Link to publication in the UWA Research Repository

Transcription

1 Automated analysis of Varian log files for advanced radiotherapy treatment verification: a multicenter study Hughes, J. (2015). Automated analysis of Varian log files for advanced radiotherapy treatment verification: a multicenter study Link to publication in the UWA Research Repository Rights statement This work is protected by Copyright. You may print or download ONE copy of this document for the purpose of your own non-commercial research or study. Any other use requires permission from the copyright owner. The Copyright Act requires you to attribute any copyright works you quote or paraphrase. General rights Copyright owners retain the copyright for their material stored in the UWA Research Repository. The University grants no end-user rights beyond those which are provided by the Australian Copyright Act Users may make use of the material in the Repository providing due attribution is given and the use is in accordance with the Copyright Act Take down policy If you believe this document infringes copyright, raise a complaint by contacting repository-lib@uwa.edu.au. The document will be immediately withdrawn from public access while the complaint is being investigated. Download date: 22. Apr. 2018

2 Automated Analysis of Varian Log Files for Advanced Radiotherapy Treatment Verification: A Multicenter Study by Jeremy Lee Hughes B.Sc. (Hons), University of Western Australia, 2013 This thesis is presented in partial fulfilment of the requirements for the degree of MASTER OF MEDICAL PHYSICS of The University of Western Australia Jeremy Lee Hughes, 2015 School of Physics (Medical Physics) University of Western Australia

3 Abstract Varian linear accelerators store information of their deliveries in log files which can be extracted from the machine and analysed. These files contain a wealth of information, including the actual and expected position of each leaf throughout the treatment. The analysis of Varian log files has been a growing field of interest ever since the turn of the millennia. The high resolution log files allow deep analysis into treatment plans delivered by Varian treatment machines. This thesis undertook an international multicenter study that sought to find trends in the positional error of the MLC when compared to factors such as; the velocity of the leaves, the gantry angle, the age of the machine, the treatment site, the treatment modality, and the treatment machine. A program was devised in MATLAB to perform statistical analysis on the Varian log files. Additionally another MATLAB program was written to clinically assess these Varian log files. By analysing the Varian log files, this thesis found a positive trend between the error of the MLC to the age of the machine. VMAT treatments had greater error than their dimrt and Step and Shoot counterparts. The prostate treatment site had less error than the H&N and PPN treatment sites. And Varian Truebeam machines possessed less error, by a full order of magnitude, than Varian Clinac ix or Varian Trilogy machines. There was a trend towards greater error for leaves with increased velocity. In general there was greater error for leaves at positions more vulnerable to the force of gravity. This trend was not present for Sliding Gap tests delivered at discrete gantry angles. Varian log files were used to assess thousands of deliveries performed by Varian linear accelerators. The information available in each file allowed trends to be drawn between the positional error of the MLC and the age, model, modality, treatment site, and leaf velocity. I

4 Contents Abstract I Acknowledgements VI List of Figures VIII List of Tables XIV Abbreviations and Acronyms XVII 1 INTRODUCTION What is Cancer? Cancer Treatments Surgery Chemotherapy Radiation Therapy Linear Accelerator Basic Overview Treatment Methods Intensity Modulated Radiation Therapy Volumetric Modulated Arc Therapy Introduction to Quality Assurance Quality Assurance of the Multileaf Collimator Varian log files Aim of thesis BACKGROUND AND LITERATURE REVIEW Treatment Machine Multileaf Collimator Gantry Quality Assurance Tests Ling s Test Ling s Test Ling s Test II

5 2.2.4 Sliding Gap Quality Assurance Gamma Map Analysis Literature Review Varian Machine Models Verification of Varian log files Varian log files for Quality Assurance Limitations of Varian log files Effect of Gravity upon deliveries Overshoot Varian log files for analysis MATERIALS AND METHOD Varian Log Files Dynamic Log Files Truebeam trajectory log files Multileaf Collimator model Matlab Software Importing log files Computing Positional Errors Calculating Leaf Velocity Computing Velocity Errors Computation of Fluence Maps Gamma Map Analysis Linear Accelerators Leaf Error Root Mean Squared Deviation Picket Fence (Ling s Test 1) Step and Shoot and dimrt VMAT Comparing Treatment Methods - Clinac ix Positional Errors due to Velocity Inner/Outer Leaves III

6 3.7 Velocity Leaf Error VMAT Sliding Gap Treatment Modalities Gamma Map Analysis Gravity Picket Fence (Ling s Test 1) Sliding Gap VMAT Age Statisical Analysis RESULTS Development of software for automatic Varian log file analysis Confirmation of MATLAB software Leaf Error Picket Fence (Ling s Test 1) Step and Shoot dimrt VMAT Comparing Treatment Methods - Clinac ix Positional Errors due to Velocity Inner/Outer Leaves Velocity Leaf Error VMAT Sliding Gap Treatment Modalities Gamma Map Analysis Gravity Picket Fence (Ling s Test 1) Sliding Gap VMAT IV

7 4.8 Age Errors and Treatment Machine Model Errors and Treatment Site Error Distribution of Prostate Treatments Errors and Different Treatment Modalities Errors and Leaf Velocity Errors and Inner and Outer Leaves Gamma Map Analysis Errors and Gantry Angles Errors and Treatment Machine Age CONCLUSION Future Work Appendix 107 V

8 Acknowledgements As I sit down to write my Acknowledgements a bittersweet feeling washes over me. On one hand I am happy to be finally done with this thesis that was the cause of many tea fuelled sleepless nights but on the other, a part of me is sad to see it go. But I suppose that s Stockholm Syndrome for you. Of course this thesis would never have been completed if it was just me tapping away at the keyboard. I had a lot of help in order to complete this project. A lot of help. I would like to thank all my supervisors, Dr Pejman Rowshanfarzad, Prof Martin Ebert, and Prof Mike House, for all the hard work they put into this project so that it was an otherwise rewarding and enjoyable experience. I would like to especially thank Dr Pejman Rowshanfarzad for all the advice he had to give and all the insights he gave. Thank you for working tirelessly away and giving me direction whenever my biannual freak out, What am I even doing? sessions came and went. You were a lot more help than you realise. I would also like to thank the wonderful people over in Belfast, Dr Conor McGarry and Christina Agnew, for their massive windfall of treatment files without which this thesis would be severely diminished. My thanks also goes out to Dr Mahsheed Sabet and Nikki Caswell (Chief Medical Physicist) for their immense help in supplying me with treatment files from all around the country without which my thesis would be non existent. Also a thank you to Dr Alison Scott, Dr Sivakumar Somangili, Anthony Walsh and Garry Grogan for all your help and input. The clinical point of view was very much welcomed and appreciated. VI

9 And it would be remiss of me not to mention the wonderful people that made up the research group I was a part of for two years. It was a privilege to belong to such an incredible bunch of people who always inspired me to work harder and get smarter. I not only thank my family but offer my apologises for the long hours spent tapping away at this infernal machine (especially near the end). Fear not though, after this thesis is submitted I shall return to my duties with only the solace of long term unemployment to look forward to. Penultimate thanks goes to Blake Segler who printed out the final copy of this thesis while I was in Japan and to Mike and Pejman who sorted out the rest. And finally, my thanks to the WA Department of Health whose funding allowed me to undertake this Master s degree, without which I would not know what I would be doing. And as this sentence culminates two years worth of work, to you dear reader, it is just the beginning. So enjoy as much as humanely possible otherwise, what s the point? VII

10 List of Figures 1.1 Healthy cells undergo apoptosis and die off whereas malignant cells bypass apoptosis and grow in number. Figure courtesy from the National Cancer Institute [1] A megavoltage linear accelerator. Image courtesy of NELCO Worldwide A basic schematic of a linear accelerator. Taken from Rowshanfarzad [5] A Varian MLC plus carriage Dose distribution for IMRT (a,b) and VMAT (c,d) for prostate radiotherapy. Figure courtesy of Department of Medical Physics, Royal Surrey County Hospital, UK A head on schematic of adjacent MLC leaves from Siemens, Elekta, and Varian showing the tongue and grooves of each leaf. The protrusions minimise interleaf leakage. Image modified from [29] Varian s M80 MLC. The leaves move independently to collimate the beam into different shapes for treatment. Image courtesy of Varian Basic schematic of M120 MLC. Individual leaves extend and retract from opposing banks A picture of two different leaf motors, one for the inner leaves and one for the outer leaves of the M120 MLC. Image taken from Institute of Radio Oncology, KFJ Hospital Vienna [34] The encoder quadrature channels on each leaf motor. Image taken from Varian [33] Radiation penetration: (a) Penetration dependence upon leaf position is minimised. (b) Penetration through square leaves is dependent on leaf positioning which is why Varian MLC leaves are curved The gantry rotates in the horizontal axis and the collimator rotates in the vertical axis. The point in space where the radiation beams intersect when the Gantry is rotated is called the radiation isocenter Dose distribution for Picket Fence QA recorded by an EPID VIII

11 3.1 Different Position Readout Scale Conventions. Modified from High Energy C-Series Clinac, Installation Product Acceptance Booklet Schematic of a typical 120 leaf MLC showing the relative positions of the leaf banks and upper and lower jaws. Leaf 54 of Bank B and leaf 8 of Bank A are both extended across the centreline A basic workflow schematic of the importation of Varian log files The Expected and Actual positions from Varian log files of a leaf during a step and shoot treatment. When the beam is off the expected position jumps to the next segment whereas the actual position slowly changes A basic workflow schematic of how the Fluence Maps are created Basic structure of how the program works The velocity of the leaves with respect to time for a Varying Sliding Gap QA computed by a) Varian log file analysis, and b) EPID analysis. The white circles highlight the discrepancy of leaf number The RMSD of 32, 25, and 1805 Picket Fence deliveries were computed for Trilogy, Clinac ix, and Truebeam machines respectively. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers Error histogram of 1643 Step and Shoot prostate treatments delivered on machine ix 6. In total points were assessed. A double gaussian distribution was fitted with R 2 = Errors of 7 different dimrt prostate treatments from Clinac ix machines. A double gaussian distribution was fitted with R 2 = IX

12 4.5 RMSD of VMAT treatments at treatment sites Head and Neck (H&N), Prostate Pelvic Node (PPN), and Prostate (PROS), for treatment machines a) True 1 and b) True 2. A total of 76 H&N, 182 PPN, and 203 PROS treatments were analysed for True 1 and 444 H&N, 570 PPN, and 193 PROS treatments were analysed for True 2. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers The RMSD of VMAT Head and Neck treatments for treatment machines Clinac ix (32 files) and Truebeam (520 files) RMSD of different treatment modalities for Clinac ix Prostate Step and Shoot, 7 Prostate dimrt, and 32 Head and Neck VMAT treatments were analysed. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers Absolute errors from three different Sliding Gap QA. The Sliding Gap QA was performed 3 times for 30mm/s Sliding Gap QA and 4 times for 10mm/s, and 20mm/s Sliding Gap QA. The errors of each delivery were: a) uncorrected, b) corrected for the 55+ms delay present in older Varian systems. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers RMSD of inner and outer leaves for Bank A and Bank B of: a) Clinac ix and Trilogy machines, and b) Truebeam machines. 13, 19, 19, 6, 732, 704, and 369 picket fence deliveries were analysed for ix 1, ix 2, Tri 1, Tri 2, True 1, True 2, and True 3 respectively X

13 4.10 The Velocity RMSD for VMAT treatments delivered at different treatment sites on True 1 (left) and True 2 (right). A total of 76 H&N, 182 PPN, and 203 PROS treatments were analysed for True 1 and 444 H&N, 570 PPN, and 193 PROS treatments were analysed for True 2. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers The error of velocity in different Sliding Gap QA deliveries on ix 3. Four 10mm/s, four 20mm/s, and three 30mm/s sliding gap deliveries were analysed. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers Velocity RMSD of different treatment modalities for Clinac ix. 7 dimrt (PROS) and 32 VMAT (H&N) treatments were analysed. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers Gamma map analysis analysing the expected and actual fluence of a H&N treatment from ix 6 using 3%/3mm. Colour intensity details how similar the two fluence maps are. The formula for calculating these values was presented in section Polar plot of the error (mm) in the radial component that has been averaged over 2 degree control points, vs the gantry angle (IEC convention) for a) ix 1 (13 tests) b) Tri 1 (14 tests) and c) True 1 (732 tests). This was further split into the error present on i) Bank A, ii) Bank B, and iii) the leaf gap width XI

14 4.15 The error (mm) for Sliding Gap QA delivered at four different gantry angles on ix 3 for a) Bank A, and b) Bank B. Leaves moved with velocity 10mm/s. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers The error (mm) for Sliding Gap QA delivered at four different gantry angles on ix 3 for a) Bank A, and b) Bank B. Leaves moved with velocity of 20mm/s. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers A histogram of the collimator angles (IEC convention, degrees) of 1451 VMAT treaments delivered on Truebeam machines Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for treatments with collimator angle (IEC convention) i) 30, ii) 330, and iii) all collimator angles. These were also split into a) Bank A and b) Bank B. There were 540, 398, and 1451 treatments for collimator angles 30, 330, and all collimator angles respectively RMS error for of routine Picket Fence deliveries vs date delivered for: a) True 1, b) True 2, c) True 3, and d) all of the above. Results were averaged on a monthly basis. A linear trend was fitted to each of the machines with R-value: a) R = , b) R = , and c) R = Error histogram of 1643 Step and Shoot prostate treatments delivered on machine ix 6. In total points were assessed.a double Gaussian was also fitted Errors of 7 different dimrt prostate treatments from Clinac ix machines. A double Gaussian was also fitted Errors of 7 different dimrt prostate treatments from Clinac ix machines split into Bank A and Bank B errors XII

15 4.23 The position of the MLC at gantry angles 90 and 270 for Ling s Test 1. The collimator angle is at Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 6 tests were analysed Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 19 tests were analysed Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 704 Picket Fence tests of True 2 were analysed Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 369 Picket Fence tests of True 3 were analysed XIII

16 List of Tables 1.1 Verification and delivery parameters for a VMAT treatment planned on Pinnacle SmartArc Module. Higher Γ values indicate a more accurately delivered treatment. Table taken from Rowbottom et al [11] Structure of the header in a DynaLog file. Modified from DynaLog File Viewer Reference Guide Structure of the contents in a DynaLog file. Modified from DynaLog File Viewer Reference Guide Structure of the header in a Truebeam trajectory log file. Modified from Truebeam Trajectory Log File Specification [89] Structure of the subbeam in a Truebeam trajectory log file. Modified from Truebeam Trajectory Log File Specification [89] Summary of linear accelerators used in this study Comparison between Argus QA and Timberwolf RMSD of Picket Fence deliveries. Results are presented as the mean of the RMSD and standard deviation RMSD of VMAT deliveries at different treatment sites for True 1 and True 2. Results are presented as the mean of the RMSD and standard deviation RMSD of H&N VMAT deliveries for Truebeam and Clinac ix machines. Results are presented as the mean of the RMSD and standard deviation The RMSD of Sliding Gap QA moving at different velocity presented as mean and standard deviation. The corrected deliveries were corrected for the 55ms+ delay present in older Varian systems a) RMSD and b)std of inner and outer leaves of Bank A and Bank B for different machines. 13, 19, 19, 6, 732, 704, and 369 picket fence deliveries were analysed for ix 1, ix 2, Tri 1, Tri 2, True 1, True 2, and True 3 respectively XIV

17 4.7 Velocity RMSD of VMAT deliveries at different treatment sites for True 1 and True 2. Results are presented as the mean of the Velocity RMSD and standard deviation The Velocity RMSD of Sliding Gap QA moving at different velocity presented as mean and standard deviation Gamma passing rates calculated through Varian log file analysis for Head and Neck (H&N), Prostate Pelvic Node (PPN), and Prostate (PROS) treatments. Results are presented as mean and STD Gamma passing rates calculated through Varian Log file analysis for 20 LT1, 44 LT2, and 17 LT3 QA tests. All tests were delivered on Clinac ix machines. Results are presented as mean and STD P-values of Student T-tests between the averaged Control Point error located in 40 degree arcs centered on gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Picket Fence tests from ix 1 were analysed and data was only taken into account if the leaves were stationary P-values of Student T-tests between the averaged Control Point error located in 40 degree arcs centered on gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Picket Fence tests from Tri 1 were analysed and data was only taken into account if the leaves were stationary P-values of Student T-tests between the averaged Control Point error located in 40 degree arcs centered on gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Picket Fence tests from True 1 were analysed and data was only taken into account if the leaves were stationary The error for Sliding Gap QA (10mm/s) delivered at four different gantry angles on ix 3 on Bank A and Bank B. Results are displayed as mean and STD XV

18 4.15 The error for Sliding Gap QA (20mm/s) delivered at four different gantry angles on ix 3 on Bank A and Bank B. Results are displayed as mean and STD P-values of Student T-tests between the error of Sliding Gap treatments (20mm/s) delivered at gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Comparison of treatment modalities by two different groups XVI

19 Abbreviations and Acronyms AAPM American Association of Physicists in Medicine ACMP American College of Medical Physics CBCT Cone-beam computed tomography D Dose Tolerance DC Direct current DICOM Digital Imaging and Communications in Medicine dimrt Dynamic Intensity Modulated Radiation Therapy DTA Distance to agreement DVH Dose volume histogram DynaLog Dynamic Log EBRT External Beam Radiation Therapy EMF Electromotive force EPID Electronic Portal Imaging Device GUI Graphical User Interface HD MLC High Definition MLC H&N Head and Neck IAEA International Atomic Energy Agency ICRP International Commission on Radiological Protection ICRU International Commission on Radiation Units and Measurements IEC International Electrotechnical Commission IMRT Intensity Modulated Radiation Therapy Linac Linear Accelerator LT1 Ling s Test 1 LT2 Ling s Test 2 XVII

20 LT3 Ling s Test 3 MLC Multileaf Collimator M52 MLC 52-leaf Millennium MLC M80 MLC 80-leaf Millennium MLC M120 MLC 120-leaf Millennium MLC MU Monitor unit NCRP National Council on Radiation Protection and Measurements OAR Organs at risk PCB Printed Circuit Board PPN Prostate Pelvic Node PROS Prostate PTV Planning treatment volume PWM Pulse Width Modulation QA Quality Assurance RMS root mean squared RMSD root mean squared deviation RMSE root mean squared error TLD Thermoluminescent Dosimeter TPS Treatment planning system VMAT Volumetric Modulated Arc Therapy XVIII

21 CHAPTER 1 INTRODUCTION 1.1 What is Cancer? Cancer is a serious disease of the body s cells. Cells normally divide and grow in a regulated manner. Healthy cells have preprogrammed cell death or apoptosis ingrained into their genetic code so that there is never an over-abundance of cells. Instead of an overcrowded environment a percentage of cells die off before they can subdivide further ensuring an environment suitable for cell growth. However the genetic code of a cell can be altered and damaged whether simply by random mutations or from external factors, one of which is radiation. If these mutations survive the multiple checks put in place by the body to stop such a thing occurring then their mutations grow and spread to their daughter cells. Eventually the mutation gets to a point of uncontrolled cell growth, this is known as cancer. The cells now no longer perform their original task and instead focus on dividing and growing. Additionally they mutate so that they no longer undergo preprogrammed cell death meaning they effectively live for however long the body can sustain them. The main difference between healthy and cancerous cells is represented graphically in figure 1.1. The cancer cells grow and multiply such that the healthy cells have restricted access to the blood supply causing them to die off. During the later stages of cancer, the primary tumour spreads microscopic cancer cells to the rest of the body, this is known as metastasis. They travel mainly through the lymphatic or circulatory system to infect other areas. The spread and growth of the tumours eventually leads to the death of the patient. Cancer is a leading cause of death in Australia accounting for approximately 3 in 10 1

22 Figure 1.1: Healthy cells undergo apoptosis and die off whereas malignant cells bypass apoptosis and grow in number. Figure courtesy from the National Cancer Institute [1]. deaths [2]. It is a serious disease and there are multiple treatments available that try to cure the patient of cancer. These fall under three main categories; surgery, chemotherapy, and radiation therapy. These treatments are often performed in conjunction with each other for maximum effect. 1.2 Cancer Treatments Surgery Surgery is an effective measure against cancer in which the tumour is simply excised from the patient. This method is most effective during the early stages of cancer growth when the tumour is still localised to a small area. During the later stages of cancer development the tumour grows and metastasises meaning the cancer spreads throughout the body via microscopic cancer cells. Only one microscopic cancer cell would have to survive and regrow for the treatment to be considered a failure [3] which is why surgery is normally confined to the early stages of tumour development. That being said surgery is still used in conjunction with other treatment options in the later stages of cancer growth. 2

23 1.2.2 Chemotherapy Chemotherapy uses drugs to combat cancer. Traditional chemotherapy drugs are chosen so that they target and kill rapidly dividing cells, which is one of the main properties of cancer cells. Unfortunately this means that healthy cells that divide at a rapid rate are also targeted and killed. These include cells in the digestive tract, hair follicles, and bone marrow and as such the patient may experience a whole raft of side effects including nausea and vomiting, loss of hair, and myelosuppression. Fortunately these healthy cells repair themselves at a greater rate than the cancer cells so although the patient will experience negative side effects to the treatment the majority are only temporary. Chemotherapy is often used in conjunction with radiation therapy [4] Radiation Therapy Radiation Therapy, or Radiotherapy, is an effective yet potentially hazardous treatment for cancer. Radiotherapy works on the basis of the ionisation and destruction of malignant cells by bombarding the tumour with ionising radiation. Ideally all malignant cells will be eradicated from the patient during treatment. However, as radiation does not discriminate between healthy and malignant cells, healthy cells will also be affected and harmed. Targeting the cancer cells whilst sparing the healthy cells then becomes a top priority and this is achieved in many different ways that differ depending upon if the radiation source is internal or external to the body. Internal radiation therapy, also known as brachytherapy, is the placement of sealed radioactive sources near or in the cancerous tumour to deliver high doses to the tumour with a rapid dose falloff to the surrounding healthy tissue. Whilst effective, internal radiation therapy is an invasive treatment. External beam radiation therapy (EBRT) is a subset of radiation therapy that delivers high energy radiation to a targeted area from an external source, usually a linear accelerator. The patient lies on a treatment couch and the linear accelerator delivers the treatment as ascribed by a Radiation Therapist. The aim is to destroy the malignant cells whilst doing as little damage as possible to the surrounding healthy tissue. This is achieved by the careful planning of the treatment as well as ensuring 3

24 Figure 1.2: A megavoltage linear accelerator. Image courtesy of NELCO Worldwide. that this treatment is accurately delivered to the patient. The linear accelerator must perform within certain specifications to provide a clinically safe treatment. 1.3 Linear Accelerator Linear accelerators, or more colloquially linacs, are an integral component of EBRT due to their ability to produce and deliver X-rays or high energy electrons. Linacs work by accelerating charged particles, such as electrons, to high energies along a linear tube. This electron beam can be used to treat superficial tumours or alternatively be used to hit a metal target thus producing bremsstrahlung radiation and characteristic X-rays. The radiation is then delivered to the patient in such a way to maximise the damage to the tumour, but minimise the impact on the healthy cells. Although there are variations to the linear accelerator s design, those used in EBRT are fundamentally similar. For the purposes of this thesis we will only be focusing on linear accelerators that produce radiation Basic Overview Figure 1.3 showcases a basic schematic diagram of a linear accelerator. The DC power supply is injected into the pulsed modulator which converts the continuous input into 4

25 Figure 1.3: A basic schematic of a linear accelerator. Taken from Rowshanfarzad [5]. pulse form. The high voltage pulsed power is then injected into the RF power source, either a Klystron or Magnetron, which causes the production of pulsed electromagnetic waves that are then injected into the accelerating waveguide. The pulsed power from the modulator is also injected into the electron gun resulting in a pulsed stream of electrons also entering the accelerating waveguide. The electrons enter the accelerating waveguide with an initial energy of approximately 50keV [6] where they interact with the pulsed electromagnetic waves and emerge from the accelerating waveguide with a vastly increased energy, on the order of a mega electron volt. The electron beam then travels through the treatment head where it strikes a metal target to produce bremsstrahlung radiation and characteristic X-rays, with energy on the order of mega voltage. The radiationis then collimated by the primary collimator and in Varian (Varian Medical Systems, Palo Alto, CA) linacs the radiation is further collimated through slabs of tungsten known as Jaws and more recently a component known as a Multileaf Collimator (MLC). The radiation is then used to treat the patient. It is important to note that the treatment head resides upon a gantry which can rotate around the patient throughout the duration of the treatment. The added degree of freedom of the gantry rotation allows more conformal and technical treatment options. The MLC is affixed to the treatment head to further collimate the radiation beam during treatment. The idea behind the MLC is relatively simple. Affixed to two opposing parallel carriages, also known as bank A and bank B, are rows of tungsten 5

26 Figure 1.4: A Varian MLC plus carriage. leaves that can move one dimensionally towards or away from the opposite carriage. These leaves can move independently from each other and as such a multitude of shapes may be formed to collimate the radiation. An image of the MLC and subsequent carriages can be found in figure 1.4. The MLC itself can also rotate about the axis of the radiation beam allowing for a greater degree of freedom. This is known as the collimator rotation. The angle of the collimator, or MLC, generally remains constant throughout delivery. More detail pertaining to the physical aspects of the MLC can be found in section Treatment Methods Intensity Modulated Radiation Therapy In traditional EBRT the intensity profile of the radiation beam normal to the beam axis is relatively uniform. Wedges or compensators are sometimes used to modulate the field such that more radiation is delivered to the planned treatment volume (PTV) and such that the organs at risk (OARs) are spared the brunt of the radiation which reduces the acute and late toxicity [7]. This is known as intensity modulated radiation therapy (IMRT). The invention of the MLC made IMRT easier to perform due to the reduction of the wedges and compensators that were needed to be manufactured. Instead shapes can simply be made with the MLC that emulate the wedges and compensators. Modern 6

27 linacs generally do not use physiacl wedges which removes the problem of manufacture however MLCs are still an easier, more practical implementation. The most basic form of IMRT is known as the Step and Shoot method. The Treatment Planning System (TPS) optimises the treatment through the design of optimal field apertures, appropriate beam directions, the number of fields, beam weights, and MLC leaf positions. The linac rotates to the first gantry angle (normally there are around 7 different fields in total), moves the MLC leaves into position, and delivers radiation until the dose for that segment has been reached. The beam is then held and the linac rotates to the second gantry angle whilst moving the MLC leaves into their new position. The beam then stops being held and the appropriate dose is delivered. This process repeats until the treatment is completed. The discrete radiation beams of the treatment is further split into subfields that can have different intensity from each other (hence intensity modulated radiation therapy). This aids the sparing of OARs and increases the conformity to the targeted area [7]. A slightly more complex version of IMRT is known as dynamic IMRT (dimrt). It is similar to the Step and Shoot method in many ways however when the beam is on and delivering radiation, the MLC leaves are not constrained to a static location. During delivery the MLC leaves move unidirectionally and are optimised to minimise treatment time [6] Volumetric Modulated Arc Therapy Volumetric modulated arc therapy (VMAT) is an even more complex form of treatment. Instead of limiting the treatment to a discrete number of gantry angles the beam continuously delivers radiation as the gantry rotates around the patient. The MLC leaves are also moving bidirectionally to continuously shape the radiation. The continuous nature of this treatment not only increases the conformity to the PTV but it also reduces patient treatment time [8]. The treatment varies the gantry rotation speed, dose rate, and the collimation to deliver a highly conformal dose [7]. A comparison between the dose distribution as a result of VMAT treatments and IMRT treatments can be seen in figure 1.5. This method allows for greater control in treatment planning and dose distribution 7

28 Figure 1.5: Dose distribution for IMRT (a,b) and VMAT (c,d) for prostate radiotherapy. Figure courtesy of Department of Medical Physics, Royal Surrey County Hospital, UK. but is computationally more expensive to produce than the other treatment types. Additionally due to its continuous delivery the dose gradient at the edge of the field is not as sharp as treatments with discrete gantry rotations [9]. Treatments that deliver from a discrete number of gantry angles are split into segments however as there are a continuous number of gantry angles that are used for delivery in VMAT a separate method to divide the treatment into smaller components for treatment planning is introduced. The VMAT treatment is divided into subarcs that are known as control points which outline the MLC shape, MLC segment dose, and a gantry-angle window across which each shape sweeps dynamically [10]. VMAT treatments typically consist of 177 control points with each control point roughly spanning a subarc of 2 degrees. Of course the number of control points in the total treatment can be changed. If the total number of control points increases then the angle that the subarcs span decrease and you can hypothetically create a better treatment plan due to the higher number of control points allowing for greater control over the evolution of MLC shape. However this has limitations in the physical velocity of the leaves as well as the downside of increased computational hours and treatment time. A table showing the different delivery times and their accuracy, in the form of gamma map analysis can be found in table 1.1 for Pinnacle TPS. As the number of control points increases the accuracy of the treatment increases but so does the delivery time of the treatment. For 8

29 Table 1.1: Verification and delivery parameters for a VMAT treatment planned on Pinnacle SmartArc Module. Higher Γ values indicate a more accurately delivered treatment. Table taken from Rowbottom et al [11]. this system a compromise of 121 control points would be optimal [11]. As a consequence of the continuous treatment beam delivery, the treatment delivery time in VMAT treatments is used much more efficiently than Step and Shoot techniques. Therefore the treatment delivery time for VMAT treatments are much shorter when compared to Step and Shoot methods [12]. A shorter treatment delivery time is beneficial to both the patient and Radiation Therapist as an increased treatment delivery time can impact on the patient s comfort on the treatment couch, the reproducibility of the treatment position and if the delivery uses discrete gantry angles then there may be some intrafraction motion [7]. Also with a reduced treatment delivery time there can be a higher throughput of patients. Abbas et al. recorded VMAT treatment times 55% less than their IMRT counterpart (median of 120s versus median of 254s [12]). Additionally Varian records a VMAT treatment time of 1.5min compared to the equivalent IMRT treatment time of 5.5min but as they rely on hospitals upgrading to their new systems these numbers should be viewed sceptically. [13]. In addition to the reduced treatment delivery time, VMAT treatments also have a reduction in monitor unit (MU) usage when compared to IMRT techniques[7]. MUs are used to measure the amount of radiation delivered and hence VMAT treatments reduces the potential for secondary malignancies arising due to the treatment. There is still concern that the greater amount of healthy tissue being irradiated may offset this reduction but more research still has to be done on the matter. VMAT is a relatively new treatment technique that has greater conformity and greater sparing of the OARs when compared to basic conformal radiation therapy tech- 9

30 niques [7]. There is still debate on the whether this is also true when VMAT is compared to IMRT treatments. Many different studies have come to the conclusion that for the most part the conformity of the dose distribution is largely the same, as is the sparing of the OARs [7]. However nearly all studies have reached the same conclusion that VMAT treatments provide highly conformal treatments with a reduced treatment delivery time and reduced MUs when compared to IMRT [7], [14] [22]. 1.5 Introduction to Quality Assurance Quality Assurance (QA) is a set of policies and procedures implemented to objectively monitor the quality and appropriateness of patient care. This is achieved by ensuring the mechanical and dosimetric characteristics of a linac lie within an acceptable range of a baseline value which in turn ensures that the patient treatments are delivered within specified spatial and dosimetric tolerances. Without QA patients may receive harmful or irrelevant treatments through either complacency or neglect. It is for this reason that many professional organisations have proposed their own QA guidelines for centres to follow. These organisations include the International Atomic Energy Agency (IAEA), the American Association of Physicists in Medicine (AAPM), and the American College of Medical Physics (ACMP). These guidelines in turn utilise data published by the International Commission on Radiological Protection (ICRP), the International Commission on Radiation Units and Measurements (ICRU), the International Electrotechnical Commission (IEC), and the National Council on Radiation Protection and Measurements (NCRP). Although these guidelines exist it is ultimately the choice of the individual centres as to what QA program they wish to implement, as long as international or national standards are fulfilled. A QA program encompasses many aspects ranging from how the equipment is performing to pre and post treatment evaluation. For the sake of relevancy this thesis will discuss QA focusing on the quality assurance of the equipment as well as treatment evaluations. Guidelines published by AAPM [23] suggest different checks are to be performed on linear accelerators annually, monthly, or daily depending on how often the part that 10

31 is being assessed is prone to failure, its clinical impact, and how long it will take to perform the test. This periodic QA is mainly in place to assure the staff that their measurements are consistent with the machine s performance from a mechanical and dosimetric standpoint. It ensures that the linear accelerator is delivering radiation and moving as per the staff s instructions, within certain tolerance percentages. It is for this reason that periodic QA is also known as linac QA. Some examples of these tests include checking the functionality of the multileaf collimator, the collimator size indicator, and the electron output constancy. Whilst periodic QA is implemented to keep the machine s performance in check, a different form of QA, patient-specific QA is implemented to verify that the treatment being delivered to the patient is the treatment that was planned by the physicists. If the machine is not physically capable of delivering the prescribed treatment, or the machine is incorrectly calibrated, patient-specific QA is designed to pick up on these inconsistencies. Patient-specific QA normally takes the form of a dry run, that is the treatment is delivered where the patient is replaced by a dosimeter and the resulting dose distribution is compared against the desired outcome. This comparison is usually performed by gamma map analysis, first suggested by Low et al. [24] in Gamma map analysis measures how well two different dose distributions agree with each other. To function effectively as a radiation dosimeter the dosimeter in question must possess a physical property that is dependent upon the measured dose quantity. They then can be calibrated to give results that are useful in a clinical setting. Additionally the properties of accuracy and precision, linearity, dose or dose rate independence, a flat energy response (although in reality corrections have to be made), directional dependence and high spatial resolution are desirable for a radiation dosimeter [9]. 1.6 Quality Assurance of the Multileaf Collimator The MLC is a complicated instrument that is responsible for correctly collimating the radiation beam. Therefore it is important that it performs to the best of its abilities and does not fall out of calibration otherwise a hazardous delivery could be made. Therefore the alignment of the MLC leaves is an area of concern as misalignment of the leaves results in treatments being delivered incorrectly. Different quality assurance 11

32 tests are performed to ensure this does not occur. One such test is performed by a quality assurance tool built into the MLCs that are produced by the company Varian. This exists as a collimated beam of light that is directed across the paths of the leaf ends. Each leaf is moved individually up until they block the beam of light and their default position is reset. This ensures the leaf positions are calibrated to the same position every time this process is run, which is every time the machine is turned on [25]. There exists independent methods to verify the leaves positions. These take the form of dosimetric quality assurance. Two such examples are Ling s Test 1, also known as the Picket Fence test, and Sliding Gap QA. The leaves are moved in a predetermined fashion while radiation is being continuously delivered to a detector. When viewing the resulting dose map from film, EPID, or ion chamber array, the movement of the MLC leaves make it obvious when a leaf is misaligned. If a leaf is out of alignment then the dose will not be uniform with the surrounding leaves. Corrections can then be made. 1.7 Varian log files The analysis of the quality assurance tests of the MLC have traditionally been performed dosimetrically however there exists an alternative method. Linear accelerators produced by Varian log information at regular intervals during treatment. These are exported to dynamic log files (DynaLog files) or Truebeam trajectory log files depending on the machine model. Each log file contains useful information such as the normalised MUs delivered, whether the radiation beam was delivering or not, the angle of the gantry, and most importantly, the expected and actual location of each individual leaf of the MLC, for both bank A and bank B at each time step. By reading these logs into appropriate software the errors of each individual leaf can easily be obtained. By assessing the Varian log files of QA tests instead of assessing the QA tests dosimetrically, the user introduces the variable of time. Although analysing the QA tests dosimetrically can tell you whether a leaf does fail, it cannot tell you when this occurs as well as what else the machine is doing at the time. There exists multiple commercial treatment software that analyse Varian log files. 12

33 In fact Varian Medical Systems have produced their own software entitled DynaLog File Viewer, however DynaLog File Viewer is limited in its functionality. It assesses the error RMS for each leaf, which gives the user an insight as to which leaf needs replacement, but it does not give individual data of the leaves with respect to time. Other independent third party commercial software, such as MobiusFX, or Argus QA, provides a more rigorous analysis with the options to view the error against time or gantry angle. These programs also produce fluence maps from the log files and perform subsequent gamma map analysis between the expected fluence and the actual fluence. Such software is designed and intended for clinical practice and as such is not programmed for large scale statistical analysis. Moreover the user has no way of modifying the code to perform their own analysis using the information present in Varian log files. If the user wished to do so then their only course of action would be to create their own program. 1.8 Aim of thesis The principle aim of this thesis is to analyse and compare multiple different treatment methods and machines through the use of Varian DynaLog and Truebeam trajectory log files. We hypothesise that different types of leaf errors could be associated with the machine model of linear accelerator used in treatment, the treatment complexity, and the treatment site that is being targeted. Multiple factors will be analysed including the effect of the gantry angle on the positional and velocity errors of the individual leaves in the MLC. Furthermore, analysis of different treatment techniques and treatment machine models will be undertaken to ascertain if factors such as velocity, age, leaf number, or treatment location have any bearing on the positional or velocity errors of the MLC. Gamma map analysis will also be performed on the actual and expected fluence maps constructed from log files as an added measure of how well a treatment was delivered. Additionally this thesis aims to construct an accessible program that analyses Varian log files for clinical use that is freely available. 13

34 CHAPTER 2 BACKGROUND AND LITERATURE REVIEW 2.1 Treatment Machine Multileaf Collimator The multileaf collimator (MLC) was developed in Japan in the 1960s by Takahaski [26] and in the past two decades has become a widely used piece of equipment when using linear accelerators. The MLC is affixed to the treatment head to further collimate the radiation beam during treatment. The idea behind the MLC is relatively simple. Affixed to two opposing parallel carriages, are rows of tungsten leaves that can move one dimensionally towards or away from the opposite carriage. These leaves can move independently from each other and as such a multitude of shapes may be formed to collimate the radiation. Unfortunately the benefit of numerous independently moving leaves comes at a cost. Radiation leaks through adjacent leaves of the MLC [27]. To combat this, each leaf is designed with a protrusion on one side, a tongue, and a depression on the other, the groove. Each tongue interlocks into the adjacent groove of the neighbouring leaf as can be seen in figure 2.1. This arrangement minimizes interleaf transmission [28]. This arrangement can cause tongue and groove underdosage which results in a reduction of dose by as much as 10-25% in the treatment fields for static and dynamic IMRT [30]. To overcome this, leaf sequencing algorithms have been created that either synchronise the movement of adjacent leaves such that this effect is reduced [31] or by 14

35 Figure 2.1: A head on schematic of adjacent MLC leaves from Siemens, Elekta, and Varian showing the tongue and grooves of each leaf. The protrusions minimise interleaf leakage. Image modified from [29] a partial synchronisation technique based on the sum of the lengths of all tongue-andgroove junctions [32]. These techniques decrease underdosage but increase the number of segments that have to be delivered. TG-50, released by AAPM suggests that the average leaf and interleaf transmission should be less than 2% [25]. Physical Characteristics The 120-leaf Millennium MLC (M120 MLC) is produced by Varian and has 60 tungsten leaves on each carriage (also known as bank A and bank B). The leaves on each bank are classified as either inner or outer leaves depending on their position on the bank and are numbered by the machine in sequential order as seen in figure 2.3. Leaf numbers 1 through 10 are considered outer leaves and Leaf numbers 11 through to 50 are considered inner leaves. Leaf numbers 51 through to 60 are also considered outer leaves. Thus each bank has 20 outer leaves sandwiching the 40 inner leaves. Most outer leaves are 1.0cm wide whereas inner leaves have a width of 0.5cm. Leaf pairs 1 and 60 both have width of 1.4cm. This allows for greater control over the collimation in the inner region which is where the majority of treatments take place. The M120 MLC has a field size of 40 x 40cm at isocenter level which is 100cm away from the source. There exists other versions of the MLC such as the 52-leaf Millennium MLC (M52 MLC) or 80-leaf Millennium MLC (M80 MLC) however as their name suggests, they possess fewer leaves than the M120 MLC. Their field size is 26 x 40cm for M52 MLC and 40 x 40cm for M80 MLC however all of the leaves have a width of 1.0cm meaning 15

36 Figure 2.2: Varian s M80 MLC. The leaves move independently to collimate the beam into different shapes for treatment. Image courtesy of Varian. Figure 2.3: Basic schematic of M120 MLC. Individual leaves extend and retract from opposing banks. 16

37 Figure 2.4: A picture of two different leaf motors, one for the inner leaves and one for the outer leaves of the M120 MLC. Image taken from Institute of Radio Oncology, KFJ Hospital Vienna [34]. less control is provided over the collimation. These leaves all have sufficient thickness to attenuate the X-rays to less than 2% [25]. There exists another variation of the MLC known as the high definition MLC (HD120 MLC). This MLC also has 60 tungsten leaves on each bank but it consists of 32 inner leaves of width 0.25cm and 2 sets of 14 outer leaves of width 0.50cm. This allows for even greater control at the reduced field size of 22 x 40cm. However for the purposes of this thesis we will be focusing on the most common M120 MLC. Each carriage in the M120 MLC weighs 36kg. The weights of the inner and outer leaves are approximately half a kilogram and a kilogram respectively. Leaf Motors Each individual leaf has its own individual permanent magnet DC motor which can be seen in figure 2.4. The motors work by drawing current equal to the motor voltage. As the motors increase speed, the motors produce an increasing back-emf voltage which decreases the voltage across the motor. As a result the current being drawn into the motor also decreases [33]. When a motor drives a leaf the amount of torque it experiences is increased. Additionally for a permanent magnet DC motor that is experiencing a constant voltage, the motor speed is proportional to the torque. This means that when a constant voltage is applied to a motor, the speed of the leaf that requires excessive torque is relatively slower than normal. This can be taken advantage of by the addition of a gearbox in which the torque can be increased or decreased. This results in a slower or faster motor speed and hence slower or faster leaf speeds [33]. On the shaft of each motor lies a quadrature encoder. This is made up of two channels, Channel A and Channel B which pulse out of phase with each other when 17

38 Figure 2.5: The encoder quadrature channels on each leaf motor. Image taken from Varian [33]. the motor rotates. The direction of motion of the leaves can be obtained from checking if Channel A leads Channel B, in which case the motor is rotating clockwise, or if Channel B leads Channel A, in which case the motor is rotating counter clockwise. A basic schematic of this setup can be found in figure 2.5. The distance travelled by individual leaves can also be computed by counting the number of pulses, with each pulse containing four counts. The settings of the encoder, gearbox, and drive screw then determine the number of counts per millimetre of leaf travel which allows for the position of the leaves to be computed [33]. MotComm printed circuit board Inside the MLC there resides a MotComm printed circuit board (PCB) for each carriage. This PCB deals with communications to the head transceiver PCB, motor quadrature signal decoding, servo control/pulse width modulation (PWM), and controlling the carriages [33]. Fiber optic cables are used for fast communication between the MotComm PCB and the head transceiver PCB. Each MotComm PCB has two fiber optic cables, one that transmits information from the head transceiver PCB to the MotComm PCB and 18

39 one that transmits information from the MotComm PCB to the head transceiver PCB. There exists one head transceiver PCB per MotComm PCB. Outgoing messages from the MotComm PCB include motor pulse width modulation commands which is important for leaf accuracy. Incoming messages to the MotComm PCB include motor quadrature signals which instructs the leaves on how to move. On a basic level the MotComm PCB transmits information about the status of the MLC and the head transceiver PCB interprets this information then sends commands back to the Mot- Comm PCB. Cyclic redundancy checks are also performed on these messages to double check that nothing nonsensical is received by the other party [33]. The MotComm PCB also decodes the information regarding the encoder counts for each individual motor. It calculates how far each motor has rotated and hence how far each leaf has travelled. This information is used to report the actual position of each leaf to a resolution of 100nm [33]. The machine can compare this actual position of the leaf with where the leaf was originally planned to be, from the treatment plan. If the leaves are ever out of tolerance (usually set at 2mm), the treatment head stops delivering radiation until the MLC has corrected itself. Newer models continue delivering radiation but heavily decrease the intensity of the beam until the MLC corrects itself. This combats the dosimetric errors that arise from the overshoot effect which will be discussed later in section A microprocessor is present inside the MotComm PCB to provide motor servo support via the use of low-level motor control algorithms. The actual position of each motor is compared with the command position of each motor every 4.5ms [33]. From here a PWM is assigned to each motor ranging from 100% to 100%. These are then converted into direction and enable signals and are sent to the head transceiver PCB via the fiber optic cables. This tries to minimize error by minimizing the difference between the calculated actual positions and the expected leaf positions from the treatment plans. For each motor, the MotComm PCB produces a 10kHz PWM enable frequency with 2% resolution [33]. Manufacturers Differences Due to the existence of three different MLC manufacturers, Siemens (Siemens AG, Erlangen, Germany), Elekta (Elekta AB, Stockholm, Sweden), and Varian, each MLC model is slightly different in what it can and can not 19

40 Figure 2.6: Radiation penetration: (a) Penetration dependence upon leaf position is minimised. (b) Penetration through square leaves is dependent on leaf positioning which is why Varian MLC leaves are curved. do [35]. One such property is interdigitation whereby the leaves on opposing banks are allowed to pass each other. Depending on the model the MLC may either be able to perform full interdigitation, no interdigitation, or no interdigitation with gap. This corresponds to the leaves from opposing banks possessing the ability to pass each other, the leaves from opposing banks being able to touch but not being able to pass each other, and the leaves from opposing banks not being able to pass a certain distance from the opposing bank, respectively [35]. The leaves of the Varian and Elekta MLCs have rounded edges which can be seen in figure 2.6. If the leaves possessed flat edges then by extending or retracting them from the center you are extending the penumbra by a non linear amount, so by curving the leaves we minimise the perturbation of the penumbra width as a function of leaf position [36]. Seimens produces MLCs that extend and retract along a curved track as opposed to the straight tracks present in Varian or Elekta machines. By curving the leaves such that they are normal to the radiation output the penumbra remains independent upon leaf position. The curved nature of the MLC leaves also helps to reduce the radiation transmission through the leaves as the leaves are always parallel to the radiation [33], [36]. The multileaf collimator is a complicated and integral component of the linear accelerator and as such it is important that it is working properly. Quality assurance of 20

41 Figure 2.7: The gantry rotates in the horizontal axis and the collimator rotates in the vertical axis. The point in space where the radiation beams intersect when the Gantry is rotated is called the radiation isocenter. the multileaf collimator was discussed in section Gantry Most linear accelerators that are currently manufactured allow for the rotation of the radiation source around the patient via a gantry. The accelerator waveguide and treatment head are mounted onto a gantry that can rotate around the patient on a horizontal axis [6] as seen in figure 2.7. The collimator may also rotate in the vertical axis. The point in space of the gantry and collimator rotations, or alternatively where the radiation beams intersect when the gantry is rotated is called the radiation isocentre. The rotating gantry allows for a greater degree of freedom in the treatment. By rotating the gantry head around the patient, healthy tissue is spared greater doses of radiation and the tumour can be targeted more precisely [6]. However the gantry does not rotate perfectly around the patient [37]. It is affected by outside forces such as gravity and when coupled with factors such as the collimator misalignment it leads to a perturbation of the radiation isocenter. This leads to spatial inaccuracies of deliveries [38], [39]. Due to the wobble present about the isocenter, the radiation isocenter can also be viewed as a radiation isosphere [40]. 21

42 Figure 2.8: Dose distribution for Picket Fence QA recorded by an EPID. 2.2 Quality Assurance Tests Ling s Test 1 Ling s Test 1 [41], sometimes referred to as the Picket Fence test, is a quality assurance test that is performed on the MLC of the linear accelerator. The leaves on the left bank of the MLC are extended by the same amount. Similarly the leaves on the right bank of the MLC are also all extended such that there is a small constant gap between the two opposing banks. The leaves idle in that position for a predetermined amount of time before moving on to their next position, preserving the same gap length in between the two banks. This process is repeated a total of 10 times with radiation continuously being delivered throughout this test. Additionally the gantry constantly rotates through its full range of motion however variants do exist where the test is delivered at a constant gantry. The resulting dose distribution resembles something akin to figure 2.8. The dose distribution can be analysed to determine if the leaves were out of position in relation to the rest of their bank or not. If the pickets that arise in the dose distribution are not uniform then at least one of the leaves are misaligned. 22

43 2.2.2 Ling s Test 2 Ling s Test 2 tests the machine s performance in relation to the dose rate of the machine and the gantry speed. While the dose is being delivered, the two banks repeat the same motion as Ling s Test 1 only this time only 7 peaks are created instead of 10. The gantry rotates at varying speeds between the peaks and the dose rate also varies. The aim of Ling s Test 2 is to see if the variable dose rate effectively cancels out the varying gantry speed. There should be an even distribution of MUs to the different parts of the field even though the gantry speed and dose rates vary Ling s Test 3 Ling s Test 3 tests the machine s performance in relation to the dose rate of the machine and the speed of the MLC. The field is broken into four different sections of varying MLC speeds, each with a constant but different dose rate. This assesses the accuracy of the leaf velocity as each segment should have a unique uniform dose distribution. If they do not then the leaf velocity is incorrect Sliding Gap Quality Assurance Another prevalent test for performing quality assurance on the MLC is known as sliding gap (SG). In sliding gap QA the leaves are swept across the field at a constant velocity with an approximately 5mm gap between the two banks while radiation is being constantly delivered. Much like Ling s Test 1, this test assesses the positional accuracy of each leaf with respect to it s bank. If the resulting field is not uniform then at least one of the leaves is out of position. Variations exist that include the leaves travelling across the field in one direction then back the other way, as well as tests that have the gantry rotating at a constant velocity during beam delivery, or different velocities during delivery Gamma Map Analysis Gamma map comparison, first introduced by Low et al. in 1998 [24], is a method that analyses the agreement between two different distributions. It is widely used to judge 23

44 the agreement between the expected dose of the treatment plan and the actual dose measurement. The basic premise is to determine if the dose delivered to a particular point is similar (within a small margin) to the corresponding point (also within a small margin) on the treatment plan. Gamma Map Analysis relies on two factors, the Distance to Agreement (DTA) and the Dose Tolerance ( D). A point on distribution A, r i is compared with a circle of radius DTA that is drawn around a corresponding point, j i on distribution B. The circle consists of many positions, j m. A gamma value, γ m is calculated for all of these points through the following equation. γ m = [Dist/DTA] 2 + [DoseDiff/ D] 2 Where Dist is the distance between j m and j i and DoseDiff is the percentage difference between the dose at j m and the dose at r i. The smaller Dist and DoseDiff is, the lower γ m will be and hence the greater the similarities between the two distributions for those points. r i. The minimum value for all of γ m is then taken to be γ i, the gamma value for position Typical values for DTA and D are 3mm and 3% respectively. If Dist >DTA or DoseDiff > D then γ m > 1. If it is also calculated that γ i > 1 then it is said that the position r i failed the gamma map comparison. The percentage of passes for the whole distribution can be calculated and it is this number that is often used for QA. Whilst there are no set rules in place that tell centers whether or not a particular planned distribution is similar enough to the actual distribution to continue with treatment the general consensus is that anything less than an 85% pass rate at a 3%/3mm level, should be revised before treatment. Although the use of Gamma Map Analysis is widespread it is not a perfect system. The system no longer provides useful information pertaining to the clinical significance or source of disagreements if there is ever a gamma value 1 [42]. There has been work done with vector fields that hopes to solve this problem. 24

45 2.3 Literature Review The advent of the multileaf collimator has allowed for more complicated and precise treatments that can more easily create complex dose distributions that would otherwise be impossible to achieve. Like any other piece of equipment integral to delivering radiotherapy treatments the multileaf collimator must be calibrated and undergo routine quality assurance checks to make sure it is performing within tolerance levels. Task Group 142 from the American Association of Physicists in Medicine s Science Council (AAPM) suggest qualitative tests, more specifically, picket fence tests to be undertaken on a weekly basis. AAPM also suggest the travel speed of the leaves to be assessed on a monthly basis. They recommend either using vendor software or MLC log file evaluation to do so noting that this is only available for use on the Varian systems and that similar analysis software for different machines can be created if leaf and time dependent data can be extracted [23]. Elekta machines do have their own log file system in place however reviewing this information has many practical limitations. The information can only be viewed on the service page of the control system, the MLC leaf and jaw positions are both stored as raw potentiometer readings, and the values recorded for each control point are saved in their own individual binary file. Recent headway has been made in analysing Elekta log files with Arumugam et al. creating a program that analyses the error of treatment plan parameters as well as comparing the DVH to the original plan [43]. Although it has not been verified by third party detectors, the program still shows promise Varian Machine Models As a manufacturer, Varian is constantly revising and upgrading their linear accelerators to more advanced models. Three such models that this project will be focusing on are Varian - Clinac ix, Varian - Trilogy, and Varian - Truebeam. Varian - Clinac ix and Trilogy machines are more similar than different. Trilogy machines are more advanced in that they have access to a higher dose rate (1000MUs/min) 6MV photon delivery mode as well as Gated Rapid Arc which monitors and adjusts for tumour motion during delivery. However the treatment heads are both designed to the same specifications which allows for similar treatment deliveries. Additionally, 25

46 both machines have the capacity to produce DynaLog files after every treatment. By comparison, Truebeam machines are completely different to the Clinac ix and Trilogy machines as they possess a different treatment head with related components. In addition to the mechanical differences between the treatment machines, there are two main differences residing in the software between Truebeam and previous Varian models. First, the log files that Truebeam machines produce are slightly different than Varian produced DynaLog files. They provide additional information and the formatting of the log files and sampling frequency differs between the two. The main difference is the increased sampling rate from the 50ms present in DynaLog files to the 20ms of Truebeam trajectory log files. Truebeam machines also add a previously absent component called the Supervisor Module. The Supervisor Module connects the multiple subsystems which allow the logical modules (beam generation, collimation, couch, Imaging system, detector arms, Stand) to provide detailed control and monitoring of subordinate functions [44]. The Supervisor polls TrueBeam subsystems for information and adjusts functions, including the collimator and MLC, every 10ms. This allows for a more intensive supervision of the leaf positions than in the previous models which do not possess the Supervisor module Verification of Varian log files Verification of Varian log files from Varian systems date back to the early 21st century. In 2003 Li et al. verified DynaLog files against a commercial two-dimensional diode array by way of measuring the MU delivery. Their research found excellent agreement for their measurements and they go on to suggest log-file validation to become standard procedure when commissioning a linac for IMRT [45]. The following year Zeidan et al. compared the accuracy of DynaLog files against an electronic portal-imaging device (EPID) and found their values agreed within 5%, providing further evidence that the log files fluence and leaf positioning data are accurate representations of actual delivery [46]. 26

47 2.3.3 Varian log files for Quality Assurance Numerous studies have been undertaken regarding Varian log file analysis as part of a Quality Assurance program. In 2002 Litzenberg et al. created a program for a semiautomated quality assurance system. Using DynaLog files it calculated position versus time related quantities (and their derivatives), dose scaled quantities versus position, differences between the expected and actual values for the previous two quantities as well as statistical data. On top of this Litzenberg et al. also included utilities to compare the reconstructed dose distribution with third party detectors. They stress it is still important to keep routine MLC QA as it is required to ensure that the MLC is working correctly and the data presented in the DynaLog files are accurate [47]. Litzenberg et al. followed this up in 2006 with a paper detailing a program that utilised the DynaLog files by utilising their data as flags for a Monte Carlo simulation. A photon would only fire if it read from the DynaLog files that the beam state was on. The leaf positions were also taken into account and a fluence map was created which reproduced the film data generally well [48]. However the fluence map does not give a quantitative measurement of the delivered dose. This is due to the Varian log files normalising the recorded MU values to 25,000. As such any fluence measurements derived solely from the Varian log files are qualitative in nature. A few years later in 2009, Tae-Suk et al. converted data received from the DynaLog files into dmlc format which enabled the delivered plan to be recalculated by the original Treatment Planning Software (TPS). The actual positions of the leaves recorded by the DynaLog files were entered into the TPS and the delivery was recalculated. This allowed quantitative analysis of the treatment as the TPS calculates the total radiation delivered for the plan as well as the dose volume histograms (DVH). The DVH for the reconstructed plan as well as the original plan were compared and showed that the regions requiring steep dose gradients needed complex modulated segments. This method is known as inverse dose verification [49]. Schreibmann et al. added to this by comparing their inverse dose verification with MatriXX, an ion chamber device and found good agreement with the detector [50]. In 2010 Qian et al. expanded inverse dose verification into Volumetric Modulated Arc Thereapy (VMAT) treatments. They utilised pre-treatment cone-beam computed 27

48 tomography (CBCT) and the post-treatment DynaLog files to create a valuable clinical tool to assure the VMAT dose delivery [51]. In 2012 Sun et al. compared the efficiency and effectiveness of independent dose calculation followed by machine log file analysis against conventional measurement based methods in detecting errors in IMRT delivery. In their clinic the independent dose calculation and log file analysis on average took 47 ± 6 minutes to complete whereas the experimental approach using an ion chambers and 2D diode array measurements took an average of about 2 hours to complete. Fluence maps were created by exporting the beam parameters into a DICOM file for reconstruction in the TPS as well as via a program that analysed the beam on/off flags and leaf positions. They found their DynaLog file QA program can verify the calculation inaccuracies as well as verify the data transfer and also evaluate the performance of the MLC delivery. They concluded that their QA process using DynaLog files is a reliable tool to verify the IMRT treatment [52]. In addition to the software produced for statistical analysis, Varian Medical Systems have produced their own software entitled DynaLog File Viewer for clinical use. Version 7.0 was released in 2007 and is the subject of the following few paragraphs. It analyses DynaLog files and calculates the error of the leaves at each time step via the following formula: X A = X A exp X A act Where X A exp is the expected position of the leaf on bank A, X A act is the actual position of the leaf on bank A, and X A is the error of the leaf. The error of the leaves on bank B is also calculated in a similar manner. The software then uses this data to create an error histogram as well as calculate Error RMS values for each leaf that moved during treatment on their respective banks. Finally DynaLog File Viewer also uses the DynaLog files to create Beam On and Beam Hold Off plots [53]. As stated previously, DynaLog File Viewer is limited in its functionality. It assesses the Error RMS for each leaf which gives the user an insight as to which leaf needs replacement but it does not give individual data of each leaf. The user receives no 28

49 information with regards to when the leaf error spikes but rather only that it does which can be frustrating if you are trying to figure out why something is occurring. More in depth, independent software exists that can be used for DynaLog analysis. One such software is MobiusFX produced by Mobius Medical Systems. Their website boasts their software uses measurements of the Varian log files to calculate and verify the delivered 3D dose in the patient as well as verifying DVH objectives and analysing MLC performance [54]. Their website also claims their software saves 30 minutes per patient for IMRT QA. A similar commercial software available is Argus QA. Again it analyses the leaves and shows where in their travel they failed. It also compares the actual and expected position of the leaves as well as the actual and expected velocities of the individual leaves, plotting the results in graphs for easy viewing. Gamma map analysis is also performed, although on the fluence and not the dose[55]. Much like Mobius QA, this software has been introduced to streamline the IMRT QA process by reducing the time needed to run these important tests thus alleviating pressure from physicists. Additionally many centres develop in house programs that analyse DynaLog files for QA purposes. These centres include; Washington University [56], Memorial Sloan Kettering [57], and Northern Ireland Cancer Centre [58] Limitations of Varian log files. Varian log files have been used for analysis on numerous occasions [58] [66]. Other quality assurance programs have also been created that are now in clinical use [56], [67], [68]. Rangaraj et al. noted that machine log file analysis is a robust, efficient, and reliable QA process capable of detecting errors originating from human mistakes, flawed planning, and data transfer problems [56]. However although numerous analysis has been undertaken via the use of Varian log files there are disadvantages associated with the format. Problems can arise due to the dependent nature of the Varian log files. In studies by Agnew et al [66] Varian log files recorded the position of the leaves incorrectly due to how the data is recorded. Each leaf of a MLC is extended and retracted by an individual motor. Varian log files measure how many rotations the motor undergoes 29

50 and from there works out where the leaf should be extended to [33]. However if the t-nut connecting the motor to the leaf is loose then even though the motor is rotating the leaf would not move. As such the log file would record that the leaf has extended a set distance even though it has not moved. Calvo-Ortego et al. raised concerns over DynaLog files pointing out that DynaLog files do not reflect leaf positional errors related to a miscalibration (offset) and that they are not sensitive to errors pertaining to the absolute calibration of the linac [64]. Other limitations include the resolution of the DynaLog file which records data every 50ms [33], [53]. Later Varian models (Truebeam) improve upon this by recording information into their Truebeam trajectory log files every 20ms [52], [66], [69]. Even though Varian log files still hold a wealth of information and can be incredibly helpful when analysed, it should be stressed that due to its dependency upon the treatment machine it is recommended an independent measuring tool is used to verify the accuracy of the Varian log files [56], [61], [64], [66], [68] Effect of Gravity upon deliveries With the dawn of IMRT and VMAT treatments, concerns facing the dependence of error upon the gantry angle were raised. There has been ample research into assessing the effect of gravity on the linear accelerator s: gantry head [38], [63], [70], [71], MLC [70], [72] [75], or external imagers[71], [76]. In 2005, Wijesooriya et al. used EPIDs to discover that when the MLC is positioned such that the direction of motion of the leaves are parallel to the force of gravity the outer leaves fighting against gravity move 6% slower and the leaves moving with gravity move 2.3% faster than their perpendicular counterparts [74]. However in 2010 Buckey et al. analysed DynaLog files and observed that Gravity effects were not noted when reviewing the dynalog files, agreeing with the measured MatriXX and film results. [77]. Sharma et al. also researched into the gravity effect on the MLC (in this case, a high definition MLC or HDMLC) and found the effect of gravity on HDMLC alignment was not observed, as the alignment accuracy was unchanged at all gantry rotations [78]. This is in stark contrast to the numerous papers who all accept the effect of gravity on the MLC as fact, attributing errors and discrepancies to this phenomenon as well 30

51 as simulating and warning of it [60], [74], [79] [83]. Lee et al. used a MatriXX ion chamber to measure the dose distribution at different gantry angles. Their research conclusively reveals that the DMLC gravity definitely affects IMRT dose distribution [84]. Furthermore the manual for the Varian MLC even suggests preforming tests at different gantry angles so that the leaves have to fight gravity [33]. Whilst there has been work in using Varian log files to compare the effect of gravity on the velocity of individual leaves [74] and indeed in the final fluence map there has been no research, to the author s knowledge, undertaken comparing the positional error of individual leaves to the gantry angle Overshoot In 2001 Ezzell et al. discovered that the control system required 65ms to monitor and halt the irradiation of a segment, for Varian DMLC system (V4.8). Basically, the control system records that the correct MU has been delivered for the segment. Ideally the irradiation would halt immediately however in actuality the beam halts some 65ms later which can end lead to the segment ending on a higher than expected monitor unit [85]. Conversely the 65ms delay has the consequence that the radiation will start delivering after the segment has begun leading to lower MU. This is known as the overshoot phenomenon and is prevalent in step and shoot IMRT deliveries due to the latency present in deliveries [48], [62], [69], [80], [85] [87]. As a side note due to this delay, the leaves lag behind their intended positions which are defined by the fractional MU or index in the leaf sequence files. Although it may seem like the fault lies in the leaf speeds not being able to keep up with the treatment plans, the error arises from the communication delay [88]. For static MLC delivery the first segment receives more dose and the last segment receives less. The dose delivered to the middle segments tend to balance out as the segment starts later but also ends later [85], [88]. Ezzel et al. concluded that the errors were insignificant in high dose regions and clinical treatments at 400 or 600 MU/min should not be compromised by this overshoot effect [85]. In 2004 Stell et al. analysed DynaLog files and found that the MU redistribution errors arising from the Overshoot phenomenon were shown to be clinically insignificant [87]. 31

52 However problems arise in dynamic MLC delivery due to the communication delay. If the leaves breach their tolerance then the beam is halted until the leaves correct themselves. The beam is then resumed. These are all subjected to the 65ms delay. If the leaves continue to breach tolerance then the dose distribution is rippled even though a smooth distribution was planned [88]. This can be corrected for by implementing leaf sequence algorithms into the treatment planning software [88]. In 2011 Agnew et al. performed analysis on DynaLog files and the overshoot effect was factored into the analysis. They reached the conclusion that the errors in the Stepand-Shoot files were dominated by dose errors originating from sub optimal segment distribution which compounded the overshoot effect [62]. In 2014 Agnew et al. found that the overshoot effect was eliminated in clinical treatments delivered by the new Varian TrueBeam model [69]. The study also found that the leaf errors were reduced in the Truebeam model when compared to the Varian C-series 2100CD model. This is due to the presence of the new Supervisor module that detects and corrects the MLC every 10ms which is not present in previous Varian models [69] Varian log files for analysis. Varian log files have also been used to assess the limitations that are placed on the TPS to ensure better treatment plans. In 2002 Litzenberg et al. analysed DynaLog files to place a limiting velocity on the MLC. They found the main cause for delay to be the time delay between the MLC control and leaves. The faster the leaves move, the further they can move out of tolerance before being checked on. If this delay is not taken into account DMLC delivery sequences are often forced to have larger-thandesired tolerances, and also experience numerous beam hold-offs [86]. Using DynaLog files Litzenberg was able to modify the motion equations to apply to the TPS. Rangel et al. followed this up in 2009 by using DynaLog files to analyse the positional errors of the MLC. They found random errors constrained by a tolerance of the MLC controller to 2mm result in negligible changes in the dosimetry of all structures of interest and hoped their study provides guidance for the development of performance standards for multileaf collimators [83]. 32

53 The following year Rangaraj et al. analysed Varian log files from VMAT treatments to improve VMAT treatment planning. They concluded that optimal VMAT delivery solution is characterized by shorter treatment time when machine constraints include the limitations on maximal gantry rotation speed, maximal admissible beam dose rate, and maximal admissible leaf speeds where these parameters were indirectly determined from the delivery log files. Through the use of Varian log files Agnew et al. determined that variability in MLC speed affected MLC positional accuracy [58]. They postulate that a reduction in the variability of MLC speed could be used to further optimise the VMAT planning process [58]. Varian log files have been in use for the last 15 years for a vast array of applications including QA programs and statistical analysis. This thesis aims to analyse and compare multiple different treatment methods and machines through the use of Varian DynaLog and Truebeam trajectory log files. Additionally this thesis aims to construct an accessible program that analyses Varian log files for clinical use that is freely available. 33

54 Table 3.1: Structure of the header in a DynaLog file. Modified from DynaLog File Viewer Reference Guide Line Format Description 1 B Letter Indicating Version 2 <Lastname>,<Firstname>,<ID> Patient information, up to 25 characters 3 <Field Serial Number> if VARViS 6.2 or or <PlanUID>,<BeamNumber> Treat 6.5 or greater or or <Plan Filename> Standalone 4 <long> Tolerance 5 <long> Number of Leaves in MLC 6 <long> Clinac Scale 0 = Varian Scale 1 = IEC Scale CHAPTER 3 MATERIALS AND METHOD 3.1 Varian Log Files Dynamic Log Files For each treatment delivered by a Varian Trilogy or Varian Clinac - ix model, two DynaLog files are created, one for bank A of the MLC and one for bank B. They are both identical in structure and each file is split into two main components. The header and the content. The header takes up the first six lines of the file and is structured as outlined in table 3.1 whereas the content size differs depending on the length of the treatment as information is constantly being appended on millisecond intervals. Information is consistently logged every 50ms. This data is located underneath the header in the content of the DynaLog file. Each line represents a new entry and lines 34

55 Table 3.2: Structure of the contents in a DynaLog file. Modified from DynaLog File Viewer Reference Guide. Column Description 1 Current Dose Fraction or Gantry Angle 2 Previous segment number (starting with zero) 3 Beam hold-off state 2 = LFIMRT carriage group transitions 1 = MLC beam hold-off signal asserted 0 = MLC beam hold-off signal not asserted 4 Beam on state 1 = Clinac beam is on 2 = beam is off 5 Previous segment dose index or previous segment gantry angle 6 Next segment dose index or next segment gantry angle 7 Gantry rotation in 10th of a degree 8 Collimator rotation in 10th of a degree 9 Upper Y1 jaw position in mm in the isoplane 10 Upper Y2 jaw position in mm in the isoplane 11 Lower X1 jaw position in mm in the isoplane 12 Lower X2 jaw position in mm in the isoplane 13 Carriage expected position in 100th of a mm 14 Carriage actual position in 100th of a mm The remaining columns contain the following values for each leaf in the carriage. (nleaf = 0,1,2 and so on) 4*nLeaf + 15 Expected position in 100th of a mm 4*nLeaf + 16 Actual position in 100th of a mm 4*nLeaf + 17 Previous field position 4*nLeaf + 18 Next field position are continually appended until the treatment is stopped. The structure of these lines are outlined in table 3.2. It should be noted that the dose fraction is normalised to a value of The dose fraction represents the fraction of the total dose that has been delivered thus far, that is a dose fraction of 0 tells us that 0% of the dose has been delivered and a dose fraction of tells us that 50% of the total dose has been delivered. Because of this oddity it is impossible to know the quantity of radiation that is delivered from the DynaLog files alone however this can be easily corrected by extracting the total dose of radiation that was delivered from the treatment planning files. The beam on state is dependent upon the beam hold-off State. For the beam to switch off, the MLC beam hold-off signal is first asserted, denoted by the beam hold-off 35

56 Figure 3.1: Different Position Readout Scale Conventions. Modified from High Energy C-Series Clinac, Installation Product Acceptance Booklet. State changing to value 1. At the next time step the beam is switched off, denoted by the beam on state changing to value 0. When the beam is switched on the MLC beam hold-off signal is first released, denoted by the beam hold-off State changing to value 0. At the next time step the beam in then switched on (beam on state = 1). Varian DynaLog files record the gantry rotation and collimator rotation in the Varian Scale. This convention is shown in figure 3.1. It is important to note that in the Varian Scale at 180, the gantry head is directly above the patient. However in many clinical settings the treatment machines follow conventions similar to Varian IEC or the Varian IEC 1217 scale. These conventions list 0 as the angle at which the gantry head is directly above the patient while 180 is directly underneath the patient. The differing conventions are also evident in the measurement of the collimator rotation. For the purposes of this thesis the IEC convention will be used. It is important to note the Jaw Positions are all recorded in the isoplane of the machine. The carriage and leaf positions are all recorded in the leaf (physical) plane. 0 denotes the centreline, positive values represent the leaf or carriage being extended across the centreline and negative values represent the leaf or carriage being retracted from the centreline. This can be seen in figure 3.2 along with the location of the upper and lower jaws. The example shows that both leaf number 54 of Bank B and leaf number 8 of Bank A will have positive values as they are both extended over the centreline. The remaining leaves of Bank A and B will have negative values as they are retracted from 36

57 Figure 3.2: Schematic of a typical 120 leaf MLC showing the relative positions of the leaf banks and upper and lower jaws. Leaf 54 of Bank B and leaf 8 of Bank A are both extended across the centreline. the centreline. For further information regarding how the leaf positions are physically recorded refer back to section Truebeam trajectory log files Varian Truebeam trajectory log files are formatted differently to traditional DynaLog files that are created by Trilogy or Clinac-iX machines. For the purposes of this thesis I will be focusing on the trajectory log files created by Truebeam V1.5 and higher (ie: trajectory log files V2.1). Unlike DynaLog files, Truebeam trajectory log files do not differentiate between bank A and bank B of the MLC, rather they include all the data in one file. This file is subdivided into four main components, the header, subbeams, Axis data, and CRC. The header is structured as outlined in table 3.3 A treatment is split into multiple subbeams. Subbeams are created when a series of treatment fields are made automatically [89]. The information pertaining to subbeams is located directly under the header and has the following format located in table 3.4. Each subsequent subbeam is appended to the bottom of the trajectory log file which means that n subbeams will result in the subbeam section being n lines long. 37

58 Table 3.3: Structure of the header in a Truebeam trajectory log file. Modified from Truebeam Trajectory Log File Specification [89]. Line Data Description 1 Signature VOSTL 2 Version Header Size (fixed for now at 1024) 4 Sampling Interval in milliseconds 5 Number of axes sampled. (Indicates length of next entry) 6 Axis enumeration Coll Rtn - 0 Gantry Rtn - 1 Y1-2 Y2-3 X1-4 X2-5 Couch Vrt - 6 Couch Lng - 7 Couch Rtn - 9 MU - 40 Beam Hold - 41 Control Point - 42 MLC Samples per axis. For MLC, it is the number of leaves and carriages 8 Axis Scale 1 - Machine Scale 2 - Modified IEC Number of subbeams 10 Is Truncated? (Did the plan exceed 20 minutes?) 11 Number of snapshots 12 Reserved Table 3.4: Structure of the subbeam in a Truebeam trajectory log file. Modified from Truebeam Trajectory Log File Specification [89]. Line Data Description 1 Control Point. Internally-defined marker that defines where the plan is currently executing. 2 Dose delivered in units of MU 3 radtime. In units of seconds it is the expected irradiation time of the subbeam. When actual irradiation time exceeds the expected irradiation time the plan is terminated. 4 Sequence number of the subbeam 5 Name of the subbeam 6 Reserved 38

59 The Axis data is stored immediately after the subbeam data. It is presented as a series of snapshots, that is data is recorded at regular intervals throughout the treatment. It has the general format; Values[Axis1], Values[axis2],..., Values [AxisN] With each value having two fields, the expected and actual values. All data is recorded in Varian scale with units of cm for linear axes, degrees for rotational axes, and MU for dose. In addition to the format changes, Truebeam trajectory log files also record information in 20ms intervals instead of 50ms intervals. Additionally when the beam is paused, be it from a minor fault arising in the treatment or whether it was instigated by the user, the system stops recording information to the trajectory log files. When the beam is resumed the trajectory log files start recording again. 3.2 Multileaf Collimator model All machines that were a part of this study were equipped with a M120 MLC. A M120 MLC consists of two separate banks of 60 tungsten leaves, each bank sitting on a carriage. The tungsten leaves are independently powered by individual DC motors allowing independent movement between the leaves. The leaves are split into two categories; leaf numbers 1-10, and are labelled outer leaves whereas leaf numbers are labelled inner leaves. Most of the outer leaves are 1.0mm in width and have approximate mass of 1.0kg each whereas inner leaves are 0.5mm in width and have approximate mass of 0.5kg. Leaf pairs 1 and 60 both have width of 1.4mm. Each carriage weighs 36kg [90]. More information pertaining to the M120 MLC and MLCs in general is located in section Matlab Software Multiple MATLAB programs were created and used in the analysis of Varian log files. A summary of their function is outlined below. The MATLAB programs were amalgamated into a program entitled Timberwolf which was used for statistical analysis. 39

60 Figure 3.3: A basic workflow schematic of the importation of Varian log files. A graphical user interface (GUI) was also formed using the following programs as a base for clinical use. This GUI was entitled MountainGoat Importing log files In order to analyse the Varian log files the data is first converted into a more user friendly format. A basic schematic (figure 3.3) documents this process. It should be noted that Varian log files record the Gantry Angle in units 10th of a degree using the Varian scale. Due to the widespread use of IEC and IEC 1217 scale, all angles are converted from Varian scale to IEC scale. That is, gantry angle 0 becomes gantry angle 180 and vice versa whilst gantry angle 90 and 270 remain where they are. The expected and actual positions of each leaf recorded in the log files are in units of 100th of a millimetre and are in the plane of the MLC. Whilst this information might have some merit, for the purposes of this project we are interested in the expected and actual positions of each leaf in the plane of the isocenter, or rather where the patient is located. This will give a more realistic look into the effects of the leaf error as the radiation is incident at the isocenter. To convert the leaf positions to the isocenter a scaling factor must be applied which for Varian M120 MLC is [33]. All of this information is then collated into two different structures depending on whether the information originated from bank A or bank B of the MLC. These structures are named dynaloga and dynalogb respectively. A list of the information that is 40

61 stored in the structures for each time step is as follows: Cumulative MU delivered Previous Segment Number Beam Hold Off State Beam On State Gantry Rotation (IEC convention) Collimator Rotation (IEC convention) Locations of each Jaw in the isoplane with respect to the centerline Locations of each leaf position in the isoplane with respect to the centerline (both actual and expected positions) Carriage positions (actual and expected) Previous and Next field positions Previous and Next segments Computing Positional Errors The two structures, dynaloga and dynalogb are read into a separate function that calculates the positional errors of each leaf. The errors of each leaf at each time step were calculated as per equation 3.1 for bank A and bank B. PositionError(n, t) = ExpectedPosition(n, t) ActualPosition(n, t) (3.1) Where n is the leaf number and t is the timestep of the treatment. The leaf motion was analysed so that only the leaves that moved in the duration of the treatment were recorded into the positional error array. Additionally all errors were ignored when the radiation was not being delivered due to how the treatment is planned. 41

62 20 Expected and Actual Position of a Leaf vs Time 15 Position (mm) BeamOff BeamOn 5 Actual Position Expected Position Beam Status Time Step Figure 3.4: The Expected and Actual positions from Varian log files of a leaf during a step and shoot treatment. When the beam is off the expected position jumps to the next segment whereas the actual position slowly changes. For Step and Shoot treatments the treatment is split into multiple segments which are separated by sections of no radiation delivery. After a segment has finished delivery the radiation beam halts and the leaves move to the next position. However the Varian log files record the expected position of the leaves such that they are already at the next segment. As soon as the beam stops delivering the expected positions of the leaves become the values of the next segment while the leaves actually have to physically move from the last segment to the next segment. This is shown graphically in figure 3.4. As such the errors computed while the beam is off are nonsensical. Additionally as they are calculated when no radiation is being delivered the errors have no bearing on the treatment and can be ignored. If the treatment rapidly toggles between beam on and beam off, for instance due to the overshoot effect, then there is a possibility that the positional error of the leaves will be higher than normal. This is because the location of the leaves have not had time to get to their new location Calculating Leaf Velocity The two structures, dynaloga and dynalogb were read into a function to compute the velocity of each leaf at each time step. A five point stencil method was used to compute 42

63 the resultant actual and expected velocity. The five point stencil method is derived from expanding out the Taylor series of a function f of a real variable at a point x to the third order. It takes the general form: f (x) f(x + 2h) + 8f(x + h) 8f(x h) + f(x 2h) 12h Where h is the time between subsequent timesteps. By applying this to the Varian log files we receive equation 3.2 and 3.3 for the actual and expected velocity respectively. AP(t + 2, n) + 8 AP(t + 1, n) 8 AP(t 1, n) + AP(t 2, n) ActualVelocity(n, t) = t 12 (3.2) EP(t + 2, n) + 8 EP(t + 1, n) 8 EP(t 1, n) + EP(t 2, n) ExpectedVelocity(n, t) = t 12 (3.3) Where AP is the actual position of each leaf, EP is the expected position of each leaf, n is the leaf number, t is the timestep, and t is the difference in time between subsequent timesteps. For Clinac ix and Trilogy machines, t=0.05s, and for Truebeam machines t=0.02s. Care was also taken at the boundaries of the arrays to exclude velocity values that pulled non existent positional values from outside of their arrays. Those that did were assigned the value NaN Computing Velocity Errors The structures with information pertaining to the velocities of bank A and bank B are read into a function that calculates the velocity error of each leaf. The velocity error is calculated as per equation 3.4. VelocityError(n, t) = ExpectedVelocity(n, t) ActualVelocity(n, t) (3.4) Where n is the leaf number and t is the timestep of the treatment. Much like the error of leaf position, the error of leaf velocity is only computed for leaves that are used 43

64 Figure 3.5: A basic workflow schematic of how the Fluence Maps are created during the treatment, that is it is only computed for leaves that move. Additionally the velocity error is only calculated for timesteps for which the beam is delivering radiation and their two neighbouring timesteps are also delivering radiation. This is due to how the velocity is calculated. As the five point stencil method is applied, each of the time steps used must be delivering radiation. If not all of the time steps are delivering radiation then the position and hence resulting computed velocity will be invalid. This is why the velocity error is only computed for timesteps, t, which satisfy the condition that timesteps; t-2, t-1, t, t+1, and t+2 are all delivering radiation Computation of Fluence Maps The two structures, dynaloga and dynalogb are read into a function to compute the expected and actual fluence of the delivery. A basic schematic of this function can be seen in figure 3.5. An empty map of the treatment area is created from the maximum and minimum positional values of the leaves. The expected positional values of the leaves were used to determine which areas on the map were exposed to the radiation beam, timestep by timestep. These areas are then incremented by the MU that was delivered at that time step. This process is then repeated for all timesteps, until the treatment is complete, to create the expected fluence map. The actual leaf positions are also used in the same manner to calculate the actual fluence map. The actual and expected fluence maps are built from the same empty map which means that the two maps cover the same area. The data points of the actual fluence map are spatially identical to the data points of the expected fluence map. This is important when comparing the two maps to each other. 44

65 As Varian log files normalise the MU to there is no possible way to calculate the quantitative dose map without further information. If you were privy to the total dose delivered by the treatment then you could easily calculate the dose distribution instead of the fluence map by scaling appropriately. As an aside, this program also has the option to capture the fluence distribution, the current leaf positions, and jaw positions at each timestep. Using this, a fluence movie can then be built which showcases the evolution of the fluence map as well as the position of the jaws and MLC with respect to time. This process can take upwards of 10 minutes to render on a system with 4.00GB RAM and 2.50GHz processor, and does not add anything of value to the analysis, however the resulting video has pleasing aesthetics. These fluence maps are primarily used for gamma map comparison. As gamma map comparison is concerned with the dose percentage difference, a normalised dose will not affect the analysis and identical values would be obtained Gamma Map Analysis The gamma map analysis software was modified from the one used by Rowshanfarzad et al[91]. It was modified to include the distance between each adjacent position in the array by taking into account the Varian log files resolution. The expected and actual fluence maps are read into the program for analysis. Variables are assigned from prompts to determine the DTA and D. A field threshold value can also be entered that determines if particular data points on the fluence maps are computed in this analysis. The dose value at these points has to be above the field threshold to be analysed. For each point of the actual fluence map a square of data points centred around the corresponding data point was scrutinised in the expected fluence map. Each data point in this square was analysed to determine the gamma map value, γ m, as per the formula: γ m = [Dist/DTA] 2 + [DoseDiff/ D] 2 Where Dist is the distance from the data point to the centre of the square and 45

66 Figure 3.6: Basic structure of how the program works. 46

67 DoseDiff is the percentage difference between the two fluence values. The minimum γ m from the square is then chosen as the gamma value for that point between the two fluence maps. This is repeated for all of the points present. This is shown graphically in figure 3.6. If the data point lies on an edge then the program would simply compare it to the data points available. If the point has a gamma value of <1 then it is said to have passed. Conversely if the gamma value is >1 then that point is said to have failed. The percentage of points that passed the analysis is computed and recorded as the gamma map pass rate. 3.4 Linear Accelerators A total of 10 different linear accelerators from multiple different treatment centres were analysed as a part of this project. Their details are summarised in table 3.5. Table 3.5: Summary of linear accelerators used in this study. Code Name Model Year Installed ix 1 Varian - Clinac ix 2007 ix 2 Varian - Clinac ix 2007 ix 3 Varian - Clinac ix 2008 ix 4 Varian - Clinac ix 2008 ix 5 Varian - Clinac ix 2010 Tri 1 Varian - Trilogy 2009 Tri 2 Varian - Trilogy 2012 True 1 Varian - Truebeam v True 2 Varian - Truebeam v True 3 Varian - Truebeam v DynaLog files were extracted from the Clinac ix and Trilogy machines for a wide range of treatment and quality assurance plans whilst trajectory log files were extracted from the Truebeam machines. The trajectory log files were modified post treatment by a separate party such that the content component of their files resembled the structure present in DynaLog files. 47

68 3.5 Leaf Error Multiple Varian log files of different deliveries were retrospectively analysed using the software developed in this study. The software calculated the positional error of the leaves as well as the RMS error of each leaf. Each Varian log file has data for 120 leaves recorded for approximately 1,500 time step entries Root Mean Squared Deviation The root mean squared deviation (RMSD), also known as the root mean squared error (RMSE), combines the magnitude of the errors into a single number through the following equation. RMSD = Σ N n=1 (xexp,n x act,n ) 2 N (3.5) Where N is the sample size, xexp is the expected value and x act is the actual value. The RMSD is simply the sum of the errors of the treatment squared, divided by the sample size and then square rooted. This provides a good measure of the accuracy of the treatment Picket Fence (Ling s Test 1) To compare the positional error of different treatment machine models the positional error was assessed for the same delivery, Ling s Test 1 on three different treatment models, Clinac ix, Trilogy, and Truebeam. Each treatment was analysed to calculate the positional error present for each leaf at each time step. The total RMS for the treatment was then computed from all the leaf errors present on both banks. Whilst we lose information regarding which bank and which leaf was prone to error this method gives us a quick metric to measure how accurately the treatment was delivered Step and Shoot and dimrt The positional errors of multiple Step and Shoot treatments focusing on the prostate region was calculated for each leaf at each timestep. The positional errors were also 48

69 calculated for dimrt treatments that focused on the prostate region. All treatments were delivered by Clinac ix machines VMAT To assess the dependence of error upon different treatment sites the RMSD was calculated for VMAT treatments delivered to different treatment sites. All deliveries were performed by Truebeam machines. These were split into three different treatment sites; the prostate (PROS), head and neck (H&N), and the prostate pelvic node (PPN). The RMSD of head and neck treatments delivered by Clinac ix machines and Truebeam machines were also compared and contrasted to one another Comparing Treatment Methods - Clinac ix To compare the error between different treatment methods the RMSD was calculated for Step and Shoot, dimrt, and VMAT treatments for Clinac ix machines. Ideally all the treatments would be focused on the same site to exclude additional factors however the only Step and Shoot and dimrt treatments available targeted the prostate region and the only VMAT treatments available targeted the head and neck region Positional Errors due to Velocity To assess the effect of the leaf speeds on the positional error, sliding gap quality assurance was analysed. Two tests were delivered with leaf speeds of 10mm/s, 20mm/s a total of four times each. One test was delivered with leaf speed 30mm/s a total of three times. These were delivered with a static gantry at gantry angle 180. The data was also corrected to remove the dependence upon the leaf error due to the 55ms MLC system delay through the following equation: X corrected = X raw ActualVelocity Where X corrected is the corrected error, X raw is the uncorrected error and ActualVelocity is the velocity of the leaf at that time step. The leaves undergo the uncorrected error. The corrected error only seeks to take away the error due to the system delay. 49

70 3.6 Inner/Outer Leaves To assess whether there is a relationship between position error and whether the leaf belongs to the inner and outer leaf category, multiple Varian log files of picket fence tests were analysed. The RMSD of the inner and outer leaves for each test was calculated for a total of 7 different machines. 3.7 Velocity Leaf Error Multiple Varian log files of different deliveries were retrospectively analysed using the software developed in this study. The software calculated the velocity of each leaf as well as the velocity error VMAT The velocity RMSD of prostate, head and neck, and prostate pelvic node VMAT treatments were calculated for Truebeam machines. Much like the RMSD of the positional error the velocity RMSD for a single delivery was calculated from all time steps and all leaves from both banks during the treatment Sliding Gap The velocity error was also analysed for multiple sliding gap treatments on a Clinac ix machine. The sliding gap treatments were split into three distinct categories of leaf speed 10mm/s, 20mm/s, and 30mm/s to analyse the effect leaf velocity had on velocity error Treatment Modalities The velocity RMSD was calculated and compared for prostate dimrt deliveries and head and neck VMAT deliveries for machine model Clinac ix to compare the velocity error of different treatment modalities. 50

71 3.8 Gamma Map Analysis Gamma map analysis was performed on multiple different Varian log files delivered by different treatment machines to assess the gamma map pass rate, that is what percentage of the treatments were deemed similar. Head and neck, prostate, and prostate pelvic node VMAT treatments were analysed for Truebeam machines and head and neck, and prostate VMAT treatments were analysed for Clinac ix machines. Gamma map analysis was performed three different times for each treatment, each with different D and DTA. These were, 3%/3mm, 2%/2mm, and 1%/1mm in the format D/DTA. The average and standard deviation of the gamma pass rates were calculated for each treatment site for each D/DTA. Ling s Test 1, 2, and 3 which were delivered by Clinac ix machines were also analysed in a similar manner. 3.9 Gravity Picket Fence (Ling s Test 1) Varian log files of Ling s Test 1 was assessed to determine the relationship between the gantry angle and the positional error of the treatment. The data was split into three different categories; error of the left bank, error of the right bank, and the leaf gap width error. This was performed on multiple Ling s Test 1 that were delivered by 7 different machines of model, Truebeam, Clinac ix, and Trilogy. As the leaves are alternating between moving and stationary positions the error was only recorded when the leaves were stationary to remove any dependence of the error upon velocity or acceleration. The data was also categorised based on which treatment machine it was delivered on. For each machine the corresponding Varian log files were collated into one large array. This data was then averaged over 2 control point positions to allow for easier graphical representation. Additionally student T tests were performed on the error distributed in 40 arcs centred around gantry angles 0, 90, 180, and 270 to assess their similarities to one another. 51

72 3.9.2 Sliding Gap Varian log files of Sliding Gap quality assurance were assessed to determine the relationship between the gantry angle and the positional error of the treatment. Sliding Gap QA was delivered by treatment machine ix 3 at four distinct static gantry angles, 0, 90, 180, and 270. These were also delivered at discrete velocities, 10mm/s and 20mm/s. The error of each delivery was recorded for each leaf and time step VMAT Varian log files of VMAT treatments were analysed to compare the gravity effect upon the positional errors of the treatment. The Varian log files were split into groups based on their collimator angle. The sets with the greatest number of treatments, collimator angle 30 and collimator angle 330 were analysed and as before, the total error of the individual banks was averaged over 2 control points Age The RMS positional error was calculated for multiple Varian log files of Ling s Test 1 delivered on Truebeam machines on a daily basis. The RMSD was averaged over a monthly basis for each machine and a Pearson correlation was used to explore the relationship between the two variables Statisical Analysis Single Tailed Student T-tests were performed to assess the null hypothesis that the populations of two different distributions had equal means at α = To determine if the distributions were normally distributed and hence suitable for Student T-tests, the distributions was assessed graphically. If distributions appeared to not be normally distributed a Wilcoxon rank sum test was instead used. A Pearson correlation was also utilised to determine if there was a relationship between the error of QA tests and the age of the treatment machine. 52

73 CHAPTER 4 RESULTS 4.1 Development of software for automatic Varian log file analysis A program entitled MountainGoat with a basic GUI was constructed in MATLAB that imports Varian log files and performs the following functions on individual treatments: Assesses individual leaf error as a function of gantry angle Assesses individual leaf error as a function of time Produces a error histogram Produces a fluence map (planned, actual, and percentage difference) Takes note of how many times the leaves flagged (error > 1mm) Calculates Gamma Pass Rates for 1mm/1%, 2mm/2%, and 3mm/3%. Exports graphs Exports Excel spreadsheet summarising key data Primary analysis of the Varian log files takes less than 5 seconds. If the user wishes to export the graphs or an Excel spreadsheet then that will take upwards of half a minute. A users guide detailing how to use MountainGoat can be found in the appendix. 53

74 Figure 4.1: The velocity of the leaves with respect to time for a Varying Sliding Gap QA computed by a) Varian log file analysis, and b) EPID analysis. The white circles highlight the discrepancy of leaf number 29. Table 4.1: Comparison between Argus QA and Timberwolf. RMSD (mm) Delivery Type Argus QA Timberwolf Percentage Difference Picket Fence % Sliding Gap % dimrt % VMAT % 4.2 Confirmation of MATLAB software EPID measurements of sliding gap treatments were used to benchmark the accuracy of the constructed program, Timberwolf. The velocity of the leaves were both calculated for each bank using EPID measurements and Varian log files and can be seen in figure 4.1. Note the inaccuracy picked up by both methods of leaf number 29 on the left bank. Additionally Timberwolf was compared with the commercially available software, Argus QA. Four different treatments representative of the data set were chosen at random to be analysed. The total RMSD of each delivery was computed by both Timberwolf and Argus QA and is tabulated in table

75 RMSD of Picket Fence QA for different Treatment Machines RMSD (mm) Trilogy Clinac ix Truebeam Treatment Machines Figure 4.2: The RMSD of 32, 25, and 1805 Picket Fence deliveries were computed for Trilogy, Clinac ix, and Truebeam machines respectively. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. Timberwolf calculated identical values to the commercially available software of the RMSD for each treatment. 4.3 Leaf Error Picket Fence (Ling s Test 1) 32, 25, and 1805 Picket Fence (more accurately Ling s Test 1) deliveries were analysed to determine the RMSD for Clinac ix, Trilogy, and Truebeam machines respectively. This is presented graphically in figure 4.2 and tabulated in table 4.2. Small variation in the RMSD of Picket Fence deliveries were noted in Clinac ix and Trilogy models with their mean and standard deviation values both similar to each other. The error present in the Truebeam machines for routine Picket Fence tests were substantially lower than the other two machine models. Student T-tests were performed on the different sets to test the null hypothesis that they had equal means. The data of Clinac ix and Trilogy machines were statistically different (p=0.031), as was the data of Clinac ix or Trilogy machines when compared to the Truebeam machine s data (p<0.001). 55

76 Table 4.2: RMSD of Picket Fence deliveries. Results are presented as the mean of the RMSD and standard deviation. Mean (mm) STD (mm) Clinac ix Trilogy Truebeam Step and Shoot 1643 Step and Shoot prostate (PROS) treatment files were delivered on machine ix 6 and their positional errors for each leaf at each time step were analysed. This is represented graphically in figure 4.3. The mean of the positional errors is mm with a standard deviation mm. In total, points were assessed for error with an average of nearly 700 points per leaf per treatment. The majority of errors are well below the threshold that triggers the treatment flag error of 1mm. There were 5 anomalies in the points which resulted in leaves with error; -8.75mm, 16.06mm, mm, mm, and mm that were ignored. These errors are a result of the temporal proximity to the beam rapidly turning off and on. This causes errors to arise as discussed in section When the beam is halted the expected position of the leaves are assigned values of the next segment whereas the actual position of the leaves display their positional values as the leaves travel to their next location. Due to the short beam off time the leaves do not reach their destination by the time the beam starts delivering radiation again and the difference between actual and expected values of the leaves cause these error spikes to occur. A Gaussian distribution with two peaks was fitted to the histogram with R 2 =0.98. The errors lay between -0.1mm and 0.1mm. The distribution appears bimodal in nature. A Gaussian distribution with two peaks was fitted to the histogram with R 2 =0.98. The distribution appears bimodal in nature dimrt 7 prostate dimrt treatments delivered on Clinac ix machines were analysed to determine the positional errors for each leaf at timestep. This is represented graphically in figure 4.4. The prostate treatments had a mean of mm and standard deviation 56

77 15 x 105 Error of Step and Shoot Prostate Treatment 10 Coutns Error (mm) Figure 4.3: Error histogram of 1643 Step and Shoot prostate treatments delivered on machine ix 6. In total points were assessed. A double gaussian distribution was fitted with R 2 = mm. A total of points were analysed Error of dimrt Prostate Treatments Counts Error (mm) Figure 4.4: Errors of 7 different dimrt prostate treatments from Clinac ix machines. A double gaussian distribution was fitted with R 2 = VMAT The RMSD was calculated individually for 1668 VMAT treatments from machines True 1 and True 2. These treatments were split into three distinct sections based on the treatment site; Head and Neck (H&N; True 1 76 treatments, True treatments ), Prostate Pelvic Node (PPN; True treatments, True treatments), and Prostate (PROS; True treatments, True treatments). This is represented 57

78 RMSD of VMAT at different treatment sites for Truebeam a) b) RMSD (mm) H&N PPN PROS 0.01 Treatment Sites H&N PPN PROS Figure 4.5: RMSD of VMAT treatments at treatment sites Head and Neck (H&N), Prostate Pelvic Node (PPN), and Prostate (PROS), for treatment machines a) True 1 and b) True 2. A total of 76 H&N, 182 PPN, and 203 PROS treatments were analysed for True 1 and 444 H&N, 570 PPN, and 193 PROS treatments were analysed for True 2. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. Table 4.3: RMSD of VMAT deliveries at different treatment sites for True 1 and True 2. Results are presented as the mean of the RMSD and standard deviation. Mean (mm) STD (mm) True 1 H&N PPN PROS True 2 H&N PPN PROS graphically in figure 4.5 and tabulated in 4.3. For both True 1 and True 2, VMAT treatments for the prostate had the lowest mean RMSD for their respective machines. PPN treatments had the highest mean RMSD. Student T-test analysis for True 1 and True 2 deems all treatment sites statistically different from one another on their respective machines (p <0.001). The RMSD for individual H&N VMAT treatments of ix 4 (4 treatments) and ix 5 (28 treatments) were also calculated. This was compared against the RMSD of H&N VMAT treatments delivered by Truebeam machines and is represented graphically in figure 4.6 and tabulated in table

79 RMSD of VMAT H&N Treatments RMSD (mm) Clinac ix Treatment Machines Truebeam Figure 4.6: The RMSD of VMAT Head and Neck treatments for treatment machines Clinac ix (32 files) and Truebeam (520 files). The mean RMSD for Clinac ix machines (0.5855mm) and standard deviation (0.0426mm) was much higher than their Truebeam counterpart (mean: mm, standard deviation: mm) for H&N VMAT treatments. A Student T-test concludes they are statistically different (p<0.001). Table 4.4: RMSD of H&N VMAT deliveries for Truebeam and Clinac ix machines. Results are presented as the mean of the RMSD and standard deviation. Mean (mm) STD (mm) Clinac ix Truebeam Comparing Treatment Methods - Clinac ix The RMSD for different treatment modalities of all Clinac ix treatment machines are represented graphically in figure Prostate (PROS) Step and Shoot, 7 PROS dimrt, and 32 Head and Neck (H&N) VMAT treatments were analysed. Unfortunately there were no H&N treatments available for Step and Shoot or dimrt, nor were there PROS treatments available for VMAT so a direct comparison could not be made. The Step and Shoot modality for PROS had the lowest mean RMSD (0.0384mm) and STD (0.0235mm). dimrt for PROS had a higher mean RMSD (0.3317mm) and 59

80 RMSD of different treatment modalities of Clinac ix RMSD (mm) Step & Shoot (PROS) dimrt (PROS) VMAT (H&N) Figure 4.7: RMSD of different treatment modalities for Clinac ix Prostate Step and Shoot, 7 Prostate dimrt, and 32 Head and Neck VMAT treatments were analysed. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. STD (0.0231mm) which is less than the mean RMSD value of VMAT treatments for H&N (mean: mm, STD: mm). All treatment modalities were statistically different from each other (p<0.001) Positional Errors due to Velocity The absolute positional errors of three different Sliding Gap QA deliveries is presented graphically in figure 4.8. The deliveries were segregated into different leaf sweeps with velocities 10mm/s, 20mm/s, and 30mm/s. The data was corrected by subtracting the absolute error by the delay present in the MLC system (55+ms) multiplied by the leaf velocity. This removes the error that arises from the delay in the MLC system. Four 10mm/s, four 20mm/s, and three 30mm/s sliding gap deliveries were analysed. For the uncorrected data the mean of the absolute error for each sliding gap delivery increased with increasing velocity. For the corrected data the mean of the absolute error also increased with increasing leaf velocity although to a lesser extent. This is summarised in table 4.5 Student T-tests on the uncorrected data and Wilcoxon rank sum tests on the cor- 60

81 Error of Sliding Gap QA for different velocities on ix_3 2 a) 1 b) Error (mm) Leaf Velocities (mm/s) Figure 4.8: Absolute errors from three different Sliding Gap QA. The Sliding Gap QA was performed 3 times for 30mm/s Sliding Gap QA and 4 times for 10mm/s, and 20mm/s Sliding Gap QA. The errors of each delivery were: a) uncorrected, b) corrected for the 55+ms delay present in older Varian systems. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. Table 4.5: The RMSD of Sliding Gap QA moving at different velocity presented as mean and standard deviation. The corrected deliveries were corrected for the 55ms+ delay present in older Varian systems. Mean (mm) Standard Deviation (mm) Uncorrected 10mm/s mm/s mm/s Corrected 10mm/s mm/s mm/s

82 RMSD (STD) Machine a) InnerA RMSD InnerB of OuterA Picket OuterB Fence InnerA InnerB OuterA ix_ ( ( ( (0.2252) ix_ ix_ ( ( ( (0.2304) ix_ Tri_ ( ( ( (0.2398) Tri_ Tri_ ( ( ( ( Tri_ True_ ( ( ( (0.0138) True_ True_ ( ( ( (0.0140) True_ True_ ( ( ( (0.0150) True_ RMSD (mm) 0.24 ix_1 ix_2 Tri_1 Tri_2 InnerA InnerB OuterA OuterB b) RMSD (mm) RMSD of Picket Fence True_1 True_2 True_3 InnerA InnerB OuterA OuterB Figure 4.9: RMSD of inner and outer leaves for Bank A and Bank B of: a) Clinac ix and Trilogy machines, and b) Truebeam machines. 13, 19, 19, 6, 732, 704, and 369 picket fence deliveries were analysed for ix 1, ix 2, Tri 1, Tri 2, True 1, True 2, and True 3 respectively. rected data confirm that the various sets of the uncorrected data are statistically different from one another (p<0.001). The corrected data is also significantly different from one another (p <0.001). 4.4 Inner/Outer Leaves The RMSD of the inner leaves and the outer leaves for bank A and bank B of Picket Fence (Ling s Test 1) tests is represented graphically in figure 4.9 and in table 4.6. The RMSD of each category was calculated from the amalgamated picket fence data from each machine. Results show an increase in RMSD for inner leaves against their outer leaf counterparts residing on the same bank. And for some machines, notably ix 1 and ix 2, the RMSD of each bank was also distinctly different. Furthermore there is also a large difference in the errors between the Truebeam machines and the two older linear accelerator models which agrees with the results from section

83 Table 4.6: a) RMSD and b)std of inner and outer leaves of Bank A and Bank B for different machines. 13, 19, 19, 6, 732, 704, and 369 picket fence deliveries were analysed for ix 1, ix 2, Tri 1, Tri 2, True 1, True 2, and True 3 respectively. a) RMSD (mm) Bank A Bank B Inner Outer Inner Outer ix ix Tri Tri True True True b) STD (mm) Bank A Bank B Inner Outer Inner Outer ix ix Tri Tri True True True

84 Table 4.7: Velocity RMSD of VMAT deliveries at different treatment sites for True 1 and True 2. Results are presented as the mean of the Velocity RMSD and standard deviation. Mean (mm/s) STD (mm/s) True 1 H&N PPN PROS True 2 H&N PPN PROS Each set of inner leaves were compared with their corresponding set of outer leaves for each treatment machine and bank. Each inner and outer leaf pair was deemed statistically different (p<0.001). 4.5 Velocity Leaf Error VMAT The Velocity RMSD was calculated individually for 1668 VMAT treatments from machines True 1 and True 2. These treatments were split into three distinct sections based on the treatment site; Head and Neck (H&N; True 1 76 treatments, True treatments ), Prostate Pelvic Node (PPN; True treatments, True treatments), and Prostate (PROS; True treatments, True treatments). This is represented graphically in figure 4.10 and tabulated in 4.7. For both True 1 and True 2, VMAT treatments for the prostate had the lowest mean velocity RMSD for their respective machines. PPN treatments had the highest mean velocity RMSD. All treatment sites were significantly different from another (p<0.001) Sliding Gap The velocity of three different Sliding Gap QA deliveries with differing leaf speed were analysed and are represented graphically in figure The deliveries were segregated into 3 distinct groups with differing leaf velocities, 10mm/s, 20mm/s, and 30mm/s. Four 10mm/s, four 20mm/s, and three 30mm/s sliding gap deliveries were analysed. The mean velocity error and standard deviation of each leaf velocity for the sliding 64

85 0.28 RMSD (velocity) of VMAT treatments at different sites for Truebeam 0.28 RMSD (mm/s) H&N PPN PROS Treatment Sites H&N PPN PROS Figure 4.10: The Velocity RMSD for VMAT treatments delivered at different treatment sites on True 1 (left) and True 2 (right). A total of 76 H&N, 182 PPN, and 203 PROS treatments were analysed for True 1 and 444 H&N, 570 PPN, and 193 PROS treatments were analysed for True 2. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. Velocity Error of Sliding Gap QA for different velocities on ix_ Error (mm/s) Leaf Velocity (mm/s) Figure 4.11: The error of velocity in different Sliding Gap QA deliveries on ix 3. Four 10mm/s, four 20mm/s, and three 30mm/s sliding gap deliveries were analysed. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. 65

86 Table 4.8: The Velocity RMSD of Sliding Gap QA moving at different velocity presented as mean and standard deviation. Leaf Speed Mean (mm/s) STD (mm/s) 10mm/s mm/s mm/s Velocity RMSD for different treatment modalities on Clinac ix 3.5 RMSD (mm/s) dimrt VMAT Figure 4.12: Velocity RMSD of different treatment modalities for Clinac ix. 7 dimrt (PROS) and 32 VMAT (H&N) treatments were analysed. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. gap QA is presented in table 4.8. The velocity error and standard deviation increases with increasing leaf velocity. Wilcoxon rank sum confirmed that all sets were statistically different from each other (p<0.001) Treatment Modalities The velocity RMSD was calculated for different treatment modalities on Clinac ix treatment machines and is represented graphically in figure A total of 7 PROS dimrt treatments and 32 H&N VMAT treatments were analysed. No Step and Shoot modalities were analysed due to the leaves remaining stationary when the radiation is delivered. The VMAT treatments possess a higher mean velocity RMSD (3.8266mm/s) with 66

87 Table 4.9: Gamma passing rates calculated through Varian log file analysis for Head and Neck (H&N), Prostate Pelvic Node (PPN), and Prostate (PROS) treatments. Results are presented as mean and STD. H&N PPN PROS Gamma Mean(%) STD(%) Mean(%) STD(%) Mean(%) STD(%) True 1 3%3mm %2mm %1mm True 2 3%3mm %2mm %1mm ix 4 3%3mm %2mm %1mm ix 5 3%3mm %2mm %1mm STD (0.2939mm/s) when compared with the dimrt treatments mean velocity RMSD of mm/s and STD mm/s. 4.6 Gamma Map Analysis Gamma map analysis was performed on fluence maps created through the analysis of Varian log files to calculate the Gamma passing rate of VMAT treatments at 3%/3mm, 2%/2mm, and 1%/1mm parameters. The parameters are of the form, D/DTA. The gamma passing rates are presented in table 4.9. The VMAT treatments were separated based on treatment sites as well as by treatment machine. 76 H&N, 182 PPN, and 203 PROS were analysed for True H&N, 570 PPN, and 193 PROS treatments were analysed for True 2. 4 H&N treatments were analysed for ix 4, and 28 H&N treatments were analysed for ix 5. Regardless of machine, treatment site, or gamma map analysis parameters the mean gamma passing rate was always above 99.5%, that is, on average 99.5% of the data points on an expected fluence map were similar enough to their actual fluence map counterpart to warrant a pass. The lowest mean gamma passing rate (99.70%) and the highest STD (0.57%) both belonged to the prostate treatment, 1%/1mm on True 1. Ling s Test 1 (LT2), Ling s Test 2 (LT2), and Ling s Test 3 (LT3) were also analysed 67

88 Table 4.10: Gamma passing rates calculated through Varian Log file analysis for 20 LT1, 44 LT2, and 17 LT3 QA tests. All tests were delivered on Clinac ix machines. Results are presented as mean and STD. Gamma Mean(%) STD(%) LT1 3%3mm %2mm %1mm LT2 3%3mm %2mm %1mm LT3 3%3mm %2mm %1mm in a similar manner. All QA tests were delivered on Clinac ix machines. 20, 44, and 17 deliveries for LT1, LT2, and LT3 respectively. The gamma passing rates are presented in table According to the analysis, LT3 was delivered the most accurately, followed by LT1 and then LT2. An example of a gamma map analysis is shown in figure This is of a H&N treatment from ix 5. The colour intensity relates to how similar the two fluence maps were. The higher the intensity the less similar they are. This treatment had a gamma map pass rate of 100%. 4.7 Gravity Picket Fence (Ling s Test 1) Ling s Test 1 QA for True 1, True 2, True 3, ix 1, ix 2, Tri 1, and Tri 2 was analysed to assess the effect of gravity on the positional error of the leaves. Data was only recorded when the leaves were stationary. The error of all the leaves for each treatment machine was averaged over multiple 2 degree control points. Three of these results, ix 1, Tri 1, and True 1 are presented graphically in figure 4.14 with the others available for viewing in Appendix. Student T-tests were also performed on 40 degree arcs centred around gantry angles 0, 90, 180, and 270 for bank A, bank B, and the Leaf Gap to test the null hypothesis 68

89 Gamma Map Analysis (3%/3mm) Gamma Map Value Figure 4.13: Gamma map analysis analysing the expected and actual fluence of a H&N treatment from ix 6 using 3%/3mm. Colour intensity details how similar the two fluence maps are. The formula for calculating these values was presented in section Table 4.11: P-values of Student T-tests between the averaged Control Point error located in 40 degree arcs centered on gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Picket Fence tests from ix 1 were analysed and data was only taken into account if the leaves were stationary. Bank A Bank B Leaf Gap that their respective means were similar to the other gantry angles. P-values are tabulated for ix 1, Tri 1, and True 1 in table 4.11, table 4.12, and table 4.13 respectively. The tables are colour coded such that p-values <0.05 are red. For treatment machine ix 1 the mean error was statistically different between each gantry angle pair except for gantry angle pair, 0 and 180. We cannot reject the null hypothesis that the mean error between gantry angles 0 and 180 for bank A are dissimilar. This is true for both bank A and bank B of the MLC. However the mean error of each gantry angle window of the Leaf Gap was statistically different. For bank A of treatment machine Tri 1 all gantry angle windows were statistically different. For bank B we could not reject the null hypothesis between gantry angles 0 69

90 Error vs Gantry Angle Ling s Test i) iii) a) ii) b) c) Figure 4.14: Polar plot of the error (mm) in the radial component that has been averaged over 2 degree control points, vs the gantry angle (IEC convention) for a) ix 1 (13 tests) b) Tri 1 (14 tests) and c) True 1 (732 tests). This was further split into the error present on i) Bank A, ii) Bank B, and iii) the leaf gap width. Table 4.12: P-values of Student T-tests between the averaged Control Point error located in 40 degree arcs centered on gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Picket Fence tests from Tri 1 were analysed and data was only taken into account if the leaves were stationary. Bank A Bank B Leaf Gap

91 Table 4.13: P-values of Student T-tests between the averaged Control Point error located in 40 degree arcs centered on gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Picket Fence tests from True 1 were analysed and data was only taken into account if the leaves were stationary. Bank A Bank B Leaf Gap and 180. And for the leaf gap error we could not reject the null hypothesis between gantry angles 0 and 90 or 90 and 180. This is similar to Bank A of True 1 as the null hypothesis could not be rejected between gantry angles 0 and 90 or 90 and 180. For bank B we cannot say that the error between gantry angles 0 and 180, 90 and 180, or 90 and 270 are different. And for the leaf gap error we cannot reject the null hypothesis between gantry angles 0 and 180 or 90 and 180. Everything else was statistically different. There is a trend present on bank A for a greater error to be present at gantry angle 270 when compared with gantry angle 90. Conversely on bank B, ix 1 and True 1 both have marginally greater error at gantry angle 90 although Tri 1 still possesses greater error at gantry angle 270. The error present in the leaf gap width also shows dependence upon the gantry angle with greater error at gantry angles 90 and 270 save for the data collected for Tri 1 which shows minimal change in error for gantry angle 90 when compared with gantry angle Sliding Gap Sliding Gap QA from ix 3 was analysed to compare the error of the MLC to the gantry angle that it was delivered at. The QA was performed four times at discrete gantry angles, 0, 90, 180, and 270. This was repeated for multiple discrete velocities of the leaves. One such result, for when the leaves were travelling at 10mm/s is represented graphically in figure 4.15 and tabulated in table The analysis of leaves travelling at 20mm/s is represented graphically in figure 4.16 and tabulated in table

92 a) Error of SG QA at different Gantry Angles Error (mm) 0.4 b) Gantry Angle Figure 4.15: The error (mm) for Sliding Gap QA delivered at four different gantry angles on ix 3 for a) Bank A, and b) Bank B. Leaves moved with velocity 10mm/s. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. Table 4.14: The error for Sliding Gap QA (10mm/s) delivered at four different gantry angles on ix 3 on Bank A and Bank B. Results are displayed as mean and STD. Gantry Angle Mean (mm) STD (mm) Bank A Bank B Each group contained approximately 6000 data points. In both banks gantry angle 270 had the greatest mean error as well as the greatest STD. Conversely gantry angle 90 had the least mean error in both banks. Each gantry angle on each bank was analysed to determine if it was statistically significant from other gantry angles residing on the same bank. All gantry angles were found to be statistically different from each other, each with a p-value < The analysis of sliding gap QA when delivered at leaf speed 20mm/s held a similar story. Gantry angle 270 had the greatest error and gantry angle 90 had the least. All gantry angles were statistically different from one another except for gantry angles

93 a) Error of SG QA at different Gantry Angles Error (mm) b) Gantry Angle Figure 4.16: The error (mm) for Sliding Gap QA delivered at four different gantry angles on ix 3 for a) Bank A, and b) Bank B. Leaves moved with velocity of 20mm/s. The box and whisker plot represents the data as follows: The top line (or whisker) is the maximum value, the next line is the 3rd Quartile, the red line is the median, the next line is the 1st Quartile, and the last line is the minimum value. The red crosses are outliers. Table 4.15: The error for Sliding Gap QA (20mm/s) delivered at four different gantry angles on ix 3 on Bank A and Bank B. Results are displayed as mean and STD. Gantry Angle Mean (mm) STD (mm) Bank A Bank B and 0, as well as gantry angles 180 and 90 on bank A. This can be seen in table VMAT 1451 VMAT treatments delivered on Truebeam machines were analysed to determine the effect of gravity on the positional accuracy of the bank A and bank B of the MLC. VMAT treatments differ from Picket Fence and Sliding Gap tests in many ways. Besides from the obvious distinction between quality assurance tests and actual treatments, the collimator angle for quality assurance tests are set to 0 (IEC convention) for delivery. For VMAT treatments the collimator angle is not always 0. This has the consequence 73

94 Table 4.16: P-values of Student T-tests between the error of Sliding Gap treatments (20mm/s) delivered at gantry angles 0, 90, 180, and 270. Colour coded such that the value is red if it is <0.05. Values of 0 represent P-values < Bank A Bank B Collimator Angle for VMAT Truebeam treatments 500 X= X=330 Counts Collimator Angle (degrees) Figure 4.17: A histogram of the collimator angles (IEC convention, degrees) of 1451 VMAT treaments delivered on Truebeam machines. that at gantry angle 90 or 270 the motion of the MLC leaves will not always be parallel to the force of gravity. Fortunately for the 1451 VMAT treatments analysed, although there was a wide variety of collimator angles the collimator angle itself remained stationary throughout delivery. A histogram of the collimator angles for the 1451 Truebeam VMAT deliveries is seen in figure The two most prominent collimator angles that are used for these deliveries are collimator angles 30 and VMAT treatments had collimator angle 30 and 398 treatments had collimator angle 330. Subsequently the treatments were split into groups that possessed collimator angle 30 and those with collimator angle 330. For each group the mean total error was obtained for 180 different, 2 degree wide, control point arcs of the gantry angle. This is represented graphically in figure

95 a) i) Error vs Gantry Angle VMAT Truebeam treatments ii) iii) b) Figure 4.18: Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for treatments with collimator angle (IEC convention) i) 30, ii) 330, and iii) all collimator angles. These were also split into a) Bank A and b) Bank B. There were 540, 398, and 1451 treatments for collimator angles 30, 330, and all collimator angles respectively. 75

96 The polar plots possess the same basic structure as the polar plots created from Ling Test 1 data from the Truebeam machines. However it is interesting to note that the distribution of the average positional errors is dependent upon the collimator angle. The polar plot of collimator angle 30 and 330 rotate the distribution found at collimator angle 0 of Ling s Test 1. It is fascinating that when averaged over all collimator angles the absolute error is actually a whole magnitude less than the absolute error present in an unique collimator angle. 4.8 Age The RMSD for True 1, True 2, and True 3 of routine Picket Fence deliveries were averaged on a monthly basis and plotted against the month they were delivered. This is represented graphically in figure Data that was analysed for each machine was collected from their commission until July Pearson correlation (R) was used to explore the relationship between these two variables. Variables were deemed correlated for R>0.5. The monthly averaged data had R value , , and for True 1, True 2, and True 3 respectively, all of which were significant (p<0.05). A linear trend was then fitted to True 1, True 2, and True 3. There is a linear trend of increasing error as time increases in each treatment machine. The maximum difference in the monthly averaged RMSD over the course of data collection for True 1, True 2, and True 3 is mm, mm, and mm respectively. The gradients for True 1, True 2, and True 3 are 2.5e-08 mm/month, 4.0e-08 mm/month, and 2.3e-07 mm/month respectively in arbitrary units. 76

97 RMS Error of Picket Fence vs Date Delivered (Monthly Averaged) RMS Error (mm) Jan2013 Apr2013 Jul2013 Oct2013 Jan2014 Apr2014 Jul a) b) Jul2013 Oct2013 Jan2014 Apr2014 Jul c) d) Jan2013 Apr2013 Jul2013 Oct2013 Jan2014 Apr2014 Jul True_ True_2 True_ Jan2013 Apr2013 Jul2013 Oct2013 Jan2014 Apr2014 Jul2014 Date Delivered Figure 4.19: RMS error for of routine Picket Fence deliveries vs date delivered for: a) True 1, b) True 2, c) True 3, and d) all of the above. Results were averaged on a monthly basis. A linear trend was fitted to each of the machines with R-value: a) R = , b) R = , and c) R =

98 4.9 Errors and Treatment Machine Model The results of this study indicate that the positional error of the MLC depends upon the treatment machine, treatment site, treatment modality, and the velocity of the MLC leaves, as well as the gantry angle. Perhaps the most drastic changes to the positional error resulted from the type of machine model the treatment was delivered on. Both Clinac ix and Trilogy machines possessed a mean RMSD an order of magnitude higher than Truebeam machines when delivering Ling s Test 1. Clinac ix and Trilogy machines had error of the same order but were statistically different. Additionally the RMSD of VMAT head and neck (H&N) treatments are an order of magnitude higher when delivered on Clinac ix machines when compared with Truebeam machines. While this is a rather large difference in error this is to be expected. The supervisor module installed in Truebeam machines supervisors the leaf positions and minimises error in the delivery. The relative errors of the different machine models agrees with the literature. Agnew et al compared treatment plans delivered on Truebeam and C-series linear accelerators (of which Clinac ix machines are a part of) and found the RMS leaf position errors to be an order of magnitude greater in the C-series [69]. Their treatment plans had mean RMS error of 0.023mm and 0.239mm for Truebeam and C-series linacs respectively. This is of the same order as the results obtained from our analysis of Ling s Test 1 of mm and mm for Truebeam and Clinac ix respectively Errors and Treatment Site VMAT Truebeam deliveries possessed the greatest mean RMSD when delivered to the prostate pelvice node (PPN) treatment site followed by deliveries to the H&N treatment site and then deliveries to the prostate. The mean RMSD of the H&N and 78

99 PPN treatment sites both showed an increase of approximately 40% when compared to the mean RMSD of the prostate treatments. Additionally each treatment site was statistically different to one another. The literature agrees that prostate treatments have the least RMSD followed by H&N, and then PPN for treatments with no overshoot [62]. It should noted that Truebeam machines are note prone to overshoot [69]. This is echoed in the analysis of the velocity RMSD of different treatment sites. VMAT Truebeam deliveries possessed the greatest mean velocity RMSD when delivered to the PPN treatment site followed by deliveries to the H&N treatment site and then deliveries to the prostate Error Distribution of Prostate Treatments Recall the histogram of the errors from Step and Shoot prostate treatments (figure 4.20) and dimrt prostate treatments (figure 4.21). The bimodal nature of the histogram of the errors in Step and Shoot prostate treatments distribution in figure 4.20 arises due to the fact both bank A and bank B of the MLC was analysed. For each treatment, bank A and bank B had errors of opposing signs. If bank A had positive error then bank B had negative error and vice versa. In this collection of Step and Shoot treatments, bank A had positive error for 44% of the treatments and bank B had positive error for the remaining 56% of the treatments. Additionally the distribution of each bank was centred around nonzero values. As such when the histograms of the two different positive and negative error distributions are combined then the resulting distribution is bimodal. Like the Step and Shoot treatments the dimrt treatments (figure 4.21) were also split into two different distributions, one of which had positive error and the other of which had negative error. These arise from the two different banks that are analysed. However unlike the Step and Shoot treatments, all of the positive errors for all of the dimrt treatments originated from bank A and all the negative errors originated from bank B. This is demonstrated in figure Interestingly the shape of these histograms suggest that the shift of the error present in the Step and Shoot treatments are not present here which is not just a consequence 79

100 15 x 105 Error of Step and Shoot Prostate Treatment 10 Coutns Error (mm) Figure 4.20: Error histogram of 1643 Step and Shoot prostate treatments delivered on machine ix 6. In total points were assessed.a double Gaussian was also fitted Error of dimrt Prostate Treatments Counts Error (mm) Figure 4.21: Errors of 7 different dimrt prostate treatments from Clinac ix machines. A double Gaussian was also fitted. 80

101 Figure 4.22: Errors of 7 different dimrt prostate treatments from Clinac ix machines split into Bank A and Bank B errors. of the resolution of the dimrt histogram. Further analysis reveals that the median values of the distribution is 0. The positive error originating from bank A of the prostate dimrt deliveries and the negative error originating from bank B agrees with the literature [92]. This phenomenon can be explained due to how the leaves traverse in dimrt treatments. They move unidirectionally; Bank A starts retracted and extends and the leaves on Bank B start extended and retract. This translates to the positional values that are recorded to Varian log files, of the leaves on Bank A, to be continuously increasing. Conversely the positional values recorded to the Varian log files are continuously decreasing with respect to time. Recall that the error is calculated as follows: X = X A exp X A act That is, the error is the expected value of the leaf s position subtracted by the actual position of the leaf. Also recall that there is a delay of 55ms in the MLC system of Clinac ix machines. As such, for bank A, the delay ensures that the expected value will be larger than the actual value as the leaves lag behind their expected position. In bank B the delay ensures the expected value will be less than the actual value and hence the resulting error is negative. The MLC system delay causing systematic shifts in dimrt treatments could have 81

102 Table 4.17: Comparison of treatment modalities by two different groups. Mean leaf RMS error (mm) Modality Kern et al. This Work Step and Shoot dimrt VMAT notable dosimetric effects on the treatment. There are multiple studies that have shown that systematic shifts have a noticeable effect on dosimetry whereas random shifts do not [83], [93] [95] Errors and Different Treatment Modalities A direct comparison between Step and Shoot, dimrt, and VMAT treatment modalities could not be made due to insufficient Varian log files. The lowest RMSD of deliveries administered on Clinac ix machines was that of the Step and Shoot modality. Specifically at the treatment site of the prostate. dimrt treatments focused on the prostate had greater RMSD than their Step and Shoot counterparts. Additionally VMAT treatments delivering to H&N treatment sites had the greatest RMSD of the three modalities examined. This agrees with the recent work of Kerns et al.[92] which compared step and shoot, dimrt, and VMAT modalities to one another. Their results also showed an increase of RMSD in increasing complexity of treatment modality and are compared to those of this thesis in table Whilst it is unfortunate that the comparison between the three treatment modalities could not be more standardised there is a trend of greater RMSD in VMAT when compared to dimrt and dimrt when compared to Step and Shoot. The more complex treatments with greater degrees of freedom result in greater error of the leaves. This is echoed in the analysis of the velocity errors of different treatment modalities. VMAT treatments had greater velocity RMSD when compared with the dimrt treatments. That is not to say that dimrt should be preferred over VMAT. Even though the leaves may have greater error there have been multiple studies that show the dose conformity to the PTV of VMAT treatments is equal or greater than those of dimrt treatments [14] [22]. 82

103 4.13 Errors and Leaf Velocity The greatest errors for Sliding Gap deliveries occurred when the leaves were moving at 30mm/s and the least error occurred when the leaves were moving at 10mm/s. Student T-tests confirm that the mean error increases with increasing leaf velocity. This trend is a result of two different elements. The first element arises from the MLC system delay of >55ms present in Clinac ix machines. It takes >55ms for the system to move the leaves into position. By the time that they are in position they already should be in the position designated by the following timestep meaning that when the leaves are travelling at a greater velocity, the distance the leaves have to travel in between subsequent timesteps, and hence what the system delay contributes to the positional error, is greater. The second element arises from the leaf velocity itself. Greater leaf velocity is harder to control from a mechanical standpoint. After the MLC system delay was taken into account there was still a trend of greater error attributed to deliveries with greater leaf velocity. This means that there is still leaf position error purely as a result of the mechanical constraints of the MLC. Even if the system delay was reduced, like in the Truebeam machines, there would still be a trend of greater error corresponding with greater leaf velocity. This is in line with numerous studies that suggest maximum leaf velocities in order to minimize treatment error [58], [74], [86], [92], [96] [99] It should be noted that the uncorrected error is the real error. Due to the delay in the MLC system the leaves undergo that error. The corrected error seeks to remove this dependence from the system lag and as such aims to view the error due to the velocity of the leaves without the effect of the system lag. Even with the effect of system lag removed the error of the leaves showed a dependence upon their velocity. As an aside this result is echoed when analysing the velocity error due to the differing leaf velocities. The mean velocity error was highest for leaf speed 30mm/s and lowest for leaf velocity 10mm/s. 83

104 4.14 Errors and Inner and Outer Leaves This study also determined the RMSD of outer and inner leaves for picket fence treatments. Each treatment machine, whether or not they were Clinac ix, Trilogy, or Truebeam machines, had greater error present in the inner leaves when compared to the outer leaves of the same bank. This is in spite of the inner leaves being half the width and weight of the outer leaves. The motors should have an easier time accurately directing the leaves to their intended positions. That being said the inner and outer leaves both are powered by different motors. The inner leaves are powered by motors with a resistance of 33 to 44Ω and the outer leaves are powered by motors with a resistance of 14Ω [74]. Another possible reason could arise from the wear and tear of the motors. The inner leaves are more often used during treatment while the outer leaves remain stationary and shielded by the jaws of the linear accelerator. The greater use of the inner leaves could cause the positional accuracy to degrade after a period of time however the trend is present in all the machines, even the Truebeam machines which are relatively new and will not have the same level of use as the older models. Additionally the leaf motors are also replaced after a period of time[88] casting further doubt upon this theory Gamma Map Analysis This study also performed gamma map analysis for VMAT treatments. All clinical deliveries performed excellently with the lowest mean gamma map pass rate being 99.70%. Analysis of Ling s Test 1, 2, and 3 was also met with excellent gamma map pass rates, the lowest being 93.32% but the majority obtaining 100% pass rates. It is not unheard of for the gamma map pass rate calculated through Varian log files to be so high. Agnew et al found that gamma map pass rates calculated through Varian log file analysis to be of a similar value for the 3%/3mm margins. Furthermore their gamma map analysis using data from a phantom ladened with ionisation chambers yielded a smaller value [58]. Due to their independent nature, dosimeters are more valued when checking the accuracy of deliveries. Varian log files can only pick up on the inaccuracies of the MLC whereas dosimeters can pick up on the inaccuracies 84

105 of the whole linear accelerator. As such the gamma map comparison computed from the Varian log files won t pick up inaccuracies due to the gantry sag or other external problems. That being said using Varian log files for gamma map comparison is quick and does not require external equipment which decreases the time needed for patient specific QA. Because of this Kumar et al argue that log file analysis, which includes gamma map analysis, is a promising tool for IMRT QA [59]. It is interesting to note that the older Clinac ix machines performed marginally better than their newer Truebeam counterparts. This could be a result of the relatively smaller sample size of the Clinac ix machines, especially when utilising the 3%/3mm parameters as all of the computed Gamma values have to be equal to 100 to achieve a mean score of 100. It could just be that the few deliveries by the Clinac ix machines were all exceptionally well planned whereas there were some relatively poorly planned treatments that were delivered by the Truebeam machines hence the lower mean values. The increased resolution of the Truebeam Varian log files would have no effect on the decreased Gamma score. The decreased timestep means that more calculations are required to compute the final fluence map which would normally magnify errors. However the MU delivered at each timestep is also smaller which means that each timestep has less of an impact on the final fluence thereby mitigating this effect. A more comparable sample size is required to comment on the differences between the Truebeam and Clinac ix machines. The large jump between the gamma passing rate of Ling s Test 1 between 2%/2mm and 1%/1mm may be attributed to the picket like nature of the dose distribution. When assessed with the parameters of 1%/1mm the thin picket like dose structures may have been out of range on the corresponding map causing more points to fail than the 2%/2mm counterparts. The numbers obtained by the Gamma map comparison are relatively high when compared to gamma map comparisons performed by external dosimeters on similar treatments [58]. The reason that the analysis of the Varian treatment files yields such high gamma map comparisons is twofold. Firstly as this process only requires the Varian log files for analysis there is no set up error in positioning a phantom. Other methods require a phantom to be set up on the 85

106 treatment table with dosimeters throughout to measure the delivered dose. However it is difficult to precisely align the phantom to sub millimetre accuracy. Even then the dosimeters themselves have their own physical resolution that introduces uncertainty when measuring the treatment. Additionally you are also making the assumption that the dosimeter is correctly calibrated. Secondly the process assumes that the beam is delivered perfectly, without any misalignment or bias or scatter. It assumes the beam is uniform in its delivery. This is not what happens in real life. There is unwanted scatter of the beam off the treatment head as well as backscatter from the couch. The beam itself also has other imperfections simply from the random nature of radiation. Furthermore there is the dose ramp up and fall off effects that are not taken into account with the Varian log files that external dosimeters will pick up on. All of this information is not available to the Varian treatment files which assumes the beam is perfect. It has to be said that the Varian treatment files are also dependent upon the machine. Imperfections of the machine, such as the sag of the gantry or of the MLC carriage[70], [90] will also not be picked up from this test due to the fact the MLC system is located within these structures. Neither will the error arising from a miscalibration of the MLC be detected, as the system is adamant that the leaves are in a specific location when in reality they are not. This is especially painful if there is a systematic shift of the entire MLC bank which can cause large dosimetric errors[83], [93], [95]. Additionally analysis of the Varian log files does not take into account the accuracy of the gantry angle[100]. As the Varian log files do not pick up on discrepancies of the MU delivered, the subsequent gamma analysis instead focuses more on the positional errors of the MLC. When coupled with the fact that most of the treatments have minuscule error in the positional locations of the leaves the resulting gamma map score is quite high. The main advantage to using Varian log files for gamma map comparison is the speed of the analysis. The quick nature of analysis of the Varian log files when compared to external dosimeters for gamma map comparison allow for gross errors to be quickly and easily picked up. Additionally as no phantoms need to be set up gamma map analysis can be performed on the actual treatment delivery. Finally comparing the Varian log files by means of gamma analysis validates information transfer from the 86

107 planning system to the treatment machine [101] Errors and Gantry Angles This study also looked into the affect gravity played on the positional errors of the MLC. Results from Ling s Test 1 tests indicate that on Bank A there is a trend towards greater error present at gantry angle 270 when compared with gantry angles 0, 90, and 180. There is also an increase in error at gantry angle 90 but to a lesser extent. This trend is similar to Bank B except the error at gantry angle 90 is more pronounced. The increase in error can be explained due to the presence of gravity. The error is greater at angles where the MLC motion is parallel to the force of gravity and less when it is perpendicular to the force of gravity. The asymmetric distribution of errors towards 270 instead of 90 could be a result of a number of factors. At gantry angle 270 and collimator angle 0, for Ling s Test 1 the MLC is positioned as below: Figure 4.23: The position of the MLC at gantry angles 90 and 270 for Ling s Test 1. The collimator angle is at 0. At gantry angle 270, the leaves on bank A are travelling in the direction of the force of gravity. At gantry angle 90, the leaves on bank A are travelling in a direction opposite to the force of gravity. The fact that the motion of the leaves on bank A is assisted by gravity at gantry angle 270 whereas the motion of the leaves on bank B is retarded by gravity at gantry angle 90 may lead to the asymmetric distribution of positional error. However this effect is not mirrored for bank B. The position of bank B is inverted such that at gantry angle 90 the leaves are moving opposite to the force of gravity and 87

108 at gantry angle 270 the leaves are moving in the same direction as the force of gravity. However the distribution of the positional error of bank B changes depending on the machine. For ix 1 the positional error of bank B for gantry angle 90 is slightly larger than the error at gantry angle 270. For Tri 1 the error for gantry angle 90 is actually smaller than the error found at 270 and for True 1 the error at gantry angle 90 is of the same order as the error at gantry angle 270. None of the distributions of bank B mirror those of bank A where the positional error at 270 is significantly larger than the error at gantry angle 90. The error of bank B at gantry angle 90 is either smaller, of the same order, or slightly larger than the error at gantry angle 270. The fact that this phenomenon is not restricted to one machine is interesting. bank A has consistently greater error at gantry angle 270 when compared to gantry angle 90 in all treatment machines whereas bank B is completely different. Ling s Test 1 was performed for both clockwise and counter clockwise rotations which erases dependence upon any errors arising from a temporal standpoint. More research needs to be done in this area to understand this phenomenon, focusing on delivering Ling s Test 1 at different collimator rotations to see if the direction of motion of the leaves effects the outcome or if the error is solely dependent upon the gantry angle. The results obtained from sliding gap deliveries at four different static gantry angles suggest a dependence of the positional errors on the gantry angle. For both bank A and B the error was more pronounced at gantry angle 270. Interestingly the positional error at gantry angle 90 was actually less than the error of the control gantry angles, 0 and 180. The same error distribution for both bank A and bank B suggests that there is not a dependence on the direction of motion of the leaves but instead purely on the gantry angle. Again the asymmetry of gantry angle 270 having greater error than gantry angle 90 for both banks suggests that there are other factors besides gravity that are coming into play. Further research has to be done to ascertain what these factors are and whether or not this phenomenon is localised to Clinac ix treatment machines or not. As an aside for the sliding gap deliveries, the positive error found in bank A and the negative error found in bank B is a consequence of how the error is calculated, the MLC system delay, and the movement of the leaves. This phenomenon was explained 88

109 when discussing the dimrt prostate deliveries. Finally the analysis of actual VMAT treatments delivered by Truebeam machines further supports the hypothesis that gravity effects the positional accuracy of the leaves. Treatments delivered with a collimator angle of 30 and treatments delivered with a collimator angle of 330 had different distributions for the positional accuracy versus the gantry angle. When the collimator angle is set to 0 the motion of the leaves are subjected to the full force of gravity at gantry angles 90 and 270. However this changes when the collimator rotates. The rotation of the collimator ensures that the direction of motion of the leaves are never parallel to the force of gravity. As such one would expect the distribution of the positional errors to change with different collimator angles as each collimator angle would have a corresponding gantry angle that provides the least resistance to the force of gravity. As the collimator rotates gravity will then start acting on the leaves such that they are pushed into one another causing interleaf friction to come into play however this was not investigated in this study due to time constraints. It is interesting to note that if all the VMAT treatments were analysed, that is the all the collimator angles are assessed together, then the resulting positional error versus the gantry angle is very similar to those produced by Ling s Test 1 for Truebeam machines. This is most probably coincidental and a result of averaging as a heavy skew towards any non-zero collimator angle would result in the distribution being rotated. As it stands the 540 treatments with collimator angle 30 is more or less balanced out by the corresponding 398 treatments with collimator angle 330. It is also fascinating to note that the error of all the collimator angles is a whole magnitude smaller than the error of treatments with collimator angle 330 or 30. This is due to the averaging of the errors from different collimator angles. Each collimator angle moves the maxima and minima such that when the errors are averaged the nodes of the error decrease the maximum error, ie: destructive interference. This agrees with the error of Ling s Test 1, all belonging to the same collimator angle, having the same order of magnitude as errors of either collimator angle 330 or 30. It should be noted that the error difference of the leaves at different gantry angles is quite small. The maximum error difference between gantry angles 270 or 90, and 0 or 180 of the leaves when they were stationary during Ling s Test 1 is less than 89

110 0.03mm. For sliding gap tests the maximum error difference between gantry angle 270 and 0 or 180 is mm. Of course these discrepancies may be magnified during real treatments where the leaves are not moving with routine patterns. This may be the case for treatment machines Clinax ix and Trilogy but the differences found in actual VMAT treatments were less than mm. To clarify, the Varian log files only record the effect upon the leaf position. The actual gantry head, as well as the MLC carriages themselves, do sag throughout the treatment by a significant amount[38], [70], [71] however the Varian log files do not pick up on this discrepancy due to the MLC system being located inside of the gantry head. The gantry head can sag up to 0.82mm and the MLC carriage can sag up to 0.99mm throughout the course of a treatment[71]. There is a dependence of the positional error of the leaves upon the gantry angle however it is clear that there are other factors in play from the dissimilar error distributions of bank A and bank B of Ling s Test 1 and sliding gap tests. Furthermore whilst there is a dependence upon the gantry angle the change in error is minuscule raising the question if it is clinically significant or not Errors and Treatment Machine Age The results from studying Picket Fence tests over the period of 18 months suggest that the mean RMSD of QA tests have a dependence upon the treatment machine s age. That is, deliveries administrated by older machines tend to have greater RMSD than when the machines were newer. This is to be expected. As machines get older they degrade and need maintenance. However the actual increase in mean RMSD for Picket Fence tests over an 18 month period is relatively tiny. The maximum difference between the mean RMSD for two different tests on the same machine is mm. The lifespan of a linear accelerator differs based on a whole range of factors including; the country, the hospital, the centre, and their available budget. The expected lifetime of a linear accelerator in Australia is 7-10 years which is based on the beam-on time and the mechanical performance of the system and linac output. With that in mind we may extrapolate to calculate the projected mean RMSD 90

111 difference of two treatments, one performed after commission and another at the end of its lifespan. Over the period of ten years the mean RMSD will increase by nearly 0.1mm. Therefore the machines will be replaced before these errors become significant. Of course this assumes the trend remains linear. The only way to ensure the machine is still in working order is to perform routine QA. This result simply shows us that the linear accelerator degrades with age albeit by a small amount. This agrees with the results of Chandraraj et al. who found minimal variation with insignificant dose differences in the same delivery performed daily over the course of one month [102]. 91

112 CHAPTER 5 CONCLUSION Varian log files were used to analyse multiple different treatments from a wide range of linear accelerators. A program was written using MATLAB to assess multiple factors of the MLC during each delivery. This study concluded that the positional errors of the MLC depends upon the treatment machine, treatment site, treatment modality, and the velocity of the MLC leaves. Greater error was found on treatments delivered on Clinac ix and Trilogy machines when compared to Truebeam machines. Additionally VMAT treatments targeting the prostate pelvic node treatment site had greater error than those targeting the head and neck treatment site which also had greater error than treatments targeting the prostate. The study also found that for the same machine, VMAT treatments had greater error than their dimrt and Step and Shoot counterparts and that deliveries with greater leaf velocity resulted in greater positional error. The study also concluded that for picket fence tests the inner leaves of the MLC had greater positional error than the outer leaves. This is most likely due to the difference between the individual leaf motors that inner and outer leaves possess but further experimentation is required to confirm. The study also concludes that the positional error of the leaves are dependant on the gantry angle. For Ling s Test 1 the positional error is greater at gantry angles where the leaf motion is parallel to the force of gravity. For sliding gap tests the error is greater at gantry angle 270 but at gantry angle 90 the error is less than when the leaf motion is perpendicular to the force of gravity. Additionally the positional errors of VMAT treatments were also affected by gravity which is evidenced in the different distributions based on the collimator angle. 92

113 This study also confirmed that the older treatment machines become the more they are prone to error. Additionally as a part of this study multiple treatment plans underwent gamma map analysis revealing that they were all delivered excellently. Analysing Varian log files for QA is quick and simple. As with any other QA tool, care must be taken to ensure that the measurements being recorded are correct and as such Varian log files do need to be checked to ensure they are reporting correct information. Additionally as Varian log files are a part of the linear accelerator system they will not pick up certain factors such as the sag of the gantry head or patient set up errors. Because of these points I believe that QA performed by Varian log files should be relegated to a supplementary role. Provided that the Varian log files are recording the correct position of the leaves, their high resolution make them useful for drawing attention to leaves that are more prone to error. Hypothetically the leaf motor could be changed before any clinically significant errors arise thereby sparing the patient from delivery errors. But perhaps where Varian log files are the most useful and take advantage of their unique properties is in analysing a failed treatment to try and find out how a treatment went wrong. The temporal component allows the clinician to view the treatment as it progressed and find out when it failed and also, what everything else in the machine was doing at that time. MountainGoat was created which analyses Varian log files for clinical use. Using Mountaingoat, or other similar programs, the RMS error of each leaf can easily be recorded for the daily QA and indeed every delivery from the linear accelerator. These can be stored and the error of each leaf can be kept in check. MountainGoat is available for free from: Future Work There are many questions left regarding the analysis of Varian log files. Ideally a standardised comparison should be made to compare the three different treatment modalities by using large samples sizes and treatments targeting the same area. 93

114 Further research could also delve into the different errors of the inner and outer leaves to determine whether this trend is MLC dependent or not. Additional work could also be undertaken in analysing the positional error of the leaves with respect to the leaf motion. Delivering sliding gap treatments on Truebeam machines should minimise the effect of the MLC system delay enabling a closer view of the MLC leaf velocities. Finally a lot more research can be undertaken in assessing the gravity effect on the MLC. By rotating the collimator angle by 180 we can determine if the errors are bank or gantry angle dependent. A relationship between the collimator angle and resulting error distribution could also be researched into. 94

115 References [1] N. C. Institute. (2014). Understanding cancer, [Online]. Available: cancer.gov. [2] A. B. of Statistics., Causes of death, australia, 2012, Journal, Online, [Online]. Available: [3] A. S. of Clinical Oncology. (2014). What is cancer surgery? [Online]. Available: [4] R. Dunleavey, Chemotherapy, in Cervical Cancer. Wiley-Blackwell, 2009, pp , isbn: [5] P. Rowshanfarzad, Improvement of epid-based techniques for dosimetry and invesetigation of linac mechanical performance in advanced radiotherapy, Research Higher Degree Thesis, University of Newcastle, [6] F. Khan, The Physics of Radiation Therapy, ser. M - Medicine Series. Lippincott Williams & Wilkins, 2010, isbn: [7] M. Teoh, C. H. Clark, K. Wood, S. Whitaker, and a. Nisbet, Volumetric modulated arc therapy: a review of current literature and clinical use in practice, British Journal of Radiology, vol. 84, no. 1007, pp , Nov. 2011, issn: [8] T. K.-M. Lam, Ed., A Comparison of IMRT and VMAT Technique for the Treatment of Rectal Cancer, presented at the American Association of Medical Dosimetrists 39th Annual Meeting, [9] E. B. Podgorsak, Radiation Oncology Physics: A Handbook for Teachers And Students. Intl Atomic Energy Agency (IAEA), 2005, isbn: [10] S. Webb. (2009). Volumetric-modulated arc therapy: its role in radiation therapy, [Online]. Available: [11] C. Rowbottom, Investigation into the pinnacle smartarc module for vmat planning, in World Congress on Medical Physics and Biomedical Engineering, O. Dossel, Ed.,

116 [12] A. S. Abbas, D. Moseley, Z. Kassam, S. M. Kim, and C. Cho, Volumetricmodulated arc therapy for the treatment of a large planning target volume in thoracic esophageal cancer, Journal of Applied Clinical Medical Physics, vol. 14, no. 3, pp , 2013, issn: [13] V. M. Systems. (2015). Vmat plan comparisons, [Online]. Available: [14] C.-L. Tsai, J.-K. Wu, H.-L. Chao, Y.-C. Tsai, and J. C.-H. Cheng, Treatment and dosimetric advantages between vmat, imrt, and helical tomotherapy in prostate cancer, Medical Dosimetry, vol. 36, no. 3, pp , 2011, issn: [15] D. Wolff, F. Stieler, G. Welzel, F. Lorenz, Y. Abo-Madyan, S. Mai, C. Herskind, M. Polednik, V. Steil, F. Wenz, and F. Lohr, Volumetric modulated arc therapy (vmat) vs. serial tomotherapy, step-and-shoot {imrt} and 3d-conformal {rt} for treatment of prostate cancer, Radiotherapy and Oncology, vol. 93, no. 2, pp , 2009, issn: [16] S.-H. Lu, J. C.-H. Cheng, S.-H. Kuo, J. J.-S. Lee, L.-H. Chen, J.-K. Wu, Y.-H. Chen, W.-Y. Chen, S.-Y. Wen, F.-C. Chong, C.-J. Wu, and C.-W. Wang, Volumetric modulated arc therapy for nasopharyngeal carcinoma: a dosimetric comparison with tomotherapy and step-and-shoot imrt, Radiotherapy and Oncology, vol. 104, no. 3, pp , 2012, issn: [17] T.-F. Lee, P.-J. Chao, H.-M. Ting, S.-H. Lo, Y.-W. Wang, C.-C. Tuan, F.-M. Fang, and T.-J. Su, Comparative analysis of smartarc-based dual arc volumetricmodulated arc radiotherapy (vmat) versus intensity-modulated radiotherapy (imrt) for nasopharyngeal carcinoma, Journal of Applied Clinical Medical Physics, vol. 12, no. 4, [18] D. Palma, E. Vollans, K. James, S. Nakano, V. Moiseenko, R. Shaffer, M. McKenzie, J. Morris, and K. Otto, Volumetric modulated arc therapy for delivery of prostate radiotherapy: comparison with intensity-modulated radiotherapy and three-dimensional conformal radiotherapy, International Journal of Radiation Oncology Biology Physics, vol. 72, no. 4, pp , 2008, issn:

117 [19] R. W. Kopp, M. Duff, F. Catalfamo, D. Shah, M. Rajecki, and K. Ahmad, Vmat vs. 7-field-imrt: assessing the dosimetric parameters of prostate cancer treatment with a 292-patient sample, Medical Dosimetry, vol. 36, no. 4, pp , 2011, issn: [20] M. Pasler, H. Wirtz, and J. Lutterbach, Impact of gantry rotation time on plan quality and dosimetric verification - volumetric modulated arc therapy (vmat) vs. intensity modulated radiotherapy (imrt), English, Strahlentherapie und Onkologie, vol. 187, no. 12, pp , 2011, issn: [21] R. Wiehle, S. Knippen, A.-L. Grosu, G. Bruggmoser, and N. Hodapp, Vmat and step-and-shoot imrt in head and neck cancer, English, Strahlentherapie und Onkologie, vol. 187, no. 12, pp , 2011, issn: [22] C. Sale and P. Moloney, Dose comparisons for conformal, imrt and vmat prostate plans, Journal of Medical Imaging and Radiation Oncology, vol. 55, no. 6, pp , 2011, issn: [23] E. E. Klein, J. Hanley, J. Bayouth, F.-F. Yin, W. Simon, S. Dresser, C. Serago, F. Aguirre, L. Ma, B. Arjomandy, and C. Liu, Task group 142 report: quality assurance of medical accelerators, AAPM, [24] D. a. Low, W. B. Harms, S. Mutic, and J. a. Purdy, A technique for the quantitative evaluation of dose distributions., Medical physics, vol. 25, no. June 1997, pp , 1998, issn: [25] P. Biggs, P. D, J. Galvin, D. Sc, E. Klein, and M. Sc, Basic Applications of Multileaf Collimators, , p. 56, isbn: [26] S. Takahashi, Conformation radiotherapy. rotation techniques as applied to radiography and radiotherapy of cancer, Acta radiologica: diagnosis, Suppl 242:1+, 1965, issn: [27] Y. Zhang, Y. Li, H. Xia, and J. Wang, Impact of dose rates on the position accuracy of multi-leaf collimator, Radiation Physics and Chemistry, vol. 81, no. 12, pp , Dec. 2012, issn: X. 97

118 [28] W. Que, J. Kung, and J. Dai, Tongue-and-groove effect in intensity modulated radiotherapy with static multileaf collimator fields., Physics in medicine and biology, vol. 49, no. 3, pp , Feb. 2004, issn: [29] S. Hariri and M. Shahriari, Suggesting a new design for multileaf collimator leaves based on monte carlo simulation of two commercial systems, Journal of Applied Clinical Medical Physics, vol. 11, no. 3, [30] S. Kamath, S. Sahni, J. Palta, S. Ranka, and J. Li, Optimal leaf sequencing with elimination of tongue-and-groove underdosage., Physics in medicine and biology, vol. 49, N7 N19, 2004, issn: [31] R. A. C. Siochi, Variable depth recursion algorithm for leaf sequencing., Medical physics, vol. 34, no. 2, pp , 2007, issn: [32] R. Alfredo C Siochi, Optimized removal of the tongue-and-groove underdose via constrained partial synchronization and variable depth recursion., Physics in medicine and biology, vol. 54, no. 5, pp , Mar. 2009, issn: [33] Millennium mlc system and maintenance guide, Varian Medical Systems, Tech. Rep., [34] K. H. V. Medical Physics at Institute of Radio Oncology. (2013). Dosimetric parameters of the hd120 mlc, [Online]. Available: [35] A. Mundt and J. Roeske, Intensity Modulated Radiation Therapy: A Clinical Perspective, ser. Intensity Modulated Radiation Therapy: A Clinical Perspective v. 1. BC Decker, 2005, isbn: [36] M. Jeraj and V. Robar, Multileaf collimator in radiotherapy, Radiology and Oncology, vol. 38, no. 3, pp , 2004, issn: [37] P. Rowshanfarzad, M. Sabet, D. O Connor, and P. Greer, Isocenter verification for linac-based stereotactic radiation therapy: review of principles and techniques, Journal of Applied Clinical Medical Physics, vol. 12, no. 4, 2011, issn:

119 [38] W. Du, S. Gao, X. Wang, and R. J. Kudchadker, Quantifying the gantry sag on linear accelerators and introducing an MLC-based compensation strategy, Medical Physics, vol. 39, no. 4, p. 2156, Apr. 2012, issn: [39] P. Rowshanfarzad, M. Sabet, D. J. O Connor, and P. B. Greer, Verification of the linac isocenter for stereotactic radiosurgery using cine-epid imaging and arc delivery, Medical Physics, vol. 38, no. 7, pp , [40] K. H. V. Medical Physics at Institute of Radio Oncology. (2013). Diameter of the radiation isocenter, [Online]. Available: [41] C. C. Ling, P. Zhang, Y. Archambault, J. Bocanek, G. Tang, and T. LoSasso, Commissioning and Quality Assurance of RapidArc Radiotherapy Delivery System, International Journal of Radiation Oncology Biology Physics, vol. 72, no. 2, pp , Oct. 2008, issn: [42] O. Holmes, J. Darko, T. Olding, and L. J. Schreiner, A potential modification of the gamma-evaluation: mapping dose disagreements using gamma-vector fields, Journal of Physics: Conference Series, vol. 444, p , 2013, issn: [43] S. Arumugam, A. Xing, C. Pagulayan, and L. Holloway, A comprehensive tool to analyse dynamic log files from an Elekta-Synergy accelerator, Journal of Physics: Conference Series, vol. 489, p , Mar. 2014, issn: [44] Truebeam, truebeam stx technical reference guide, Varian Medical Systems, Tech. Rep., [45] J. G. Li, J. F. Dempsey, L. Ding, C. Liu, and J. R. Palta, Validation of dynamic MLC-controller log files using a two-dimensional diode array., Medical physics, vol. 30, no. 5, pp , 2003, issn: [46] O. a. Zeidan, J. G. Li, M. Ranade, A. M. Stell, and J. F. Dempsey, Verification of step-and-shoot IMRT delivery using a fast video-based electronic portal imaging device., Medical physics, vol. 31, no. 3, pp , 2004, issn: [47] D. W. Litzenberg, J. M. Moran, and B. a. Fraass, Verification of dynamic and segmental IMRT delivery by dynamic log file analysis., Journal of applied clinical medical physics, vol. 3, no. 2, pp , Jan. 2002, issn:

120 [48] D. W. Litzenberg, S. W. Hadley, N. Tyagi, J. M. Balter, R. K. Ten Haken, and I. J. Chetty, Synchronized dynamic dose reconstruction., Medical physics, vol. 34, pp , 2007, issn: [49] S. Tae-Suk, L. Jeong-Woo, P. Jeong-Hoon, C. Jin-Beom, P. Ji-Yeon, C. Bo- Young, L. Doo-Hyun, H. Semie, K. Min-Young, and C. Kyoung-Sik, Inverse Verification of Dose Distribution for Intensity Modulated Radiation Therapy Patient-specific Quality Assurance Using dynamic MLC Log Files, Journal of the Korean Physical Society, vol. 55, no. 4, p. 1649, Oct. 2009, issn: [50] E. Schreibmann, A. Dhabaan, E. Elder, and T. Fox, Patient-specific quality assurance method for VMAT treatment delivery., Medical physics, vol. 36, no. 10, pp , 2009, issn: [51] J. Qian, L. Lee, W. Liu, K. Chu, E. Mok, G. Luxton, Q.-T. Le, and L. Xing, Dose reconstruction for volumetric modulated arc therapy (VMAT) using conebeam CT and dynamic log files., Physics in medicine and biology, vol. 55, no. 13, pp , Jul. 2010, issn: [52] B. Sun, D. Rangaraj, S. Boddu, M. Goddu, D. Yang, G. Palaniswaamy, S. Yaddanapudi, O. Wooten, and S. Mutic, Evaluation of the efficiency and effectiveness of independent dose calculation followed by machine log file analysis against conventional measurement based IMRT QA., Journal of applied clinical medical physics, vol. 13, no. 5, p. 3837, Jan. 2012, issn: [53] Dynalog file viewer reference guide, Varian Medical Systems, Tech. Rep., [54] M. M. Systems. (2014). Mobiusfx: overview, [Online]. Available: mobiusmed.com/mobiusfx. [55] IAEA, International Conference on Quality Assurance and New Techniques in Radiation Medicine, November [56] D. Rangaraj, M. Zhu, D. Yang, G. Palaniswaamy, S. Yaddanapudi, O. H. Wooten, S. Brame, and S. Mutic, Catching errors with patient-specific pretreatment machine log file analysis, Practical Radiation Oncology, vol. 3, no. 2, pp , Apr. 2013, issn:

121 [57] T. LoSasso, C. S. Chui, and C. C. Ling, Comprehensive quality assurance for the delivery of intensity modulated radiotherapy with a multileaf collimator used in the dynamic mode., Medical physics, vol. 28, no. 11, pp , 2001, issn: [58] C. Agnew, D. Irvine, and C. McGarry, Correlation of phantom-based and log file patient-specific QA with complexity scores for VMAT, Journal of Applied Clinical Medical Physics, vol. 15, no. 6, pp , [59] M. Dinesh Kumar, N. Thirumavalavan, D. Venugopal Krishna, and M. Babaiah, QA of intensity-modulated beams using dynamic MLC log files., Journal of medical physics / Association of Medical Physicists of India, vol. 31, no. 1, pp , 2006, issn: [60] A. K. Bhardwaj, T. Kehwar, S. Chakarvarti, A. Oinam, and S. Sharma, Dosimetric and qualitative analysis of kinetic properties of millennium 80 multileaf collimator system for dynamic intensity modulated radiotherapy treatments, Journal of cancer research and therapeutics, pp , [61] A. Manikandan, B. Sarkar, R. Holla, T. R. Vivek, and N. Sujatha, Quality assurance of dynamic parameters in volumetric modulated arc therapy, British Journal of Radiology, vol. 85, no. July, pp , 2012, issn: [62] C. E. Agnew, R. B. King, a. R. Hounsell, and C. K. McGarry, Implementation of phantom-less IMRT delivery verification using Varian DynaLog files and R/V output, Physics in Medicine and Biology, vol. 57, no. 21, pp , Nov. 2012, issn: [63] J. Peng, Z. Zhang, L. Zhou, J. Zhao, J. Wang, L. Kong, and W. Hu, A study on investigating the delivery parameter error effect on the variation of patient quality assurance during RapidArc treatment., Medical physics, vol. 40, no. 3, p , 2013, issn: [64] J. F. Calvo-Ortega, T. Teke, S. Moragues, M. Pozo, and J. Casals, A varian dynalog file-based procedure for patient dose-volume histogram-based IMRT QA, Journal of Applied Clinical Medical Physics, vol. 15, no. 2, pp , 2014, issn:

122 [65] N. Tyagi, K. Yang, and D. Yan, Comparing measurement-derived (3DVH) and machine log file-derived dose reconstruction methods for VMAT QA in patient geometries, Journal of Applied Clinical Medical Physics, vol. 15, no. 4, pp , 2014, issn: [66] A. Agnew, C. E. Agnew, M. W. D. Grattan, a. R. Hounsell, and C. K. Mc- Garry, Monitoring daily MLC positional errors using trajectory log files and EPID measurements for IMRT and VMAT deliveries., Physics in medicine and biology, vol. 59, no. 9, N49 63, May 2014, issn: [67] T. Teke, A. M. Bergman, W. Kwa, B. Gill, C. Duzenli, and I. A. Popescu, Monte Carlo based, patient-specific RapidArc QA using Linac log files., Medical physics, vol. 37, no. 1, pp , 2010, issn: [68] D. Rangaraj, S. Oddiraju, B. Sun, L. Santanam, D. Yang, S. Goddu, and L. Papiez, Fundamental properties of the delivery of volumetric modulated arc therapy (VMAT) to static patient anatomy., Medical physics, vol. 37, no. 2010, pp , 2010, issn: [69] C. E. Agnew, D. M. Irvine, a. R. Hounsell, and C. K. McGarry, Improvement in clinical step and shoot intensity modulated radiation therapy delivery accuracy on an integrated linear accelerator control system, Practical Radiation Oncology, vol. 4, no. 1, pp , 2014, issn: [70] P. Rowshanfarzad, M. Sabet, D. J. O Connor, and P. B. Greer, Investigation of the sag in linac secondary collimator and MLC carriage during arc deliveries, Physics in Medicine and Biology, vol. 57, no. 12, N209 N224, Jun. 2012, issn: [71] P. Rowshanfarzad, M. Sabet, D. J. O Connor, P. M. McCowan, B. M. C. Mc- Curdy, and P. B. Greer, Detection and correction for EPID and gantry sag during arc delivery using cine EPID imaging, Medical Physics, vol. 39, no. 2, p. 623, Feb. 2012, issn: [72] M. F. Clarke and G. J. Budgell, Use of an amorphous silicon EPID for measuring MLC calibration at varying gantry angle., Physics in medicine and biology, vol. 53, no. 2, pp , Jan. 2008, issn:

123 [73] A. R. Hounsell and T. J. Jordan, Quality control aspects of the Philips multileaf collimator., Radiotherapy and oncology, vol. 45, no. 3, pp , Dec. 1997, issn: [74] K. Wijesooriya, C. Bartee, J. V. Siebers, S. S. Vedam, and P. J. Keall, Determination of maximum leaf velocity and acceleration of a dynamic multileaf collimator: implications for 4D radiotherapy., Medical physics, vol. 32, no. 4, pp , 2005, issn: [75] L. Parent, J. Seco, P. M. Evans, D. R. Dance, and A. Fielding, Evaluation of two methods of predicting MLC leaf positions using EPID measurements., Medical physics, vol. 33, pp , 2006, issn: [76] I. Ali and S. Ahmad, Evaluation of the effects of sagging shifts on isocenter accuracy and image quality of cone-beam CT from kv on-board imagers., Journal of applied clinical medical physics, vol. 10, no. 3, p. 2930, Jan. 2009, issn: [77] C. R. Buckey, S. Stathakis, and N. Papanikolaou, The inter- and intrafraction reproducibilities of three common IMRT delivery techniques., Medical physics, vol. 37, no. 9, pp , 2010, issn: [78] D. S. Sharma, P. M. Dongre, V. Mhatre, and M. Heigrujam, Physical and dosimetric characteristic of high-definition multileaf collimator (HDMLC) for SRS and IMRT., Journal of applied clinical medical physics, vol. 12, no. 3, p. 3475, Jan. 2011, issn: [79] D.-h. Lee, J.-h. Park, and B.-y. Choe, Discrepancy of Intensity Modulation Radiation Therapy Dose Delivery due to the Dose-Dynamic Multi-Leaf Collimator Gravity Effect, Journal of the Korean Physical Society, vol. 53, pp , 2008, issn: [80] T. LoSasso, IMRT Delivery Performance With a Varian Multileaf Collimator, International Journal of Radiation Oncology Biology Physics, vol. 71, no. 1 Suppl, S85 8, Jan. 2008, issn:

124 [81] G. Heilemann, B. Poppe, and W. Laub, On the sensitivity of common gammaindex evaluation methods to MLC misalignments in Rapidarc quality assurance., Medical physics, vol. 40, no. 3, p , 2013, issn: [82] S. Sathiyan, M. Ravikumar, C. Varatharaj, S. S. Supe, and S. L. Keshava, IMRT Implementation and Patient Specific Dose Verification with Film and Ion Chamber Array Detectors., The gulf journal of oncology, vol. 1, no. 2, pp , Jan. 2010, issn: [83] A. Rangel and P. Dunscombe, Tolerances on MLC leaf position accuracy for IMRT delivery with a dynamic MLC., Medical physics, vol. 36, no. 7, pp , 2009, issn: [84] J. Lee, S. Hong, K. Choi, J. Chung, D. Lee, and T. Suh, SU-GG-T-163: Discrepancy of IMRT Dose Delivery Due to Dynamic MLC Gravity Effect, Medical Physics, vol. 35, no. 6, p. 2763, 2008, issn: [85] G. a. Ezzell and S. Chungbin, The overshoot phenomenon in step-and-shoot IMRT delivery., Journal of applied clinical medical physics, vol. 2, no. 3, pp , Jan. 2001, issn: [86] D. W. Litzenberg, J. M. Moran, and B. a. Fraass, Incorporation of realistic delivery limitations into dynamic MLC treatment delivery., Medical physics, vol. 29, no. 5, pp , 2002, issn: [87] A. M. Stell, J. G. Li, O. a. Zeidan, and J. F. Dempsey, An extensive logfile analysis of step-and-shoot intensity modulated radiation therapy segment delivery errors., Medical physics, vol. 31, no. 6, pp , 2004, issn: [88] T. J. Losasso and D. Ph, IMRT Delivery System QA, American Association of Physicists in Medicine, pp , [89] Truebeam trajectory log file specification, Varian Medical Systems, Tech. Rep.,

125 [90] P. Rowshanfarzad, C. K. McGarry, M. P. Barnes, M. Sabet, and M. a. Ebert, An EPID-based method for comprehensive verification of gantry, EPID and the MLC carriage positional accuracy in Varian linacs during arc treatments, Radiation Oncology, vol. 9, pp. 1 10, 2014, issn: X. [91] M. Hussein, P. Rowshanfarzad, M. a. Ebert, A. Nisbet, and C. H. Clark, A comparison of the gamma index analysis in various commercial IMRT/VMAT QA systems, Radiotherapy and Oncology, vol. 109, no. 3, pp , Dec. 2013, issn: [92] J. R. Kerns, N. Childress, and S. F. Kry, A multi-institution evaluation of MLC log files and performance in IMRT delivery, Radiation Oncology, vol. 9, no. 1, p. 176, 2014, issn: X. [93] G. T. Betzel, B. Y. Yi, Y. Niu, and C. X. Yu, Is RapidArc more susceptible to delivery uncertainties than dynamic IMRT? Medical Physics, vol. 39, p. 5882, 2012, issn: [94] G. Mu, E. Ludlum, and P. Xia, Impact of MLC leaf position errors on simple and complex IMRT plans for head and neck cancer., Physics in medicine and biology, vol. 53, pp , 2008, issn: [95] S. Bai, G. Li, M. Wang, Q. Jiang, Y. Zhang, and Y. Wei, Effect of MLC leaf position, collimator rotation angle, and gantry rotation angle errors on intensitymodulated radiotherapy plans for nasopharyngeal carcinoma, Medical Dosimetry, vol. 38, no. 2, pp , 2013, issn: [96] D. a. Low, J. W. Sohn, E. E. Klein, J. Markman, S. Mutic, and J. F. Dempsey, Characterization of a commercial multileaf collimator used for intensity modulated radiation therapy., Medical physics, vol. 28, no. 2001, pp , 2001, issn: [97] P. Rowshanfarzad, M. Sabet, M. P. Barnes, D. J. O Connor, and P. B. Greer, EPID-based verification of the MLC performance for dynamic IMRT and VMAT, Medical Physics, vol. 39, no. October, p. 6192, 2012, issn:

126 [98] K. Wijesooriya, E. Aliotta, S. Benedict, P. Read, T. Rich, and J. Larner, RapidArc patient specific mechanical delivery accuracy under extreme mechanical limits using linac log files, Medical Physics, vol. 39, no. 4, p. 1846, Apr. 2012, issn: [99] P. Zygmanski, J. H. Kung, S. B. Jiang, and L. Chin, Dependence of fluence errors in dynamic IMRT on leaf-positional errors varying with time and leaf number., Medical physics, vol. 30, no. 2003, pp , 2003, issn: [100] H. C. Woodruff, T. Fuangrod, P. Rowshanfarzad, B. M. C. McCurdy, and P. B. Greer, Gantry-angle resolved VMAT pretreatment verification using EPID image prediction., Medical physics, vol. 40, no. 2013, p , 2013, issn: [101] N. Childress, Q. Chen, and Y. Rong, IMRT QA using treatment log files is superior to conventional measurement-based method, Journal of Applied Clinical Medical Physics, vol. 16, no. 1, pp. 4 7, [102] V. Chandraraj, S. Stathakis, R. Manickam, C. Esquivel, S. S. Supe, and N. Papanikolaou, Consistency and reproducibility of the VMAT plan delivery using three independent validation methods., Journal of applied clinical medical physics, vol. 12, no. 1, p. 3373, Jan. 2011, issn:

127 Appendix APPENDIX A: Polar Plots Ling s Test 1 QA for True 2, True 3, ix 2, and Tri 2 was analysed to assess the effect of gravity on the positional error of the leaves. Data was only recorded when the leaves were stationary. The error of all the leaves for each treatment machine was averaged over multiple 2 degree control points. The results of, ix 2, Tri 2, True 2 and True 3 are presented graphically in figure 5.2, figure 5.1, figure 5.3, and figure 5.4 respectively. Error vs Gantry Angle Ling s Test 1 (Tri_2) a) b) c) Figure 5.1: Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 6 tests were analysed. 107

128 Error vs Gantry Angle Ling s Test 1 (ix_2) a) 0 0 b) c) Figure 5.2: Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 19 tests were analysed. Error vs Gantry Angle Ling s Test 1 (True_2) 0 0 a) b) c) Figure 5.3: Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 704 Picket Fence tests of True 2 were analysed. Figure 5.4: Polar plot of the error (mm), averaged over 2 degree control points, in the radial component vs the gantry angle (IEC convention) for a) Bank A, b) Bank B, and c) the error in the leaf gap width. 369 Picket Fence tests of True 3 were analysed. 108

129 APPENDIX B: Mountaingoat User Guide 109

130 A Guide to MountainGoat V1.02 Jeremy Hughes March 10,

131 I would like to acknowledge the financial support of the Western Australian Department of Health. This code was completed whilst undertaking a Masters of Medical Physics at the University of Western Australia as part of the Medical Physics Research Group at SCGH. For any questions pertaining to MountainGoat, (or if you want someone to talk to), my address is: jeremy.l.hughes@gmail.com 2

132 Contents 1 MountainGoat V Requirements for MountainGoat 4 3 Navigating around MountainGoat Loading in a DynaLog File Starting the Analysis Different Sections Information The DynaLog file name Date Delivered Number of Leaf Pairs used Times Flagged Gamma Map Analysis Graphs Error vs Gantry Error vs Time Error Histogram Fluence Actual Fluence Planned Fluence Diff Exporting Export Graphs Export Excel Known Issues 17 3

133 1 MountainGoat V1.02 MountainGoat was written in MatLab 2012b. MountainGoat analyses and assesses individual DynaLog files. After loading in a DynaLog file, MountainGoat performs the following functions: Assesses individual leaf error as a function of gantry angle Assesses individual leaf error as a function of time Produces a error histogram Produces a fluence map (planned, actual, and percentage difference) Takes note of how many times the leaves flagged (error > 1mm) Calculates Gamma Pass Rates for 1mm/1%, 2mm/2%, and 3mm/3%. Furthermore the graphs and an excel spreadsheet summarising this information can be exported. MountainGoat assumes the DynaLog files are from MLC with 60 leaf pairs. It also assumes the log file separations are 50ms. (Truebeam has 20ms). 2 Requirements for MountainGoat As MountainGoat was created in MatLab 2012b, MatLab Compiler Runtime V8.0 must be installed. It doesn t matter if you have a higher version of MCR installed, it needs V8.0 to run. This is annoying but something we must all live with. Luckily I have packaged MCR V8.0 in with the MountainGoat package and as such you should get a nice little pop up that asks you to install it. After you have installed it MountainGoat should be good to go. Give it a double click and wait a bit... it takes a bit to boot up. 3 Navigating around MountainGoat When you open up MountainGoat you should be greeted by the following window. 4

134 Note the majestic looking mountain goat. 3.1 Loading in a DynaLog File Click the browse button on either Bank A or Bank B to open up a navigation menu. Navigate to your DynaLog file and load it in. MountainGoat will automatically fill in the Bank A/Bank B counterpart. 5

135 3.2 Starting the Analysis Click on the giant square button that says GO! 6

136 MountainGoat should now look like this: 4 Different Sections 4.1 Information The Information section has the following: 7

137 DynaLog File Name Date Delivered Number of leaf pairs used The DynaLog file name It pulls the information from the DynaLog file Date Delivered Depending on the format of the Dynalog File Name the DATE may be incorrect. MountainGoat assumes the format AXXXXXX....dlg. If there is anything before the A, or B, in the DynaLog file name then MountainGoat will pull the incorrect characters and give the incorrect date of delivery. An example of an incorrect date is shown below Number of Leaf Pairs used This section tells you how many of the leaf pairs were used during delivery. If both of the leaves from Bank A and Bank B were stationary then they were not considered in the analysis (unless they had leaf pairs either side moving). For example: Leaf pairs 1-7 do not move through the whole treatment. Leaf pairs 8-12 move but leaf pair 13 doesn t. Leaf pairs then move but Leaf pairs don t. 8

138 Then MountainGoat will analyse Leaf Pairs 8-40 and as such 32 leaf pairs were used in the treatment. 4.2 Times Flagged This section contains two tables, one for Bank A and one for Bank B. It contains all the leaves that were used during the treatment and records the number of times they were flagged (error > 1mm). As it records each DynaLog step that had error>1mm the result is somewhat nonsensical (due to treatments having different lengths etc) but it gives a qualitative view of which leaves were erroring more (at a glance.) 4.3 Gamma Map Analysis This section details the Pass Rate for a Gamma Map analysis of expected and actual fluence maps for 1mm/1%, 2mm/2%,3mm/3%. 9

139 5 Graphs The following graphs are available. Error vs Gantry - Bank A, Bank B Error vs Gantry - Bank A, Bank B Error Histogram Fluence Actual Fluence Planned Fluence Diff 5.1 Error vs Gantry The default graph displayed is the Error vs Gantry of the first moving leaf in Bank A. The Gantry Angle is IEC convention. Different leaves can be viewed by using the slider, or typing in the leaf number in the box provided. 10

140 B. The graphs can be selected via the drop down menu or by clicking on either Bank A or Bank You can change the axis of the graph via the circled box below. Simply type in your number then hit enter. 11

141 This currently only works for Error vs Gantry with no plans to implement it for other graphs. 5.2 Error vs Time The graph can also be changed from Error vs Gantry to Error vs Time via the drop down menu. As above, the leaf can be changed via the slider or text box. It should be reiterated that MountainGoat assumes the delay between recorded points is 50ms. 12

142 If you are looking at Truebeam log files that have been converted to DynaLog format the delay is 20ms. This means that the scale down the bottom will be incorrect. 5.3 Error Histogram An error histogram can be produced by clicking on the Error Histogram button. 5.4 Fluence Actual The actual fluence map using actual leaf positions. Colour indicates intensity but only qualitative. It messes with the UWA logo colour scheme but I won t lose any sleep over that. 13

143 5.5 Fluence Planned The planned fluence map using planned leaf positions. Colour indicates intensity but only qualitative. 14

144 5.6 Fluence Diff The percentage difference between the Fluence Actual and Fluence Planned maps. 6 Exporting 15

145 6.1 Export Graphs Select a folder to export the graphs to. The default is dir/plots/[dynalog name]. Exports the Gantry vs Error graphs, Fluence Maps, and Error Histograms. It does not export Error vs Time graphs, that might come in at a later date. 6.2 Export Excel Exports an excel file much like the one below. It is colour coded. Any error that is green has error less than 0.5mm. Yellow is error between 0.5mm and 1mm and anything colour coded red is considered to have error greater than 1mm. It has two sheets. On the first sheet it contains the date and Dynalog name of the file. It then displays the Leaves used in the delivery for Bank A, and Bank B and says how many times they flagged (error>1mm). It then displays their maximum error and at what gantry angle this occurred. The second sheet displays the maximum error of each leaf for a 20degree arc for Bank A and Bank B 16

146 7 Known Issues BIGGEST ISSUE: Error Histogram will not have statistics displayed on the figure when you go to export it IF it is not being displayed in GUI. If you want to save the Error Histogram you have to be viewing it as you click Export Graphs. I have no idea why. I will look into it. When you view the Fluence Map the colormap changes ALL the axes, including the UWA logo. This makes everything look garish. 17