominal energy, it is directed out of the cyclotron via an extraction foil at the cyclotron exit. (11) Figure 3 shows a drawing of a typical cyclotron. Figure 3. A representation of a cyclotron. The proton is accelerated in a spiral pattern via an oscillating electric field until the desired energy is reached. The beam is then extracted through a beam extraction foil. Cyclotrons are separated in two categories, depending on how the device accommodates the relativistic mass gain of the proton as it is accelerated. In an isochronous cyclotron, the particles revolve around the accelerator at the same frequency. The relativistic mass gain is accommodated via an increase in magnetic field strength as the proton orbit radius 26

27 increases, enabling the delivery of a continuous beam. (11) In contrast, a synchrocyclotron uses a varying electric field frequency to synchronize the proton s orbit frequency with the oscillation in polarization of the dees. In this case, the magnetic field strength is constant across the radius of the cyclotron. Synchrocyclotrons are only able to deliver a pulsed proton beam. (18) In a synchrotron, a proton is injected into an evacuated narrow tube in the shape of a ring. The ring is lined with magnets dedicated to bending the proton along the circular path around the tube (see Figure 4). The proton is accelerated via passage through an RF cavity in the ring. The frequency of the RF voltage is tuned to match the period of the proton around the ring. The strength of the magnetic field and the RF frequency are increased in synchrony as the proton gains energy and relativistic mass. Once the proton acquires the desired energy, it is extracted from the ring. (11) 27

28 Figure 4. A representation of a synchrotron. The initial proton is injected into the device and accelerated around the ring until the desired energy is reached. The beam is then extracted through the beam extraction nozzle. Traditional proton delivery systems include one accelerator that provides the proton beam to several different treatment rooms (Figure 5). The accelerator is located in a separate shielded location, and beam is directed by bending magnets to a particular treatment room through a complex beam transport system. In this system, the proton beam is delivered to only one room at a time. 28

29 Cyclotron Treatment Rooms Beamline Figure 5. Schematic of a multi-room traditional proton delivery system. Beam is delivered to a treatment room sequentially from a cyclotron located in a separate vault. (Courtesy of Varian Medical Systems, Inc) The major drawback to traditional proton therapy delivery systems is the cost. (12) It has been identified that the complex beam delivery system found with these systems increases the maintenance cost to a level that may inhibit growth of radiation oncology facilities that offer proton therapy as treatment modality. (13) This has led to the development of compact, single room proton therapy systems, such as the Mevion S250, shown in Figure 6. These units have an in-room cyclotron mounted on a rotating gantry that delivers therapeutic treatment beam directly to the patient, without the use of a complex beam transport system. Cost advantages from single-room proton systems stem from the smaller space needed to house a unit, reducing the construction and installation cost. When compared to traditional 29

30 systems, the simplicity of compact units also reduce maintenance costs, due to the elimination of complex beam transport systems. (14) Figure 6. Schematic of a single-room compact proton therapy system where the cyclotron is located in room and mounted on a gantry for rotation around the patient (Courtesy of Mevion Medical Systems) 1.5 Passive Double Scattering Proton Beam Delivery When the proton beam exits the accelerator, it is a nearly mono-energetic pencil beam with a diameter of only several millimeters. (15) Modulation of the beam takes place prior to its delivery to the patient to achieve the most uniform dose distribution across the treatment target. As a target is rarely the size of the Bragg peak, or the width of a pencil beam, the initial beam is spread to a reasonable field size, and the Bragg peak broadened to cover the target volume. A popular technique to achieve beam modulation is passive beam double (16, 17) scattering. 30

31 Passive beam double scattering systems make use of two beam scatterers to spread the beam to a useful treatment size, a range modulator wheel (RMW) to broaden the Bragg peak, and a patient specific aperture and compensator to shape the final field both laterally and distally to the target volume. (17) Figure 7 shows the typical sequence of the scatterers and RMW in a passive double scattering system. Figure 7. Sequence of beamline components in a passive double scattering proton therapy delivery system (18) The First Scattering Foil The first component in the beamline of a passive double scattering system is the first scattering foil. This foil is a static component, placed in the beamline prior to delivery of the therapeutic proton beam. It is fabricated from a high-z material, usually lead, that delivers a large scattering potential. In most cases, a proton system will have many different first scattering foils of varying thickness, mounted on a wheel for simple 31

32 movement into the beamline (Figure 8). While traditional systems may offer around nine first scattering foils, newer compact systems offer as many as twenty-one. (19, 20) To achieve a larger beam spread, a thicker first scattering foil is placed in the beam line. Scattering foil thicknesses range from 0.5 cm 2 /g for small field sizes, up to 3 cm 2 /g for the largest field sizes. Figure 8. Mevion S250 first scattering foils. One scattering foil is loaded into the beamline prior to treatment via rotation of the pictured component (Courtesy of Mevion Medical Systems) The Range Modulator Wheel After the first scattering foil, the beam will pass through the RMW. The RMW spreads the initial Bragg peak out to a useful size by temporally modulating the beam energy, producing a sequence of Bragg peaks that form a composite spread out Bragg peak (SOBP)(Figure 9). (21) 32

33 Figure 9. A composite spread out Bragg peak created from energy modulated Bragg Peaks The RMW resembles a spiral staircase, varying discretely in thickness on each step around the central hub (Figure 10). A low-z material comprises most of the component, serving to reduce the beam energy for a specific step of the wheel without scattering the beam. The steps with a smaller thickness have a small amount of high-z material present to give each step uniform scattering power. A single proton delivery nozzle may have multiple RMWs, (19, 20) ranging from three to fourteen. 33

34 Figure 10. A RMW from the Mevion S250 proton therapy system. Rotation of this component during irradiation creates the SOBP. RMWs are a dynamic component that continuously rotate during the delivery of the beam to the patient. The beam that strikes the thick steps of the RMW experience more range loss and deposit dose at a shallower depth in the patient. The beam that strikes the thin steps of the RMW experience less range loss, depositing dose at a deeper depth in the patient. Taken together, the full SOBP is created to cover the target volume, see Figure

35 Figure 11. A rotating RMW creates a succession of Bragg peaks to create an SOBP (Courtesy of Mevion Medical Systems) Second Scatterer The second scatterer is a contoured bi-material device used to further spread and flatten the beam, shown in Figure (12). (19) The high-z material is of a Gaussian shape, with a greater thickness in the center of the beam to spread the fluence of protons toward the field borders. The low-z material is in inverse shape to the high-z material, with the thickest area being towards the edge of the component. (22) The low-z material serves to uniformly scale the range of protons throughout the treatment field. When the high and low-z material are combined, the second scatterer has a uniform water equivalent thickness. (23) Many proton therapy systems will have multiple second scatterers, either located on a carousel or shuttle, which can be easily seated in the beam path to designate the maximum field size. The second scatterer is a static component and is placed in the beamline prior to treatment delivery. 35

36 Figure 12. A common second scatterer. (24) The low-z material is in blue and the high-z material is in green. This component will spread and flatten the proton beam Apertures and Compensators Apertures and compensators are the last components of a passive scattering system the beam passes through before delivery to the patient (Figure 13). These patient specific devices are used to shape the beam laterally and distally to conform to the target dimensions. (25) The aperture, usually constructed from brass, is upstream of the compensator and is used to shape the field to the beam s eye view of the target by collimating the beam outside the desired field. The compensator is used to correct for irregularities in the target shape and water equivalent thickness from the distal end of the target to the patient surface. (26) Compensators are usually constructed of a Lucite, and the cross-sectional thickness varies along the component, scaling the ranges of the beam within the field to match the depth of the distal edge of the target. (27) 36

37 Figure 13. Typical patient specific brass aperture and Lucite range compensator used in passive scattering proton therapy 1.6 Pencil Beam Dose Calculation Algorithm A popular method of calculating dose to a patient in proton therapy is the pencil beam dose calculation algorithm (PBA). The algorithm approximates a full field through the convolution of many individual pencil beams. (28) The patient dose at any one point is calculated as the dose summed from a contribution from each of the pencil beams. The full algorithm can be separated in two parts: the characterization of a single pencil beam; and the summed contribution over all pencil beams in the field. (28) The characterization of a single pencil beam is separated into a central axis dose (C(z )) and an off axis dose (O(x, y, z ), thus the dose, d, from one pencil beam is: d(x, y, z ) = C(z ) O(x, y, z ), Eq. 3 37

38 The central axis dose data is collected from a measurement of the broad beam depth dose in a water phantom. (28) The off axis dose is taken to be a lateral distribution resulting from the scattering of protons in the treatment beamline and patient. This distribution is modeled as a Gaussian function, where the standard deviation is calculated from summing in quadrature the individual contributions of scatter from each beamline component and patient. The off axis dose contribution from the patient is taken from the General Highland Approximation. (29) The full dose distribution from a broad beam proton treatment field is solved as a convolution of all pencil beams within the field. At a particular point of interest, the total composite dose will be a summation of the primary dose from the protons originating directly from the source (the central axis dose) and a scatter dose described by the off-axis dose from each pencil beam. Figure (14) gives a summary of the pencil beam algorithm. Figure 14. An overview of the pencil beam algorithm The PBA performs well in several areas: predicting the widening of field penumbra width as a function of depth in patient, modeling the degradation of distal falloff in dose deposition in heterogenous medium, and predicting hot and cold dose regions in a treatment 38

39 plan. (30) The algorithm has an appreciable limitation, however, when modelling scatter in inhomogenous media, particularily at low density interfaces. (31-33) The PBA also fails to take into account nuclear interactions and large angle scattering. (29) Taken together, these limitations can lead to target underdosing by as much as five percent. (34) 1.7 Machine Output Calculations The output of the proton treatment system delivering a given amount of radiation dose to a patient is specified in monitor units (MU). An MU is defined in the treatment machine as a specific quantity of charge created in monitor chambers located in the beamline. The charge is calibrated to a known dose delivered to a phantom under reference conditions. (35) As the reference conditions are rarely treatment conditions, the determination of MU to be delivered to reach a prescribed radiation dose becomes a necessary task in planning a radiation treatment. (36) Current treatment planning software is not able to determine the MU required for patient treatment. Therefore, it is desirable to have a function able to calculate MU for a treatment field. To date, there is no standard MU prediction model for proton therapy. (37) There are however, two generally accepted methods of calculating MU for a patient treatment field. Sahoo et. el. proposed a correction based formula that applies correction factors to the reference calibration to adjust for varying ranges, SOBP widths etc. (36) The second method, derived by Kooey et al. calculates the MU needed to deliver a desired dose analytically, relating the entrance dose of a field to the dose at the center of the SOBP. (35) Comparisons and implementation of these methods can be found in (37, 38) literature. The Sahoo formula, shown in Equation 4, assumes that the necessary parameters that need to be taken into account in a passively scattered proton beam are the energy, lateral 39

40 scatterers, RMW, SOBP width, range shifter thickness, the depth dose value relative to the normalization point in the SOBP, and scatter from the patient and compensators. ( d ) = ROF SOBPF RSF SOBPOCF OCR FSF ISF CPSF, MU Eq. 4 The ROF, SOBPF and RSF are the major contributing factors to the change in dose per monitor unit for a given proton beam. ROF is the relative output factor and corrects for the changes in beamline components for a specific nozzle configuration. SOBPF is the spread out Bragg peak factor and accounts for changes in output based upon SOBP width. Often in passive scattering proton therapy, a low Z range shifter is introduced after the second scatterer to pull back the range of the beam. The RSF is the range shift factor that corrects for changes in output due to the presence of a range shifter in the beam. The SOBPOCF is the SOBP off center factor and accounts for changes in dose per MU with the location of the measurement point longitudinally along the SOBP. Similarly, the off center ratio, OCR, accounts for changes in output when the location of measurement changes laterally across the field. The field size factor, or FSF, accounts for changes in output with changes in the field size. ISF is the inverse-square factor and corrects for changes in source-to-point-of-mu calibration distance. Finally, the CPSF is the compensator-patient scatter factor and accounts for changes in output due to scatter from the patient specific compensator and the patient anatomy, taken to be unity. The correction factors used in the Sahoo formula are determined experimentally through measurement by holding all conditions constant, and varying one parameter sequentially until a table of known values for each factor is acquired. When using this technique to calculate MU, 40

41 agreement within two percent between measured output to calculated output can be achieved in a water phantom. (36) In contrast to the Sahoo method, the Kooy formula (Equation 5) predicts output based upon first principles, using the observation that the output factor is the ratio of SOBP plateau dose to the dose measured in the monitor ion chamber. (35) This analytical technique predicts the output based on a single factor, r, which is derived from the range and modulation width of the user s beam. d MU = CF Ψ c D c 100 (1+a 1 r a 2), Eq. 5 In Equation 5, the CF is similar to the ROF in the Sahoo method, accounting for relative output changes due to different beamline configurations. Ψ c is the machine output under reference conditions and D c is the entrance dose under reference conditions. The term r can be expanded to (R M M), where R and M are the range and modulation width of the given beam. The terms a 1 and a 2 are fitting parameters used to fit the function to observed data acquired from measurement. This output calculation method does not take into account patient or compensator effects on output. The Kooy model predictions agree with measured output to just under three percent, when used in a water phantom. (35) 1.8 The Benefit of Monte Carlo Simulation Monte Carlo (MC) simulation has become very important in proton radiation therapy. (39) In many cases, it is treated as the gold standard in dose calculation computation. This is due to its consideration of all physics processes during particle interactions. (31) Simulation also accurately models tissue inhomogeneities by taking into account specific material 41

42 properties, such as elemental composition, mass and electron densities and ionization potentials. (40) This allows for better modeling of nuclear interactions and scatter. As a proton treatment beam has a finite range, the greatest impact of MC simulation in calculating a dose distribution may be in the distal fall off region. (40) In a study done by Sorriaux et al., MC simulation outperformed the PBA when lateral inhomogeneities are present. The gamma index was 52.7% higher for MC under a 2%/2mm condition when compared to a measured dose distribution. (41) This enhanced ability to calculate dose has led to several studies involving the use of a simplified MC method in the treatment planning (33, 42) process. MC simulation models radiation transport exceptionally well and is utilized in a wide range of study. (43) In a variety of cases, MC is the only tool for patient specific applications, where it is not possible to take physical measurements. This includes machine output calculations (44, 45) and the impact of secondary neutron contamination from nuclear cascades during (46, 47) treatment TOPAS MC Code In order for the full potential of MC simulation to be realized in proton therapy, there must be an accurate characterization of the radiation field exiting the treatment system. This requires meticulous modeling of the beamline geometry in the treatment system. Currently, MC simulation is underutilized due to the complexities of passive scattering components such as the shape of the RMW. (48) All-purpose MC codes such as GEANT 4, FLUKA and MCNPX require extensive programming knowledge and are not suited well to the inexpert user for medical physics applications. (48) 42

43 TOPAS (TOlkit for PArticle Simulation) is a recently developed MC simulation toolkit built on the GEANT 4 platform. TOPAS is designed to be a user friendly code with an emphasis on making proton therapy simulation available to the inexperienced programmer, thus facilitating MC use in proton therapy research. (48) The parameter system is simple in comparison to the extensive C++ language used in GEANT 4. Because TOPAS is layered neatly on top of the GEANT 4 platform, the TOPAS user enjoys the same accuracy and reliability associated with the native GEANT 4. (48) In a recent study, TOPAS was shown to accurately reproduce measured SOBPs, lateral dose profiles and absolute dosimetry when compared to measurement of machine output factors. (49) The conclusion from this study was threefold: 1) That TOPAS can be applied to quality assurance for proton therapy, 2) TOPAS can be a tool used for commissioning of a commercial treatment planning system and 3) TOPAS can also be used as a basis for performing routine clinical dose calculations. (49) 1.9 Significance of this Work This work aims to accomplish several goals. The first is to accurately model the Mevion S250 proton therapy system recently commissioned at the University of Oklahoma Health Sciences Center (OUHSC) in the TOPAS MC code. Due to the complexities of proton therapy MC, little work has been done to model a traditional proton therapy delivery system (19, 50) and no work has been done to model a compact proton therapy system such as the Mevion S250. Having an accurate simulation model of the Mevion S250 will enable the facilitation of use for patient specific research, beam commissioning and treatment plan dose calculation, which is currently unavailable for the Mevion system. 43

44 The next goal is to validate the simulation model against measured beam data collected during commissioning of the Mevion system at OUHSC. To enable use of the modeled system, the code must be validated and benchmarked, as omitting this crucial step will invalidate any study or results conducted using the simulation. The benchmarking process will require that measurements of Bragg peaks, SOBPs and lateral profiles be collected from the system. Simulations using the model will be conducted and compared to measurements under matching conditions. Due to their highly penetrating nature, neutron contamination is always a concern to patients undergoing proton therapy. (51) The presence of the cyclotron in the room is an aspect of compact proton therapy units which adds a new dimension to neutron production that has not been fully investigated. This work will be the first to fully characterize the neutron field around the patient during treatment using both measurement and the benchmarked MC model. The final goal of this work is twofold: To develop a workflow for patient specific MC studies and to use this workflow to investigate the effects of compensator and patient anatomy on MU calculation. As previously noted, MC is usually treated as the gold standard in patient dose calculations; however, conducting patient specific MC simulations is complex, and few groups have the capability. This work will provide the necessary foundation to give OUHSC this prestigious tool. Using the developed workflow, a patient specific study involving compensator and patient anatomy effects on MU calculations will be conducted. These two factors are largely ignored in MU output calculations, due to the restrictions of measuring output factors with respect to patient geometry. (35) MC simulation 44

45 allows for the determination of dosimetric impact due to the compensator and patient in the beamline. Taken together, this work will be a full scale, end-to-end MC simulation study involving the Mevion S250; starting from modeling the system and ending with a patient specific study. It is expected that this work will facilitate further research involving the Mevion system, as well as provide a useful dose calculation tool in the clinic that will directly enhance patient treatment Statement of Hypothesis and Specific Aims We hypothesize that TOPAS will enable us to accurately model our proton therapy system to a point where our simulated beam data will characterize the true radiation field produced by the Mevion S250. The increased understanding obtained from the neutron study we propose will enhance patient treatments and proton shielding design. It is also believed that the developed treatment simulation workflow will allow the simulation of full patient treatment plans, and that the study involving the effects of patient anatomy and range compensator on machine output will lead to more accurate MU calculations. 45

46 Specific Aim 1: To model the Mevion S250 in the TOPAS Monte Carlo code TOPAS will be installed. The blueprints for the Mevion S250 will be acquired and each of the beam line components will be modeled exactly to the specifications described by the blueprints Specific Aim 2: To validate the model against measured beam data Bragg Peaks, SOBPs and lateral beam profiles will all be calculated in simulation and compared to the commissioning data Specific Aim 3: To perform a comprehensive neutron study using MC simulation and measurement In room neutron ambient dose equivalents will be calculated around the patient using MC simulation. Measurements will then be made at locations outside the treatment vault and around the patient to compare to our MC model. Specific Aim 4: A) To develop a workflow to conduct patient specific studies using MC simulation and B) To determine the impact of patient and compensator heterogeneities on the proton system output during patient treatment Part I) Patient specific computed tomography (CT) data will be loaded into TOPAS and a Hounsfield Unit (HU) to material density and composition curve will be custom created to fit our institution s CT scanner data. (Including export aperture and compensator information from treatment plan file into 3D structure, export treatment plan parameter (gantry angle, SSD, isocenter) into simulation space, export simulated 3D dose information 46

47 into 3D Slicer based platform and perform comparison with calculated dose from treatment planning system). Part II) Patient and compensator scatter factors will be calculated for 21 different cases involving 4 different sites using both Eclipse treatment planning software and MC simulation. Comparison between the two methods of calculation will be performed and a proposed method to incorporate the specific calibration factors into treatment planning will be made. 47

48 CHAPTER II TOPAS SIMULATION OF THE MEVION S250 COMPACT PROTON THERAPY UNIT Michael Prusator, M.S., Salahuddin Ahmad, Ph.D. and Yong Chen, Ph.D. Stephenson Oklahoma Cancer Center, University of Oklahoma Health Sciences Center, Oklahoma City, OK, USA Published in: Journal of Applied Clinical Medical Physics, 2017;18(3):88-95 doi: /acm Reprinted with permission of John Wiley & Sons, Inc. All rights reserved. M Prusator planned and performed simulations, analyzed data, prepared figures and wrote the first draft and multiple revisions of the manuscript. 48

49 2.1 Abstract As proton therapy becomes increasingly popular, so does the need for Monte Carlo simulation studies involving accurate beam line modeling of proton treatment units. In this study, the 24 beam configurations of the Mevion S250 proton therapy system installed recently at our institution were modeled using the TOolkit for PArticle Simulation (TOPAS) code. Pristine Bragg peak, spread out Bragg peak (SOBP) and lateral beam profile dose distributions were simulated and matched to the measurements taken during commissioning of the unit. Differences in the range for all Percent Depth Dose (PDD) curves between measured and simulated data agreed to within 0.1 cm. For SOBP scans, the SOBP widths all agreed to within 0.3 cm. With regards to lateral beam profile comparisons between the measured and simulated data, the penumbras differed by less than one millimeter and the flatness differed by less than one percent in nearly all cases. This study shows that Monte Carlo simulation studies involving the Mevion S250 proton therapy unit can be a viable tool in commissioning and verification of the proton treatment planning system. 49

50 2.2 Introduction Success in radiation therapy is dependent on maximization of the tumor control probability while minimizing the normal tissue complication probability. The characteristics of the proton Bragg peak help to deliver large doses to the target, lower entrance doses proximal to the target and almost no doses distal to the target. (1) Advancements in Monte Carlo simulation and radiation transport calculations have enabled more accurate characterizations of the radiation fields during proton treatments that can benefit patients. (12) The cases where the greatest advantages of proton therapy could be realized (6, 52, 53) are in targets near critical organs and treatments involving pediatric patients. In order for the full potential of proton therapy to be utilized, there is a need for an accurate dose calculation in proton treatment plans. Monte Carlo simulation has traditionally been shown as a prominent method for conducting various research topics that include dosimetric, linear energy transfer (LET), and commissioning studies. (54-56) In a study done by Paganetti et al., it was shown that the modeling of the IBA proton treatment head at the Northeast Proton Therapy Center resulted in simulated data matching with measured beam data within millimeter accuracy for beam range and 3 millimeters for SOBP width. (40) This acceptable tolerance helps generation of detailed simulated beam data for use in commissioning of a treatment planning system. (19) Monte Carlo simulation has also been used to calculate the risk of secondary cancer due to neutron exposure occurring from nuclear reactions in the treatment system on patients that have undergone proton craniospinal irradiation treatments. (57) Paganetti has also reported that Monte Carlo simulations can even improve proton beam range uncertainty by up to 2.2 percent. (31) 50

51 A passive double scattering compact proton therapy unit has recently been installed at our facility. The Mevion S250 is the first proton therapy system of its kind, delivering a pulsed beam with a nominal energy of 250 MeV and utilizing an in-room superconducting synchrocyclotron mounted on a gantry that allows for 180 degree rotation. The compact nature of the unit has many cost and therapy treatment advantages, and eliminates the need for a complex beam transport system. (14) The main beam shaping components in the nozzle of the unit include a lead first scattering foil present at the cyclotron exit for initial beam spread. The beam then passes through a bi-material staircase type range modulator wheel (RMW) consisting of one track made of graphite to modulate the range of each Bragg peak, and a second track made of lead to ensure uniform scattering power over all steps of the wheel. The last step of each wheel has a brass wedge to completely stop the beam. A bi-layer contoured second scattering foil made from lead and Lexan is incorporated downstream of the RMW to further spread and flatten the beam. The last components the beam passes through are a graphite absorber for energy degradation and a post absorber for fine tuning the beam range, followed by two ion chambers to monitor beam output. For more information on the initial clinical experience with the system, the reader is referred to Zhao et al. (14) The Mevion S250 has 24 different beam configurations, divided into large, deep and small groups. The large group utilizes a large beam nozzle with an un-collimated field size of 25 cm in diameter where the deep and small group share a small beam nozzle with the field size of 14 cm in diameter. The distinction between deep and small groups occurs in their range capabilities and modulation widths, all specified in water throughout the text. The deep group has a depth range from 20.1 cm to 32 cm with a maximum modulation width 51

52 of 10 cm, whereas the small group has a shallower depth range of 5 cm to 20 cm but has a maximum modulation width of 20 cm. For any given group, a special beam configuration results from the unique order and combination of different beam line components as summarized in Figure 15. Figure 15. The configuration of beam line components for (A) the large group of configurations, (B) the deep group of configurations and (C) the small group of configurations. Where FS is the first Scattering foil, RMW is the range modulator wheel, AB is an energy degrader, SS is the second scattering foil, PA is the post absorber, IC are the ion chambers, SN is the snout and AP is the aperture. 52

53 In total, the Mevion system offers 12 beam configurations in the large group, 5 for the deep group and 7 in the small group which are composed from 18 first scattering foils, 14 RMWs and 3 secondary scattering foils. Table 1 shows the beam characteristics for all configurations in each group. Table 1. Summarization of the three beam configuration groups used in the Mevion S250. Group Number of Max. Range Min. Range Max. SOBP Min. SOBP Configurations (cm) (cm) Width (cm) Width (cm) Large Deep Small The objective of the this study has been to model the beam line components for all 24 beam configurations of the Mevion S250 system using Monte Carlo simulation in order to accurately characterize the radiation field exiting the treatment head, and benchmark the simulation results with the measured beam data obtained during commissioning of the machine. 2.2 Methods During the commissioning of the Mevion system, pristine Bragg peaks and full SOBP scans were measured with PTW s advanced Markus ion chamber (PTW 34045) in IBA s Big Blue 1 water phantom. The chamber is a parallel plate ion chamber with a sensitive volume of 0.02 cubic centimeters, enabling the detector to accurately measure the sharp 53

54 distal falloff in the proton PDD. The lateral beam profile scans were measured in the same phantom using PTW s proton diode detector (PTW 60020) with a sensitive volume of 0.03 cubic millimeters. The diode detector provided the necessary spatial resolution needed for capturing the sharp penumbra seen in lateral beam profile scans. This specific diode was chosen in part due to its resilience against radiation damage. McAuley et al. showed that a loss of sensitivity as a function of accumulated dose was only 1% per 100 Gy, which is well under the doses delivered during our measurements. (58) TOPAS (TOolkit for PArticle Simulation) version 2.0 was utilized to model the beam delivery system in this study. (48) TOPAS 2.0 is an extension of the GEANT p02. toolkits that was developed specifically as a user friendly proton therapy tool. It has been validated experimentally as a viable choice when tasked with reproducing beam data from a passive scattering proton system. (49) The dimensions and materials for each beam component were modeled in TOPAS based on the data given by the manufacturer. For all simulations, the TOPAS default modular physics list was used, which is composed of TsEmStandardPhysics_option3_WVI, HadronPhysicsQGSP_BIC_HP, G4DecayPhysics, G4IonBinaryCascadePhysics, G4HadronElasticPhysicsHP, G4StoppingPhysics and G4RadioactiveDecayPhysics. The range cut for secondary particles was set to cm, with an energy cut in water of 990 ev, 57.3 kev, 5 kev and 56.6 kev for gamma rays, electrons, protons and positrons, respectively. A water phantom with dimensions of 40x40x40 cm 3, placed downstream of the delivery nozzle, was used Pristine Bragg Peak Simulations For all configurations, the deepest pristine Bragg peak simulations were calculated by passing beam through the full beam shaping geometry including a static RMW with only 54

55 the thinnest step in the beam path. An open ring aperture was located at 180 cm downstream of the source. The diameter of the aperture was 14 cm for the small and deep groups and 25 cm for the large group. The air gap between the bottom face of the aperture to the top surface of the water was set to 10 cm. The detector mesh used to score the dose was 2x2x0.1 cm 3. The face of the detector mesh was large enough to allow for fewer particles to be run achieving a smooth PDD, while the fine resolution in the z direction prevented distortion of high gradient areas Spread Out Bragg Peak Simulations The Mevion proton system uses a pulsed beam through a RMW rotating at a constant speed of 600 rpm to deliver protons to the target. A user defined beam current modulation (BCM) sequence is applied to synchronize and scale the individual beam pulse striking on the rotating RMW to achieve a uniform, flat SOBP. To mimic such process, individual pristine Bragg peaks were created in simulation in a manner similar to Polf et al. (50) by determining the timing and landing locations of the pulses on each step of the wheel. The pulses were weighted so that each step the beam passed through received the same number of protons. For our purposes, each step received 3x10 6 protons, regardless of the number of pulses located on the step. The pristine Bragg peaks from each step of the wheel were scaled by applying weighting factors to create a SOBP that matched to the measured data (see Equation 6). j D(d) = w i p(d) i, i Eq. 6 55

56 Where D(d) is the dose as a function of depth, w i is the weighting factor applied to the ith peak, and p(d) i is the dose distribution associated with the ith pristine Bragg peak. The sum of square errors between measured and simulated data for each point at 0.1 cm depth increments were minimized to give the best agreement between the calculated SOBP and commissioning data (see Equation 7). d 2 SSE = (D(d) m D(d) s ) 2, d 1 Eq. 7 where SSE is the sum of square error, D(d) m is the dose distribution for the measured data and D(d) s is the dose distribution for the simulated data. An example of an SOBP and corresponding pristine Bragg peaks for configuration 13 are shown in Figure

57 Figure 16. Individual peaks created from the method described in the text to be summed to create the SOBP. Each peak was assigned a specific weighting factor that, when summed together with the other peaks, created a flat SOBP shown by the dotted green curve. Three SOBPs (one configuration from each of the group categories) were constructed using this method (detailed information is shown in Table 2). Table 2. Range and modulation widths of the three SOBP beams chosen for the simulation. Configuration Range (cm) Modulation Air gap Aperture Width (cm) (cm) size (cm 2 ) Large x 18 Deep x 8 Small x 8 57

58 2.2.3 Lateral Beam Profile Matching In a passive scattered proton system, scattering foils are put in place to spread the beam to a clinically relevant treatment size. For Mevion S250, the small and deep groups support a maximum field size of 14x14 cm 2, while the large group supports field sizes up to 25x25 cm 2. The spreading of the beam is achieved through the use of two scattering foils (a first scatter foil and a second scatter foil). The second scatter foil is a bi-material foil fabricated from lead and Lexan. The lead functions to further spread and flatten the beam while the Lexan is added to create a constant water equivalent thickness (WET) across the component. Three unique second scatter foils are equipped in the system (one for each group). To ensure that the scattering foils were modeled accurately, lateral profiles were simulated for two different configurations within each group. For each of these beam configurations, lateral beam profiles were calculated at a shallow and deep depth. The detector mesh was 0.1x2x2 cm 2. The base of the aperture was placed 10 cm above the water surface, and isocenter was located at a depth of the range minus half the modulation width of the configuration being simulated. For each simulation, 2.5x10 8 protons were run. Table 3 gives a summary of each of the configurations and depths used to calculate the lateral beam profile. 58

59 Table 3. Range and depths of the three configurations chosen for lateral beam profile calculation Group Range (cm) Shallow Depth (cm) Deep Depth (cm) Aperture size (cm 2 ) Large x 18 Large x 18 Deep x 8 Deep x 8 Small x 8 Small x Spread Out Bragg Peak Modulation Width Adjustments In the clinical applications of a passive scatter system, it is necessary to adjust the modulation size of the SOBP for target coverage. The Mevion system offers the adjustment for the SOBP width in 0.1 cm increments by applying stop digits in the BCM files. These stop digits designate a stop pulse, which reduces or cuts the beam current on a certain pulse of the incident beam. In these cases, a step of the wheel may receive less fluence of protons or none at all. By eliminating or lowering the fluence contribution to the shallower peaks, the modulation width will become smaller. Using a similar idea, adjustment of the modulation width in our simulation can be done through simple scaling of weighting factors for the individual pristine Bragg peaks. For each configuration, we sequentially subtracted one pulse from our fully modulated beam and calculated the new SOBP widths. 59

60 This gave a curve of SOBP width vs. stop pulse that could easily be looked up to determine what the new weighting factors were for each peak to create the desired modulation width. 2.3 Results and Discussion Pristine Bragg Peaks The simulated deepest Pristine Bragg peaks for each configuration were normalized and compared with the corresponding measured data at PDD of 0.5 cm (PDD (0.5)) and the range of the beam, defined as the depth of 90% dose (D90). For all 24 beam configurations, the range matched to within millimeter accuracy and the PDD (0.5) was within 2%. The average differences over the large group for the PDD (0.5) and D90 depth was 0.9% (ranged from %) and 0.06 cm (ranged from cm), respectively (shown in Table 4). The average differences over the deep group for the PDD (0.5) and D90 depth was 1.3 % (ranged from %) and 0.03 cm (ranged from cm), respectively (shown in Table 5). The average differences over the small group for the PDD (0.5) and D90 depth was 1.2 % (ranged from %) and 0.04 cm (ranged from cm), respectively (shown in Table 6). An example from each group is shown in Figure 17, where normalized depth dose curves from beam commissioning measurements and simulated data are plotted for comparison. 60

61 Table 4. Absolute differences between measured and simulated percent depth dose data for the large group. PDD (0.5) Measured Simulated Difference in Config. Difference (%) D90 (cm) D90 (cm) D90 (cm) Avg

62 Table 5. Absolute differences between measured and simulated percent depth dose data for the deep group. Config. PDD (0.5) Difference (%) Measured D90 (cm) Simulated D90 (cm) Difference in D90 (cm) Avg Table 6. Absolute differences between measured and simulated percent depth dose data for the small group. Config. PDD (0.5) Difference (%) Measured D90 (cm) Simulated D90 (cm) Difference in D90 (cm) Avg

63 Figure 17. Measured data from commissioning (circles) and simulated (solid lines) normalized pristine Bragg peaks from (A) a large configuration with a range of 22.5 cm, (B) deep configuration with a range of 29.5 cm and (C) a small configuration with a range of 15.3 cm. 63

64 To achieve agreement between the measured data and the simulated data for the pristine Bragg peak of each configuration, small adjustments were made to the manufacturer provided geometry of the beam delivery system. As described by Bednarz et al. (59), there is uncertainty as to how accurate the provided blueprints reflect the real geometry of the components in the commissioned treatment head. Therefore, tuning some of the physical parameters of the geometry was necessary assuming that the adjustments were within manufacturer tolerance. (60) The range of proton beam was first matched by fine adjustments of the post absorber thicknesses using stopping power conversion ratios. In most cases, after range correction, the entrance dose and Bragg peak distal fall off needed to be tuned. Paganetti and Bednarz showed that the beam spot size, energy and angular spread can influence the shape of the Bragg curve and provides one of the largest sources of uncertainty, due to the difficulty in measuring these parameters. (19, 59) Paganetti et al., also showed that beam energy spread has the largest influence on entrance dose and distal fall off. (19) The same trend is also found in our study, where increasing the energy spread increased the entrance dose and the distal fall off length. Therefore, in order to tune our pristine Bragg peaks to appropriate distal falloff, the energy spread was adjusted iteratively until the best agreement between measured and simulated data was reached. We found that the energy spread that achieved the best agreement between the measured and the simulated data for most of the configurations was 0.4%. The angular spread and spot size were then optimized in order to achieve further agreement in data. A radian standard deviation in angular spread in both the x and y directions gave the best representation of commissioned data, along with a beam spot size of 8 millimeters for the majority of configurations. 64

65 In the case of the large configuration group, some of the shallower range configurations were not able to be tuned to match within our set criteria with adjustments to the source alone. The Mevion system has a consistent distal falloff margin regardless of the energy or configuration. We found that as we decreased the energy of the beam and increased the field size, the distal fall off margins in simulation began to degrade. This is likely due to the presence of more material in the beam that introduced large uncertainty in geometry; and influenced the beam energy distribution in a way not accounted for in our simulations. In these specific cases, small adjustments (on the order of a tenth of millimeter) to the thickness of the first scattering foil were made in order to adjust the distal gradient to agree with measurements. These adjusted first scattering foils were used in the geometry for the subsequent Bragg peak simulations Spread Out Bragg Peak Matching At our institution, SOBP width is defined as the distance between the proximal 95% dose depth and the distal 90% dose depth. To evaluate the agreement between simulated results and measured data, the SOBP width, beam range, and the depth of distal 20% dose were compared. Figure 18 shows the matching of normalized percent depth dose curves for configuration 1, 13 and 20. For all three SOBPs, the distal 90% depths were matched within 0.1 centimeter difference. The differences of distal 20% depths were within 0.15 centimeters and the largest discrepancies in SOBP width was 0.24 cm for configuration 1 as shown in Table 7. 65

66 Table 7. Comparison of measured and simulated SOBP for configurations 1, 13 and 20. Config. Measured Modulation Width (cm) Simulated Modulation Width (cm) Mod. Width Difference (cm) D90 Difference (cm) D20 Difference (cm) Large Deep Small This has been a difficult characteristic to precisely reproduce SOBP width in simulation due to the nature of shallow gradient at the proximal end of the PDD curve. The main contribution to the proximal end of a SOBP is from the beam pulses hitting on the finite steps of the RMW. It has been shown as partial shining effect that even small temporal errors in the BCM and RMW synchronization can result in substantial errors for SOBP formation. (60) Furthermore, studies have been done on optimization of the BCM in Monte Carlo simulation for a continuous beam from an isochronous cyclotron, but there has been (19, 60) little work done on how to optimize this for a pulsed beam from a synchrocyclotron. Nevertheless, benchmarking simulated SOBP width within 0.3 cm to measurement data is in tolerance when commissioning a Monte Carlo simulation model for dose calculation. (40) This agreement was achieved by actually determining the step location for each beam pulse which is necessary to represent well in SOBP simulations due to the presence of peak broadening that occurs when a pulse strikes the edge of a step. 66

67 Figure 18. Measured data from commissioning (circles) and simulated (solid lines) normalized spread out Bragg peaks for (A) large configuration (configuration 1) with a range of 25 cm, (B) deep configuration (configuration13) with a range of 32 cm and a (C) small configuration (configuration 20) with a range of 15.3 cm. 67

68 2.3.3 Lateral Beam Profile Matching Three of the lateral beam profiles are shown in Figure 19 with the simulated and measured data plotted on each graph. The penumbrae were calculated as the distance between the 80% and 20% dose levels. Beam flatness and symmetry was calculated using Equations 8 and 9, respectively. F = D min D max D min +D max 100, Eq. 8 Where D min and D max are the minimum and maximum doses within the middle 80% of the field size. S = DL DR DL+DR 100, Eq. 9 Where DL and DR are the integral doses of the left and right side of the radiation field, respectively. The full width at half maximum values (FWHM) for both measurement and simulated data were calculated and compared with each other. The absolute differences in the penumbrae between simulated and measured profiles at each depth for each configuration all agreed to well within a millimeter. Flatness and symmetry values for all 12 profiles (6 configurations at 2 depths each) are listed in Table 8. All of the FWHM calculations agreed to within 0.2 cm. It has been noticed that subtle changes in the physical dimensions of the aperture cutout or air gap setting could affect in large the simulated field size due to the diverging nature of the beam. 68

69 Figure 19. Measured data from commissioning (circles) and simulated (solid lines) normalized lateral beam profiles from a (A) large configuration (configuration 1) at a depth of 10 cm, (B) deep configuration (configuration 13) at a depth of 20 cm and (C) small configuration (configuration 20) at a depth of 10 cm. 69

70 Table 8. Comparison of penumbra, flatness, symmetry and full width at half maximum for measured and simulated cross profiles for two configurations in each group. Config./ Group 1 (Large) 5 (Large) 13 (Deep) 17 (Deep) 18 (Small) 20 (Small) Depth (cm) Penum. Diff. (cm) Sim. Flat. (%) Meas. Flat. (%) Sim. Symm. (%) Meas. Symm. (%) FWHM Diff. (cm)

71 2.3.4 Spread Out Bragg Peak Modulation Width Adjustments Based on the timing of beam pulse and the speed of the rotating RMW, there are roughly 50 contributing pulses per full rotation of the modulator wheel. The intensity of each of the pulses is also modulated to achieve the preset SOBP width with 0.1 centimeter accuracy. Using the modulation adjustment algorithm described in the methods section, a reference SOBP with various modulation widths were simulated and shown together with a pulse modulation function in Figure 20. If the intended modulation width does not correspond to an integer stop pulse, one can use linear interpolation to determine the fraction of the last pulse to be applied to the corresponding pristine Bragg peak. In this example, the desired SOBP width was changed to 10 cm, resulting in pulses per revolution of the modulator wheel. The measured modulation width was 10.1 cm and the simulated modulation width was 9.94 cm, giving a difference of only 0.17 cm. The depth at 90% dose differed between the two curves by 0.02 cm. The depth of 20% dose shows an absolute difference of 0.07 cm between simulated and measured beam data. Our method of SOBP width adjustment thus shows to be a reasonable option when simulating a pulsed proton beam from a synchrocyclotron. 71

72 Figure 20. A) The stop pulse curve for configuration 21. The curve will give the appropriate stop pulse to use when adjusting the SOBP width for simulations. Linear interpolation is used to determine the fraction of the last stop pulse to be applied when the desired modulation width is between pulses. This curve was used to provide the stop pulse to simulate the reference beam at our institution. B) The measured data from commissioning (circles) and simulated (solid lines) normalized spread out Bragg peak for the reference SOBP. 72

73 2.4 Conclusion To our knowledge, this study marks the first full scale Monte Carlo simulation of the Mevion double scattering S250 proton therapy system. The beam line components for all 24 beam configurations were extensively modeled and we have shown excellent agreements for the simulated range and modulation width of given SOBP with the corresponding measured beam data taken during commissioning. This demonstrates that the simulation could serve as a promising tool for generating commissioning data for treatment planning system with acceptable accuracy and possible reduction of the intensive time required in measured data collection for the commissioning and verification of the proton treatment planning system. 2.5 Disclosure of Conflicts of Interest The authors have no relevant conflicts of interest to disclose. 2.6 Acknowledgments The authors would like to thank Dr. Hosang Jin, Dr. Maria Ferreira, Dr. Rafiq Islam, and Dr. Andy Lau for their work in acquiring the commissioning data used in this study. The authors would also like to thank Mr. Dan Green, for his guidance on the Mevion system. 73

74 CHAPTER III SHIELDING VERIFICATION AND NEUTRON DOSE EVALUATION OF THE MEVION S250 PROTON THERAPY UNIT Michael Prusator, M.S., Salahuddin Ahmad, Ph.D. and Yong Chen, Ph.D. Stephenson Oklahoma Cancer Center, University of Oklahoma Health Sciences Center, Oklahoma City, OK, USA Published in: Journal of Applied Clinical Medical Physics, 2018;19(2): doi: /acm Reprinted with permission of John Wiley & Sons, Inc. All rights reserved. M Prusator planned and performed simulations, analyzed data, prepared figures and wrote the first draft and multiple revisions of the manuscript. 74

75 3.1 Abstract For passive scattering proton therapy systems, neutron contamination is the main concern both from an occupational and patient safety perspective. The Mevion S250 compact proton therapy system is the first of its kind, offering an in room cyclotron design which prompts more concern for shielding assessment. The purpose of this study was to accomplish an in-depth evaluation of both the shielding design and in room neutron production at our facility using both Monte Carlo simulation and measurement. We found that the shielding in place at our facility is adequate, with measured annual neutron ambient dose equivalents at 30 cm outside wall/door perimeter ranging from background to 0.06 msv. The in room measurements reveal that the H*/D decreases when the distance from isocenter and field size increases. Also, the H*/D generally increases when the angle around isocenter increases. Our results from in room measurements show consistent trends with our Monte Carlo model of the Mevion system. 75

76 3.2 Introduction Proton beam radiation therapy, first introduced in 1946 by Robert Wilson, (1) is increasing in popularity in both the United States and the rest of the world. Possible benefits of proton therapy lie within the Bragg Peak, where at a finite depth the ionization rises sharply to a maximum, and then falls quickly to near zero. This characteristic can offer high conformity treatments and sparing of tissues distal to the target in comparison to conventional photon therapy. Advancements in present technology have led proton therapy to be considered as a viable and possibly improved means of providing radiotherapy to a variety of treatment sites. Hazardous neutrons resulting from intranuclear cascade interactions from incoming protons are highly penetrating in nature and can have high RBE values. (2,3) Thus, neutron production is the main concern regarding shielding applications and in room contamination dose to a patient during treatment. Typical neutron shielding for proton therapy systems is usually done in two stages, the design phase and the evaluation phase. The evaluation phase is conducted using measurements after the treatment unit vault is constructed according to the shielding design. The shielding design is done prior to construction of the therapy unit vault, using one of two accepted methods. One of these two procedures is an analytical method assuming point beam losses, based upon the Moyer model (4), shown in Equation 10; the other requires full Monte Carlo simulation. H = H 0 r 2 [E p E 0 ] α exp [ d λ ], Eq. 10 Where H is the maximum dose equivalent rate at a given radial distance r from the source of neutron production, d is the shield thickness, λ is the attenuation length of the shielding 76

77 material E 0 = 1 GeV, H 0 =2.6 x10-14 Sv m 2, deemed the source term, and α is a fitting parameter, fixed at 0.8. The Moyer model was first introduced as a method for shielding design of high energy particle accelerator facilities. (4) The drawback from the Moyer model as it is presented in equation 1 is the lack of flexibility for use in other types of shielding applications. Two major assumptions in this model is that λ and H 0 are fixed values, which is valid at high energies. (5) Much work has been done since in the introduction of this model to increase its validity for use in shielding design for proton treatment facilities. (6-9) The use of analytical models for proton shielding applications is well suited for bulk design of a facility, where high efficiency of results can be obtained. (5) The second shielding calculation method is a full scale Monte Carlo simulation of the treatment facility geometry with detectors at points of interest to score dose values. In comparison studies of the Monte Carlo method and the analytical method, the Monte Carlo method was more accurate in situations where the neutron fields were incident on barriers at angles of high obliquity when compared to measured data. (10) Shielding studies utilizing complete three-dimensional Monte Carlo simulations for proton therapy facilities show that Monte Carlo can reasonably predict dose rates at points of interest when compared to measurements. (2) Newhauser et al show that simulations agree with experimentally measured neutron dose values to within 10 percent, with the Monte Carlo values generally overestimating the measured values. (2) The drawback of this technique is the increased computing power and complexity of computer codes required to obtain relevant data. (10) In 2012, a new Monte Carlo Code specific for use in proton therapy was introduced. TOPAS, or TOolkit for Particle Simulation, is a Monte Carlo based platform that was 77

78 designed to handle the complexities in geometry introduced by proton therapy systems. (11) The concept of TOPAS is to preserve the GEANT 4 code as the underlying platform and includes additional codes adapted specifically for proton therapy. TOPAS in proton therapy has the complex geometries prebuilt into the coding toolkit. These include range modulator wheels based on commercially available wheel designs, and complex patient apertures and compensators. (11) TOPAS also has a time feature component for the simulation of moving geometries. This component consists of parameters describing the change of a time feature value, such as the rotation of the range modulator wheel. The commercially available Mevion S250 compact proton therapy unit has recently been installed at our institution. The Mevion S250 is a compact proton therapy unit in which all components of the treatment system are located in the treatment room. The cyclotron is attached to a gantry that rotates around the patient during treatment to provide beam at various angles. (12) The unit offers 24 different beam configurations categorized into large, deep and small groups, each with their own specific beam line components. The large group allows for field sizes up to 25x25 cm 2, and a maximum depth and modulation width of 25 and 20 cm in water. Both deep and small groups offer a maximum field size of 14 x 14 cm 2. The deep group has a maximum depth and modulation width of 32 and 10 cm water, while the maximum depth and modulation width of the small group is 20 and 20 cm in water. The in room design of the unit poses unique shielding challenges that are not present in the shielding design of traditional units. Thorough shielding design assessments using a hybrid technique of both the analytical and Monte Carlo methods have been recently conducted in a prior study. 23 The first objective of this study is to evaluate both 78

79 the hybrid shielding design technique used in the aforementioned study, and the shielding put in place at our institution. The in-room cyclotron design of the Mevion system provides a unique neutron signature around the patient that has not yet been fully described in a treatment setting. Neutron production from passive scattered proton systems has been extensively investigated, but the conclusions drawn are from units other than Mevion S250. (13-16) Chen et al. recently performed a comprehensive neutron assessment of the Mevion S250 using Monte Carlo methods. However, the study was conducted in a factory testing vault with dimensions about two times smaller than the actual treatment vault found at centers that currently use the Mevion system. According to the conclusions drawn from this study, the smaller vault likely confounded their results by an overestimation of the actual in-room neutron dose. (17) The second objective of the present study has been to perform a comprehensive in room neutron dose evaluation using both Monte Carlo simulation and measurements in a realistic clinical treatment vault. 3.3 Methods Shielding Design Evaluation For the evaluation of the prior shielding method, neutron ambient dose equivalent (H * ) measurements were done at 8 critical locations, shown in figure 21, using the SWENDI-2 (Thermo Scientific, MA) neutron detector. A 250 MeV beam with a range of 15 g/cm 2 and modulation width of 10 g/cm 2 delivered proton dose (D) through a 10x10 cm 2 field to a solid water block phantom measuring 30x30x40 cm 3. The MU rate used for measurement was 127 MU/min, and the machine was allowed to deliver dose until a detectable neutron ambient dose equivalent was recorded. The therapeutic dose delivered to the phantom was 79

80 calculated using the absolute machine output for the given configuration and the actual monitor units delivered. Figure 21. The locations and summary of the eight critical point locations where neutron doses were calculated and measured. 80

81 The workload was calculated based on 20 patients being treated with 2 Gy fractions per day. Under the assumption of adding equivalent treatment specific QAs, 40 beams delivering 2 Gy/per beam to isocenter were utilized per day in the calculation. Over a five day workweek, 400 Gy of beam was assumed to be delivered by the proton system to isocenter, resulting in a workload of 400 Gy/wk In Room Neutron Study Measurement Using the SWENDI-2 neutron detector, in room measurement of neutron ambient dose equivalent per therapeutic dose delivered (H * /D) were made around a 30x30x40 cm 3 solid water phantom using a 250 MeV beam at distances of 20, 40, 60, 80, 100 and 150 cm from isocenter. The SWENDI-2 detector was chosen for the in-room measurements due to the wide energy response (thermal to 5 GeV) and short dead time (1.8 μs). (22) At these same distances, measurements were done at angles of 0, 45, 90 and 135 degrees around isocenter for a large, deep and small beam configuration. For the large configuration, a 20x20 cm 2 field size was used, and a 10x10 cm 2 field size was used for the deep and small configurations. All measurements were done at a gantry angle of 90 degrees. Figure 22 demonstrates our setup and Table 9 gives a summary of the beam configurations utilized. 81

82 Figure 22. Representation of the points of measurement around the phantom. d represents the distance of the detector from the proximal phantom surface. Each tic mark represents a point where a measurement was taken. Table 9. Beam characteristics of the three configurations chosen for in room measurements Range Modulation Configuration (g/cm 2 ) Width (g/cm 2 ) Large Deep Small

83 Using the same beam configurations, neutron measurements were done with different field sizes. Field sizes with diameters of 0, 5, 10, 12 and 15 cm were used for the small and deep configurations and 0, 8, 16, 20 and 26.5 cm for the large configuration. These measurements were also done at a distance of 100 cm from isocenter and at a gantry angle of 90 degrees. To attain the H/D value at each point, dose was delivered to isocenter until an instant neutron dose equivalent rate was observed to be stable via a camera from outside the proton vault. Several measurements were taken at low (about 20 MU/min) and high dose rates (150 Mu/min) to assure that the results were not affected by pulse pile up. Assuming the proton dose rate delivered can be calculated using the dose/mu coefficient, the H/D is calculated using Equation 11. H D = H r MU r c, Eq. 11 where H r is the measured neutron equivalent dose rate in msv/hr, and MU r is the proton output rate, in MU/hr, and c is the dose/mu calibration coefficient Monte Carlo Simulation A defining characteristic of the MEVION S250 proton system is that the compact superconducting cyclotron, mounted on a rotating gantry, is present in the treatment room with patients. For the simulation, neutron production from the cyclotron was taken into account as an additive source, and the treatment vault walls were included to incorporate the effects of neutron scatter. 83

84 Using the TOPAS (TOolkit for PArticle Simulation) Monte Carlo code, the simulation model from the work of Prusator et al. (19) was used to model all field shaping devices and the nozzle of the Mevion system. A simple approach was implemented to model the cyclotron in TOPAS with the assumption that neutrons within the cyclotron primarily were produced from collisions of protons with the magnet and extracting foil when exiting the device. A cylindrical iron target (radius= 5.8 cm, length=8 cm) was simulated and bombarded with primary proton beams to model a source of neutron production as described in the work of Chen and Ahmad. (18) The physics list in TOPAS was set to the GEANT 4_Modular physics list that has shown to work well when simulating proton therapy. (11) EM-Standard was used to model the gamma, electron and proton electromagnetic processes. Hadron inelastic collisions of protons and neutrons were simulated using the Binary Cascade model and hadronic ion (Z>1) inelastic collisions were modeled using Ion-Binary Cascade. Elastic collisions of protons, neutrons and ions were simulated in the Elastic model. To model radioactive decay processes G4 Decay was used. The cutoff step range for electron, proton and gamma was set to 0.05 mm. Using the same beam configurations and point locations used for the measurement, neutron fluence scorers where placed in the simulation space around the water phantom. At each location, the fluence scorer counted the neutron fluences in bins from 0-30 kev, kev and 201 kev-250 MeV generated while beam was delivered through the nozzle to the water phantom. The fluences scored were normalized by the total number of incident protons through the beam delivery system. The ambient dose equivalent per incident proton was calculated by multiplying the particle fluence for a given energy bin with the 84

85 corresponding fluence to dose conversation factor from ICRP report number 74. (20) The ambient dose equivalents from each energy bin were then summed together to achieve the total ambient dose equivalent at a given point of interest. The contribution from the cyclotron was taken into account via addition of the source terms calculated from Prusator et al. in a previous study. (24) 3.4 Results and Discussion Shielding Design Evaluation The highest measured neutron dose equivalent was at the treatment console and lowest measured dose was in the waiting area. This coincides with what was predicted in the shielding design. Table 10 shows the annual doses calculated in the prior shielding design and those projected from measurement using the SWENDI at each of the eight critical points of interest around the proton treatment vault. The shielding design goals recommended by the NCRP at each point are listed for comparison. (23) There is reasonable agreement between the calculated and measured values, with the largest discrepancy outside the vault at point 2. In most cases, the calculated values overestimated what was measured. The differences found in this study may suggest that the prior design method could overestimate the neutron dose especially when the simplified source terms were used. Also, the lower values in measurements may be due to the restrictions on the actual beamon time as well as the allowable dose rate. During the shielding survey measurements, the SWENDI detector was set to the integral mode (instead of the rate mode) to accumulate the neutron dose due to the ultra-low signal strength. The machine parameters were set to the maximum allowable dose rate with 15 minutes of constant beam on time. Because of the time limit and low dose rate achievable with the unit, it proved difficult to accumulate 85

86 an accurate appreciable dose. It is also important to note that the workload and occupancy factors were a direct multiplier of the dose beyond the barrier, and must be accurately approximated for validation of the results. Table 10. Annual neutron equivalent doses calculated from the shielding design and measured during evaluation for each of the 8 critical points Pt. Calculated Measured Design Goal (msv) (msv) (msv) Minimal Minimal In Room Neutron Study Measurement For the deep and large configurations at all angles around the isocenter, our measurements showed that as the distance from the isocenter increased, the neutron dose equivalent decreased. However, for the small configuration, and at an angle of 135 degree, the neutron dose equivalent increased from a H/D of 0.22 to 0.26 msv/gy. This can possibly be explained with the following: at 135 degrees the distance increased from isocenter. However, the detector was moved closer to the cyclotron and nozzle, which were the main 86

87 producers of neutrons for shallow ranges due to the presence of energy degrading materials upstream of isocenter. Figure 23 shows the measurement results. Figure 23. Comparison of measured H/D versus distance at an angle of 90 degrees between the large configuration with a 20x20 cm 2 field, and the small and deep configurations with a 10x10 cm 2 field. With regards to H/D as a function of angle around the isocenter, the measurements showed for all configurations that the neutron dose equivalent increased as the angle increased. This was likely because increasing the angle brought the detector closer to the neutron producing components of the treatment unit, especially the cyclotron. The maximum H/D was found to be for the large option at an angle of 135 degrees, with a value of 0.5 msv/gy and the minimum value was at 0 degrees for the small option, measuring 0.2 msv/gy. A summary of the measurements can be seen in Figure

88 Figure 24. Comparison of H/D versus angle at a distance of 100 cm between the large configuration with a 20x20 cm 2 field, and the small and deep configurations with a 10x10 cm 2 field. For each configuration, as expected, the H/D decreased as the field size increased, as shown in Figure 25. This was due to the fact that smaller field sizes required larger aperture collimation, resulting in more beam striking brass aperture followed by more neutron production. For this reason, the large option with the smallest field size resulted in the highest H/D of 0.41 msv/gy, and the smallest H/D of 0.19 msv/gy was created by the small option with the largest field size. 88

89 Figure 25. Measured H/D as a function of field size for the three configurations at a distance from isocenter of 100 cm Monte Carlo Simulation When compared to our simulation results, the doses measured inside the vault followed the same trends, but were about an order of magnitude lower. Figures 26, 27 and 28 compare the simulated and measured data for each of the options studied. At 1 meter and at an angle of 90 degrees, the simulated values for the large, small and deep configurations were 2.2, 2.0 and 2.2 msv/gy, respectively. We believe that this may be because of the simplistic approach of modeling the cyclotron. The cyclotron was represented as a single iron cylindrical target, and it was assumed that all neutron production came from point losses within the target. In reality, the transport of protons within the cyclotron would result in neutron production throughout the accelerator, thus confounding the assumption of the cyclotron as a point source. Also, self-shielding is not taken into account when a simple 89

90 iron target is used to model the accelerator. This in turn will cause an over estimation of the neutron contribution from the cyclotron. Figure 26. Comparison of simulated and measured H/D values at an angle of 90 degrees around the water phantom for the large option. Figure 27. Comparison of simulated and measured H/D values at an angle of 90 degrees around the water phantom for the small option. 90

91 Figure 28. Comparison of simulated and measured H/D values at an angle of 90 degrees around the water phantom for the deep option. 3.5 Conclusion All points measured outside the treatment vault were beneath the NCRP recommended shielding design goals. The shielding barrier of the proton vault at our facility was found to be sufficient. We believe that because the hybrid treatment vault design model overestimated the annual neutron ambient dose calculations in each of the eight points compared to measurements, the model can thus be used as a conservative approach of calculation to design the shielding of a proton facility. Our in room measurements generally followed the same trends found through Monte Carlo simulations. In summary, we believe this data may be used to conduct further study in calculating secondary neutron dose to patients during proton treatments using the Mevion S250 system. 3.6 Disclosure of Conflicts of Interest The authors have no relevant conflicts of interest to disclose. 91

92 3.7 Acknowledgements The authors would like to thank Dr. Hosang Jin, Dr. Dan Johnson and Dr. Rafiq Islam for their assistance in data acquisition. The authors would also like to thank Mr. Dan Green, for his guidance on the Mevion system. 92

93 CHAPTER IV THE IMPACT OF PATIENT ANATOMY AND RANGE COMPENSATOR ON PROTON THERAPY MONITOR UNIT CALCULATIONS Michael Prusator, M.S., Salahuddin Ahmad, Ph.D. and Yong Chen, Ph.D. Stephenson Oklahoma Cancer Center, University of Oklahoma Health Sciences Center, Oklahoma City, OK, USA Submitted To: International Journal of Particle Therapy: 2018 M Prusator planned and performed simulations, analyzed data, prepared figures and wrote the first draft and multiple revisions of the manuscript. 93

94 4.1 Abstract A universal formalism for calibrating the dose per monitor unit (DMU) for a passive scattering proton therapy system has yet to be established. Two factors that can influence the DMU in a clinical treatment are the patient scatter factor (PSF), which accounts for changes in DMU due to patient anatomy, and the compensator scatter factor (CSF), which accounts for the presence of a range compensator during treatment delivery. Largely, these two factors are ignored due to the limitations of complex physical measurement. It is the goal of this study to use both Monte Carlo (MC) simulation and the pencil beam dose algorithm (PBA) in the treatment planning system (TPS) to investigate PSF and CSF in a variety of treatment scenarios that include sites of brain, lung, pancreas and prostate sites. Results of this study show that the CSF values are always greater than one, with some reaching nearly 4% above unity, and depend strongly on the shape of the component. MC and PBA calculated CSF factors agree very well, with average differences below one percent. PSF values calculated in this study ranged from to and are largely dependent on the type of tissue heterogeneities in the treatment field. MC and PBA calculated PSF factors show differences, with the largest discrepancies seen in lung cases, with an average difference of 1.9%. It is shown that dense bone will drive a PSF to values greater than unity, while large quantities of air decrease the PSF to below unity. Due to the magnitude of the PSF and CSF values, it is recommended that both MC and TPS should be utilized to take these factors into account in the final DMU calculation. 94

95 4.2 Introduction Proton therapy is becoming an increasingly popular method for the treatment of cancer. The sharp dose characteristics of the Bragg peak enable greater healthy tissue sparing and offer patients a treatment option with fewer side effects when compared to traditional photon therapy. (61-63) Since its proposed use by Robert Wilson in 1946 (1), sixty-three total proton facilities are treating patients around the world. (64) Technological advancements in proton therapy are continually being realized, thus increasing its availability as a treatment modality. There are currently forty proton therapy facilities under construction, and twenty-one more in the planning stage in the world. (64) A common technique for proton beam delivery is passive scattering. (9) Using this technique, a narrow, nearly monoenergetic, proton beam is transformed to a clinically useful size using scattering foils to laterally spread and flattened the beam and a range modulator wheel (RMW) to increase the width of the Bragg peak. (16, 17, 21) The final components which the beam passes through are patient specific devices, consisting of an aperture to collimate the field, and a range compensator to shape the beam distally. (65-68) A challenge that arises from use of the passive scattering proton beam delivery technique is the determination of beam output which is needed to deliver a prescribed dose, namely, calibrating the dose per monitor unit (DMU). Differences in range, modulation width and beamline components for a treatment field will all affect the dosimetric output from a passive scattering treatment unit. Currently, there is a lack of a widely established procedure for output calculations. (37) Analytical, correction based and simplified Monte Carlo (MC) techniques have been proposed, but due to the complexity and specificity of each individual treatment system, a global protocol has yet to be accepted. (35, 36, 38, 44) Many 95

96 proton facilities choose to use the calculation of DMU as a secondary check, and rather perform measurement in a water equivalent phantom to calibrate the DMU to the patient. (44) A challenge associated with machine output prediction is characterizing the effects of the patient specific compensator and patient anatomy on monitor unit calibration. (69) Analogous to photon therapy, implementation of these effects have been suggested as correction factors to the final calibrated DMU. The so called compensator scatter factor (CSF) is responsible for taking into account compensator effects and is defined as the ratio of DMU under treatment conditions with and without the compensator in place in a homogenous phantom. In many cases, the CSF is ignored due to the cumbersome nature of taking measurements for every patient s field specific compensator. In fact, Fontenot et al. concluded that the high dose gradients caused by the irregular shape of range compensators have the potential to make accurate calibration of DMU difficult, and suggested that calibrations be done without consideration of the compensator. (70) However, Akagi et al. reported that the effects of scatter from a compensator and patient anatomy could potentially result in a DMU difference of 2-3% from what is measured in a water phantom without the compensator. (71) The second factor which accounts for the effect of patient anatomy on DMU is the patient scatter factor. It is defined as the ratio of DMU in patient under treatment conditions to that in a homogenous medium. (36) During calibration, the effect of the PSF is neglected entirely due to the limitations of measuring the effects of patient inhomogeneities on the DMU. Akagi et al. has demonstrated the possibility of using the pencil beam algorithm (PBA) in the treatment planning system (TPS) as a means of calculating the PSF. (71) However, the study suggested that errors on the order of 2-3% are possible in the TPS calculated PSF 96

97 due to limitations of the PBA. (71) In one recent study involving prostate cases, measurements in an anthropomorphic phantom showed that the PSF was negligible on the DMU. (69) However, a self-proclaimed limitation of this study was that the results were valid for only prostate cases, and to come to a more global conclusion, other sites needed to be investigated. (69) Monte Carlo simulation is widely accepted to be the most accurate technique for proton treatment dose calculation. (72, 73) MC dose calculation is especially valuable in highly heterogeneous fields, where analytical techniques show limitations. (40, 41) Due to the difficulties in estimation of the PSF and CSF through measurement, the most accurate way to calculate these factors and implement them in the DMU calibration is likely through Monte Carlo simulation. However, there are technical drawbacks associated with this technique. Perhaps the most significant hurdle with this technique includes the long calculation times to run enough histories to gather significant data. (42) A previous study from Paganetti indicated that calculation time could reach upwards of six hours to calculate dose in patient, which proved to be too inefficient for clinical use. (40) Along with the extended calculation times, challenges also arise with MC in the complexities of simulating patient treatments in general. There are three different distinct categories of difficulties in particular for simulating a patient treatment plan. The first is modeling the proton treatment conditions accurately. This includes precise beamline modeling of the proton treatment unit, accurate characterization of the pencil beam exiting the cyclotron, and the ability to export and load patient specific devices such as apertures and compensators from the TPS into the simulation space. The second category of technical challenges to patient simulation is the import of the patient into the simulation. This requires export of the CT 97

98 data from the TPS into the simulation space, and conversion of the CT HU values to an accurate material density and atomic composition. Lastly, the user has to be able to obtain the dose data from the simulation, visualize it on the patient CT scan, and finally, make a quantitative comparison to the TPS. This entire process requires a significant amount of technical skill by the user, and there is yet to be a standardized protocol to achieve patient treatment simulation. Currently, the PSF and CSF are largely ignored, or only the CSF is factored in DMU calculations due to the limitations previously described. However, in one study the combined effect of these two factors is shown to influence the DMU by nearly 4%, which is a large enough value that the authors believe it should be accounted for. (36) In order for these factors to be included in output calculations, a deep investigation into the properties of the CSF and PSF is needed. To the best of the author s knowledge, the current literature lacks detailed analysis on what causes a PSF or CSF to deviate from unity, and most conclusions drawn are limited to specific sites. With a thorough understanding of what influences the values of these factors, conclusions can be drawn that can apply to any treatment site, regardless of method of calculation used. There are several goals of this work, the first is to develop a procedure for simulating patient treatments including a standard workflow, and a second is to use this workflow to calculate PSF and CSFs for a wide range of treatment sites and fields that can be compared to the PBA calculation technique. Finally, the properties that influence the PSF and CSF will be investigated. Taken all together, this will result in a way to incorporate these factors into the DMU calculations. 98

99 4.3 Methods Developing the Patient Treatment Simulation Workflow To develop the treatment simulation workflow, a proposed process is broken down into a sequence of discrete steps that a user can follow. Figure 29 shows the proposed process and the steps incorporated into the workflow. Figure 29. Proposed workflow for conducting patient treatment Monte Carlo simulations The first step in the process was to load the treatment beam with the correct parameters into the simulation, i.e. the correct range and modulation width. In this study, the recently modeled Mevion S250 was used as the proton source, and the appropriate beam parameters were created in a phase space file via the technique shown by Prusator et al. (20) The next step was to load the patient CT images into the simulation space. In order to achieve this, the patient CT was exported directly from the TPS using the export function to a DICOM file that was readable by the TOPAS simulation code. For TOPAS to recognize the CT DICOM file as a structure, the HU values of the data set must be converted to a material 99

100 density and subsequent atomic composition. This was perhaps one of the more challenging steps in the process, and this study closely follows a technique proposed by Schneider et al. (74) A CT scan of the CIRS electron density phantom was acquired from Schnell et al. (75) and a table of HU values for each plug and their material density was created, shown in Table 11. Table 11. HU values of each electron density phantom plug and their corresponding material density Plug Material Density (g/cc) HU Value Lung Inhale Lung Exhale Adipose Breast Water Muscle Liver Bone Bone Bone Nine linear functions were created (one used to interpolate the density between each of the HU values) and used by TOPAS to calculate the material density for each possible HU value in the CT data set. Figure 30 shows the curve created by the functions that relate HU to material density. 100

101 Figure 30. The conversion curve used by TOPAS to calculate material density for each HU value in a CT data set Once TOPAS has calculated the density for each CT voxel, it then assigned each voxel an atomic composition. This required a look up table of density to atomic composition to be loaded into TOPAS. Schneider et al. proposed a table of this data that was used in this study, shown in Figure 31. Any density values that fall in between data points were assigned an atomic composition based on the lower density data point. 101

102 Figure 31. Look-up table of density to atomic composition developed by Schnieder et al. used by this study to convert HU values to atomic composition (74) Next, the treatment parameters (beam angle, couch angle, air gap etc ) were loaded into TOPAS. This was done by looking up these parameters directly from the TPS, and manually coding them in the geometry file of the simulation setup. In order for patient treatment simulation, the patient specific aperture and compensator must be created in the simulation geometry. Both of these components were complex in shape, and required a careful technique to incorporate in simulation. TOPAS does have the capability to import geometric components as stereolithographic (STL) files. Therefore, to create the patient specific devices, the aperture and compensator was exported from the TPS as a part of the RN treatment plan (DICOM proton plan file). Using an in-house tool developed by Muller et al., range compensators were created in an STL file with the aid of MATLAB (The Mathworks, Inc.). Similarly, aperture STL files were created using MATLAB to draw the 102

103 cutout shape, and Autodesk Fusion 360 to add the physical thickness of the component, shown in Figure 32. (76) A) B) Figure 32. The sequence of steps for STL file creation of A) compensator and B) Aperture (76) Once the STL files were created, they were imported into TOPAS as would be any other geometry component. The treatment was then simulated by activating the phase space file once the geometry was complete. The last part of the process was to extract the dose data from the simulation for visualization and comparison to the TPS. A 3D dose scoring mesh was created to match the CT voxels (512x512 matrix). The calculated dose matrix from simulation was extracted in a DICOM RT Dose format where every CT voxel was associated with an accumulated dose value. Using the method described in Muller et al., (76) 3D slicer with the radiotherapy extension was used to read the exported RT Dose file and display the simulated doses on the CT scan. Small adjustments to the position of the exported dose volume were performed in order to place the dose overlay properly over the CT image. The simulated dose matrix was then 103

104 re-sampled to match the calculation grid in the TPS, (77) and a gamma analysis was performed to quantitatively compare the dose distributions from simulation to that calculated by the TPS. To benchmark this process, a two field treatment was simulated for a case using an anthropomorphic lung phantom provided by Imaging and Radiation Oncology Core (IROC) Houston. (78) The dose distributions were calculated using MC and TPS and compared in the axial, coronal and sagittal planes using the 2D gamma analysis. Measurements were also taken using film for all fields, and a 2D gamma analysis comparison against the MC and measured doses were conducted by the IROC group. Patient plans for four different sites, including brain, lung, pancreas and prostate were also simulated and compared to TPS. For each of the gamma comparisons, a criteria of 7% and 5 mm and a threshold of 10% was used PSF and CSF Study PSFs and CSFs for brain, lung, pancreas and prostate (three plans for each site) treatment sites were calculated using MC simulation and PBA from the TPS software to evaluate the agreement between the two PSF and CSF using the PBA To calculate the PSF and CSF using the PBA in the TPS, the same method was followed as described by Sahoo et al. (36) Three patient fields were taken from real patient plans for each of the treatment sites previously mentioned. The dose was calculated at isocenter, which was located in the center of the PTV along the central beam axis, in each field. A corresponding plan was then created using the verification plan tool in the TPS. The proton fluence, aperture and compensator were copied from the treatment field and dose 104

105 was calculated at the same water equivalent depth (WED) along the central beam axis in a homogenous water phantom. The value of the PSF was then acquired for each field using Equation 12. The compensator was then removed from the verification plan, and the dose at the same equivalent WED was recalculated. The value of CSF was then acquired using Equation 13. Eq. 12 PSF = d p/c Eq. 13 dvp/c CSF = d vp/c dvp/nc Where d p/c is the dose in patient, d vp/c is the dose in the verification plan at the same WED, and d vp/nc is the dose at the same WED in the verification plan without the compensator in place. The workflow has been shown in Figure 33 for a typical brain case. 105

106 106

107 Figure 33. An example of the PSF and CSF calculation points using the treatment planning system verification plan tool for a brain case at a WED of 5 cm. A) Shows the calculation point in the patient, B) shows the calculation in the phantom via the verification plan tool with the compensator in place and C) shows the same calculation as B without the compensator PSF and CSF calculated in MC Using the same workflow derived previously for patient treatment simulation, the same fields used to calculate the PSF and CSF factors in the TPS were also simulated using the MC technique. The TOPAS (TOolkit for PArticle Simulation) version 3.1 (48) simulation code was utilized for all of the MC calculations. TOPAS has been shown to be a prominent 107

108 and user friendly simulation code for proton therapy. (49) The recently modeled beamline for our Mevion S250 double scattering proton therapy system was used for simulating the patient treatment plans. (20) The treatment parameters for each field were exported from the TPS into the simulation space, along with the patient CT images. HU values from each pixel were then converted to corresponding materials and density. Patient specific apertures and compensators were exported from the TPS and converted to STL files for implementation in the MC space Phantom Study A benchmark study was carried out to ensure the integrity of our simulation technique in an anthropomorphic phantom. A two field treatment plan was first created for a brain case in a RANDO phantom. The brain case was chosen due to the simplicity of the site, as well as the wide range of HU values that would be present in the CT dataset. For the physical measurement, one hundred monitor units (MU) were delivered to the RANDO phantom under treatment conditions and a dose measurement at isocenter was taken using a thermoluminescent detector (TLD). Two more TLD measurements were taken under the same treatment parameters with a water equivalent phantom substituted in for the RANDO phantom. One measurement was taken with the compensator in place, and the other without. The dose measurements taken from the TLD were used in equations 1 and 2 to calculate the measured PSF and CSF. The same treatment fields were then simulated in MC and also calculated using the PBA in the TPS for comparison purposes Evaluation of PSF The main function of the PSF is to account for patient anatomy on DMU. Therefore, to investigate how different tissue types influence the PSF, each of the prostate and lung 108

109 treatment fields were broken down into groups of materials based on the HU values within the field. The prostate cases were chosen to isolate how bone influences the PSF, where the lung cases were chosen to investigate the influence of air. Using the TPS, the 50% isodose curve was contoured to a structure, and a histogram of HU values from the created structure was produced for each field. Five different material types were grouped and assigned their percentage contribution in the field according to their HU, namely air (-1000 to -650 HU), thin tissue (-651 to -500 HU), soft tissue (-501 to 125), bone (126 to 500 HU) and dense bone (501 to 1245 HU). The percentage of materials present were then compared to the calculated PSF values for the field Evaluation of CSF From the twenty-eight CSF factors calculated, we chose seven that best reflected the range of values attained and calculated the interquartile range (IQR) of thickness values of the compensator around the central beam axis. The IQR was chosen as a metric of irregularity of compensator shape, due to its ease of calculation, and it s stability in the presence of extreme values in the data set. The area around the central beam axis chosen to calculate the IQR was determined through calculation of the maximum scattering distance of the beam with respect to beam energy by the General Highland approximation. (28) 4.4 Results and Discussion Patient Treatment Simulation Procedure To validate the workflow of treatment simulation proposed in this study, a two field treatment plan was made using the IROC lung phantom, and a 2D gamma analysis was conducted between the simulated, TPS calculated and film measured dose distributions. 109

110 Figure 34 shows a comparison of simulated and TPS calculated doses for the phantom, along with the gamma analysis. Figure 34. A) Side by side comparison of the TPS (left) and TOPAS simulated (right) dose distributions for the IROC lung phantom. B) displays the gamma analysis with points in green taken to be passing. (76) 110

111 For each of the three measured dose planes, the simulated dose distributions had a higher gamma passing rate than the TPS calculated dose when compared to film. On average between the three planes, the TPS and film measurement comparison achieved a passing rate of 93%, the simulation and film measurement averaged a 99% comparison passing rate, and the simulation and TPS calculated dose planes averaged a gamma passing rate of 98%. Table 12 displays the gamma passing rates for each comparison on each dose plane. Table 12. Gamma passing rates of the comparisons between MC, TPS and film measurement for the axial, coronal and sagittal planes of the IROC lung phantom Plane TPS vs. Film MC vs. Film MC vs. TPS Axial 94% 99% 98% Coronal 94% 99% 97% Sagittal 92% 99% 98% The MC and measurement agreed better than the TPS and measurement. It is believed that this was due to the limitations of the PBA dose calculation algorithm. With regards to the true patient plans, the MC and TPS agreed better for fields with less heterogeneity in the BEV. Figure 35 shows the visualization and gamma analysis of the four treatment sites studied. 111

112 A) 112

113 B) 113

114 C) 114

115 D) Figure 35. Visualization and dose comparison of a A) prostate, B) pancreas, C) lung and D) brain site between MC and TPS. The top dose distributions are from the TPS and the middle dose distribution are those calculated from MC. The lowest row is the gamma analysis where the points in yellow and red are the failed points. (79) 115

116 The gamma pass rates for the prostate, pancreas, lung and brain case were 97%, 93%, 83% and 90%, respectively. The lung was the site with the lowest passing rate, and this was likely due to the large amount of air in the field which inherently produced inaccuracies from the PBA in the TPS. As clearly shown by the phantom and patient cases, the proposed workflow for conducting patient treatments was viable to use in both the clinic and in research. The excellent gamma passing rate for the measurement and MC calculated doses in the phantom study proved that the developed MC model for treatment simulation was accurate and could provide reliable results. The dose calculation, visualization and dose comparison study for the patient cases can be very useful for further research in the reliability of TPS dose calculation for passive scattering proton therapy PBA and MC Comparison The results from the benchmark case in the Rando phantom showed good agreement between MC simulation and measurement. The measured PSF was and the MC calculated PSF was 1.031, giving a difference of only 1.4%. The difference between the measured and MC calculated CSF was only 0.7%, with the measured and MC factors being and 1.016, respectively. This agreement suggested that our MC model was adequate for further PSF and CSF calculations and was a good point of comparison for the PBA calculation method. Table 13 shows the PBA and MC calculated PSF and CSF factors for the twelve fields studied. The MC and PBA calculated PSF factors agreed to within 3% of each other. The average differences between MC and PBA factors for the brain, pancreas, prostate and lung sites were 1.6, 1.0, 0.6 and 1.9% respectively. For each field, both prediction techniques 116

117 were in agreement as to when the factors were greater than or less than unity. In nearly all cases, the MC calculated PSFs were larger in magnitude than the PBA calculated PSFs. This suggests that our MC model predicts the scatter from surrounding patient anatomy to play a more prominent role in the DMU than does the PBA. The limitations of the PBA in TPS to accurately account for proton scatter in heterogeneous media has been well documented prior to this study, (28, 41) and it is believed that the discrepancies in the current data are a result. This is clearly shown in the lung cases, where there was the most inhomogeneity present in the patient anatomy, and disagreement between the MC and PBA was the largest. The CSFs calculated using both techniques agreed very well. The largest discrepancy between MC and PBA factors was 1.4%, with the average differences for each site being at or below half a percent. This data suggested that the PBA was adequate for predictions on the impact of the compensator on the output calculation. The improved agreement of the CSF between the two methods of calculation when compared to the PSF was likely because the PBA needed only to consider scatter from Lucite/air interfaces for the CSF factors. The PSF was much more complex in that many different types of material interfaces exist with a wide range of atomic makeup and densities. Due to the limitations of the PBA to model scatter, it was reasonable to see larger discrepancies in the PSF calculation methods than for the CSF. 117

118 Table 13. A comparison of the CSF and PSF factors calculated in both MC and the PBA in the TPS Location TPS CSF MC CSF TPS PSF MC PSF Brain Brain Brain Pancreas Pancreas Pancreas Prostate Prostate Prostate Lung Lung Lung

119 4.4.3 PSF and CSF PBA Results Table 14 shows the CSF and PSF factors calculated using the PBA in the TPS for each of the brain, pancreas, prostate and lung treatment plans studied. Table 14. The PBA calculated CSF and PSF factors for each patient plan Patient Location CSF PSF CPSF A Brain B Brain C Brain D Brain E Brain F (Field1) Pancreas F (Field2) Pancreas G Pancreas H Pancreas I (Field1) Pancreas I (Field2) Pancreas J (Field1) Pancreas J (Field2) Pancreas K (Field1) Prostate K (Field2) Prostate L (Field1) Prostate L (Field2) Prostate M (Field1) Prostate M (Field2) Prostate N (Field1) Prostate N (Field2) Prostate O Lung P Lung Q Lung R Lung S Lung PSF Evaluation The value of the PSF appeared to be heavily dependent upon material in the beam s eye view (BEV). Sites with large quantities of air in the BEV (such as lung cases) yielded a 119

120 PSF less than one. Air, having much less scattering power than water, resulted in less scatter contribution to the point of measurement than that in phantom. Figure 36 shows a histogram of CT values for Patient S, where there are large amounts of air present. Figure 36. A histogram of CT values in the treatment field for Patient S The first peak shown at -800 HU was representative of the quantity of air in the field, while the second peak around 0 HU was soft tissue. The area under these two peaks gave a quantitative breakdown of the percentage of each material present. For Patient S shown, the field contained 32% air, 62% soft tissue, and 6% bone, yielding a PSF of According to the data, as the quantity of air present in the BEV of the field increased, the corresponding PSF decreased, see Figure 37. This data suggested that the PSF for lung cases was important to account for, having values that could be nearly 6% lower than unity. The wide range of values could be attributed to the many different locations a target might be present. Situations where the target was located at or near the chest wall had less air 120

121 present in the beam path, resulting in a PSF closer to unity. However, the inverse situation where the target was located toward the midline of the chest where the beam must traverse greater amounts of air resulted in a large deviation of the PSF from unity. Figure 37. The relationship between the PBA calculated PSF and the percentage of air in the treatment field for the lung cases. Fitting a linear function to the results yielded Equation 14. PSF = (A) , Eq. 14 Where A is the percentage of air in the field. Due to the wide range of scenarios present in lung treatments, this equation could be used as a cursory guideline to estimate PSF from the amount of air present in the treatment field. 121

122 Conversely, sites such as the prostate that have a significant amount of dense bone (HU greater than 500) in the BEV produced a PSF greater than one. This was due to the fact that dense bone had more scattering power on the incoming fluence of protons, resulting in more scatter delivering dose to the point of interest than in a homogenous water phantom. A histogram of HU values for field 1 of Patient K is shown in Figure 38. Figure 38. A histogram of CT values in the treatment field for Patient K The first two peaks in the histogram are representative of the soft tissue present in the BEV, with the first peak representing the prostate gland. The shoulder of the second peak beginning at about 125 HU is where the soft tissue values turn to bone and the tail of the shoulder starting at 500 HU represents the dense bone. The material breakdown for the particular case of field one of patient K is 0% air, 80.8% soft tissue, 15.1% bone and 4.1% dense bone. The data range of values for sites with bony anatomy present was narrow. For 122

123 the prostate cases, the spread of values of dense bone present was 2.1% resulting in only a 1.2% range of PSF values. The greatest value of PSF for all sites studied was just 1.9% greater than unity. This low variability in the data could create difficulties in analyzing how strongly the effects of dense bone were on the PSF. However, according to Figure 39, there did appear to be a correlation between the amount of dense bone present in the treatment field and the PSF. Figure 39. The relationship between the PBA calculated PSF and the percentage of dense bone in the treatment field for the prostate cases Fitting a curve to the data yielded Equation 15, which could be used as a preliminary estimate to predict the magnitude of the PSF based upon the amount of dense bone in the field. PSF = (B) , Eq

124 where B is the percentage of dense bone in the field. The data clearly showed that the prostate cases studied always resulted in a PSF greater than 1 due to the bone in the field, while the opposite was true for lung cases, where there were large amounts of air present. However, similar conclusions could not be drawn for the other treatment sites examined in this study. Both the brain and pancreas cases could result in a PSF greater than one or less than one. This was because these sites could have bony anatomy and air present in the treatment field, depending on the beam angles and locations of the targets. For example, a pancreas site might have bone in the field from a rib or air in the field from the bowel. This presented two competing factors on the PSF that was difficult to predict using analytical methods. The same was true for brain cases where bone was present from the skull, but there might also be air in the field from a sinus. In some special circumstances, a gap between the immobilization device and the patient surface can introduce an air cavity, which could also influence the PSF as was the case for Patient E, shown in Figure 40. Sites that have multiple types of inhomogeneities will need to be evaluated on a case by case basis to insure that the proper PSF be assigned. 124

125 Figure 40. An example of a beam crossing an air pocket caused by a gap between a patient immobilization device and the surface of the patient CSF Evaluation A general observation of the CSF was that all calculated factors were greater than one. The range of values extended from to This suggested that the presence of the compensator tended to increase the amount of scatter to the point of measurement, thus adding dose. The wide range in CSF factors could be attributed to the degree of irregularity of compensator shape. Figure 41 shows the CSF factors as a function of IQR of compensator thicknesses around the field area that could contribute scatter to the calibration point for seven cases. 125