The University of Toledo The University of Toledo Digital Repository Theses and Dissertations 2010 Implementation of the Dosimetry Check software package in computing 3D patient exit dose through generation of a deconvolution kernel to be used for patients' IMRT treatment plan QA Brian James Bismack Medical University of Ohio Follow this and additional works at: http://utdr.utoledo.edu/theses-dissertations Recommended Citation Bismack, Brian James, "Implementation of the Dosimetry Check software package in computing 3D patient exit dose through generation of a deconvolution kernel to be used for patients' IMRT treatment plan QA" (2010). Theses and Dissertations. 793. http://utdr.utoledo.edu/theses-dissertations/793 This Thesis is brought to you for free and open access by The University of Toledo Digital Repository. It has been accepted for inclusion in Theses and Dissertations by an authorized administrator of The University of Toledo Digital Repository. For more information, please see the repository's About page.
A Thesis entitled Implementation of the Dosimetry Check Software Package in Computing 3D Patient Exit Dose Through Generation of a Deconvolution Kernel to be Used for Patients IMRT Treatment Plan QA by Brian J. Bismack Submitted to the Graduate Faculty as partial fulfillment of the requirements for the Master of Science in Biomedical Science degree in Radiation Medical Physics. E. Ishmael Parsai, PhD; Committee Chair David Pearson, PhD; Committee Member Michael Dennis, PhD ; Committee Member Patricia Komuniecki, Ph.D. Dean, College of Graduate Studies The University of Toledo December 2010
Copyright 2010, Brian J. Bismack This document is copyrighted material. Under copyright law, no parts of this document may be reproduced without expressed permission of the author. ii
An Abstract of Implementation of the Dosimetry Check Software Package in Computing 3D Patient Exit Dose Through Generation of a Deconvolution Kernel to be Used for Patients IMRT Treatment Plan QA. by Brian Bismack As partial fulfillment in the requirements for the Masters of Science of Biomedical Sciences Degree in Radiation Medical Physics The University of Toledo June 2010 Using the Dosimetry Check IMRT QA package a deconvolution kernel to be used in exit image dose calculations was created. This kernel modeled Electronic Portal Imaging Device response by incorporating the various machine characteristics along with the variant patient thickness and composition. To properly achieve this, Dosimetry Check first had to accurately model the beams for the machines being used. This was done by taking a series of in air and in water measurements including central axis depth dose values at various field sizes and in-air off center ratios to model beam flatness and symmetry. A deconvolution kernel for patient CT dose computation using pre-treatment (in-air) EPID images was created. This baseline helped establish the necessary iii
measurements for the exit image kernel. The necessary measurements for the exit image kernel included EPID images of various field sizes with various thicknesses of water in the beam as well as a characterization of off axis narrow beam transmission. The fit was performed and a report generated for its variance. The kernel was then successfully used in the evaluation of an IMRT prostate plan that was created for an anthropomorphic phantom and compared to a baseline evaluation of the same plan using in air EPID images. Volumetric, planar, and point dose comparison between measured and computed dose distributions agreed favorably indicating the validity of technique used for IMRT QA. iv
Dedication This would not have been possible without the support and opportunities afforded to me by my parents, friends, and the staff of the University of Toledo Radiation Oncology and the Physics and Dosimetry departments. Countless resources and endless time and effort have been expended in helping me to the position that I am in now and for that I am deeply appreciative. I have my parents to thank for inspiring me to find medical applications for physics. They have always encouraged applications for charity and the practical use of physics in medicine allows for that desire to be fulfilled. Lastly I would like to remind myself that the hard work, time and dedication really do affect people, never lose sight of that. v
Acknowledgments: E. Ishmael Parsai, PhD. Primary Advisor and Committee Chair Wendel D. Renner, President of Math Resolutions and contact for help David Pearson, PhD. Clinical Instructor and general motivator Nick Sperling, M.S. Staff Physicist and answerer of many questions Jerome and Theresa Bismack, Parents and life crisis rescue workers Jalpa Patel, Fellow Student and partner in crime on this project vi
Table of Contents Abstract Acknowledgments Contents List of Tables List of Figures List of Abbreviations 1. Introduction and Literature Survey 1.1. Various Methods of Quality Assurance.... 1.2. Useful Analytic Quantities..... 2. Theory 2.1. Pre-Treatment Image Deconvolution..... 2.2. Exit Image Deconvolution.. 3. Commissioning of Dosimetry Check For in Air Measurements 3.1. Beam Data.. 3.1.1. Issues with Beam Characteristics... 3.1.2. Equipment and Setup... 3.1.3. Data Processing iii vi vii x xi xiii 1 3 7 9 9 12 16 16 16 18 20 vii
3.2. Necessary Curve fits... 3.2.1. MU to Signal Calibration Curve.. 3.2.2. The Deconvolution Kernel... 3.2.3. CT Number to Density Curve.. 4. Experimental Setup for Exit Dose Measurements 4.1. Narrow Beam Transmission... 4.2. EPID Image Gathering... 5. Fit of the Exit Kernels and Results 5.1. Fit of the Narrow Beam Transmission and Exit Kernel. 5.1.1. Narrow Beam Transmission 5.1.2. Exit Kernel Fit.. 5.2. Results of Deconvolution of Exit Images.. 5.3. Discussion... References Appendices A Basic Design and Operation of an EPID B Welhoffer Data Excel Sheet C In Air OCR Text File D Central Axis Depth Dose Curve Data Edited for Dosimetry Check E Output Factor Text File F Text File of a Signal to MU Calibration Curve G In Air Deconvolution Kernel Text File H Organized Narrow Beam Data File 21 21 22 24 27 27 29 31 31 31 33 36 42 43 46 46 47 48 50 52 53 54 56 viii
I In Water Transmission Fit File for 10x J Exit Read File for the 10x Exit Kernel Fit K Fit Deconvolution Kernel File for the 10x Beam 57 59 66 ix
List of Tables 3.2.1 In air deconvolution kernel evaluation..... 3.2.2 Data for the CT number to density curve.... 4.1.1 Raw data for the 10x beam on the SL15..... 5.1.1 Results taken from the last section in the text file delineating the fit. The data is for 30 cm water depth.. 5.1.2 A sample of the results section from the Exit Kernel. This particular result is for a depth of 15.7 cm... 5.2.1 Point dose differences and GVH percentages. 5.2.2 A number of relevant region of interest average doses... 23 25 29 32 35 40 40 x
List of Figures 2.1.1 Shows a projection of pencil beams through an axial patient CT slice. Every other pencil beam is shaded.. 11 2.2.1 Field size response of deconvolution kernel with phantom in the beam (dotted) and without the phantom (solid), courtesy of Wendel Dean Renner President of Math Resolutions... 12 3.1.1 A picture of the Blue Phantom 2, very similar to the Blue Phantom that was employed for the annual QA and many other measurements delineated in this section. 3.1.2 In air OCR data taken on the SL15 at University of Toledo..... 18 19 3.2.1 An example fit of a signal to MU calibration curve. For an EPID the response should be linear and intersect at or near the origin.. 22 3.2.2 CT number to density curve employed by our Dosimetry Check program. The fit was performed by linear interpolation point to point... 26 4.1.1 A diagram of the setup used to measure the narrow beam transmission. The red dots are locations for the placement of the ion chamber... 28 xi
5.1.1 Graph of exponential fit by Dosimetry Check for the narrow beam transmission 32 5.1.2 Example screen plot of an evaluation of an initial guess for a deconvolution kernel... 5.2.1 Dose profile in the "X" direction through the calculation point. 5.2.2 Dose profile in the "Y" direction through the calculation point. 5.2.3 Dose profile in the "Z" direction through the calculation point.. 34 37 37 38 5.2.4 Overlay of two isodose lines from both the treatment planning system and from Dosimetry Check. 38 5.2.5 Gamma Volume Histogram of the reconstruction using pre-treatment images, the PTV is on the left while the body is on the right. 39 5.2.6 Gamma Volume histogram of the reconstruction using the exit images, the PTV is on the left and the body is on the right. 40 xii
List of Abbreviations IMRT...Intensity Modulated Ration Therapy QA Quality Assurance JACMP...Journal of Applied Medical Physics MU...Monitor Unit CT Computed Tomography TPS..Treatment Planning System LINAC...Linear Accelerator OCR...Off Center Ratio MLC...Multi-Leaf Collimator RMU...Relative Monitor Unit DTA.Distance to Agreement HU...Hounsfield Units SL15/SL25...Designation of LINACs at University of Toledo xiii
Chapter 1 Introduction and Literature Survey The most widely used form of radiation therapy today is external beam radiation therapy. Its versatility allows for the treatment in some form of almost any type of cancer. Of the approximately 1.4 million people who will be diagnosed with cancer this year in US alone, about 50% - 60% receive radiation therapy as part of their adjuvant therapy. This means that external beam therapy is widely used and employed for the treatment of over 700,000 people today. External beam radiation therapy went through a number of developments throughout its history. For a while it was done with a radioactive source known as cobalt-60. In this case the sources had to be replaced at regular intervals or the dose rate would fall so low that it would be impractical to use. In addition to the issue of dose rate, there was the problem of energy for deep-seated tumors, and penumbra because of the large size of the source. Then came Linear Accelerators (LINACS) that allowed for varying energies, two types of radiation (electrons and photons), and variable dose rates. Along with advances in radiation generation came advances in collimation. Rectangular fields gave way to 3D conformal radiation therapy by which cerrobend blocks were shaped to the tumor from a particular angle. This allowed for the improved sparing of normal tissue but was cumbersome in that blocks had to be customized for 1
each patient, were heavy, and very often could not be re-used. This resulted in them being cumbersome. The Multi-Leaf Collimation systems (MLC) were developed and became an integral part of modern linear accelerators by early 1900. These are a series of sheets of metal approximately 1cm wide (and thick enough to sufficiently attenuate a megavoltage beam) lined up together to spread across a field and attached to small motors that move them. This allows for quick shaping of the beam for each patient and encourages the use of more complex plans that focused the dose on the cancerous tissue. An advancement in the theory of dose delivery showed that beams of non-uniform fluence could be used to generate steep dose gradients in the patient that allowed for high dose conformality to the target and minimized dose to the surrounding healthy tissue. (Khan, 2003) This was called Intensity Modulated Radiation Therapy (IMRT). One way to do this is with brass compensators that are patient and beam specific. This might be likened to the IMRT equivalent of cerrobend blocks used for 3D conformal therapy. Another more popular way to modulate the intensity of the beam is with the MLC s. Further advances in MLC technology allowed for many shaped beams within beams (known as control points). (Khan, 2003) A plan of this sort would consist of 50-90 total control points from 5-9 different angles depending on the treatment. This new type of treatment revolutionized external beam radiation therapy. With the decrease in dose to the surrounding tissue came the ability to significantly increase the dose to the target. This aided in tumor response and patient quality of life. In some cases this allowed for the viable treatment of cancers that would otherwise not be feasible to treat properly. 2
Since its inception it has spread to many radiation therapy centers and has become the standard of treatment in one form or another for many different types of cancer. With this added precision and flexibility of treatment came new challenges. The plans being generated for IMRT could push LINACS to their operational limits in terms of Monitor Unit (MU) delivery, beam shape (the MLC s have finite resolution and motion), and pushed the boundaries of measurable and verifiable dosimetry. Daily, monthly, and annual tests for the MLCs on LINACS had to be developed to ensure their precision and accuracy for the treatment they would employ. (Khan, 2003) Additionally, the complexity of treatment plans called for the need to assure the deliverability of each individual plan. The plans needed to undergo a Quality Assurance (QA) process to ensure that what the computer calculated for delivery and what the machine actually delivered matched within a designated tolerance. This triggered the development of various techniques of ensuring the accurate delivery of a plan. (Khan, 2003) Patient specific QA is a field that continues to be explored as new means of broadening our tests to incorporate all aspects of treatment delivery are developed to ensure the best possible quality of patient care. 1.1 Various Methods of Quality Assurance QA of IMRT treatment plans has developed quite significantly over the past decade. One method developed early-on that is still in wide use today is film, sometimes called film and point dosimetry when used in conjunction with an ion chamber. If the film s characteristic optical density is linear with dose for the dose range and energy being used then exposed films can be scanned into a computer and directly compared to a 3
computer generated fluence using a software program designed for densitometry. (Khan, 2003) If the film is not absolutely reliable for the dose or energy being administered or an absolute calibration curve is not obtained then one may use an ion chamber to measure a point within the treated volume to compare with computed dose. However, it is a standard of practice to use some method of measuring planar fluence to a given depth for measuring the goodness and agreement of data computed vs. that being delivered. The idea is to take dose measurement of one point with a small ion chamber and use the exposed film to relate all other points in the plan to each other. In this method there are two components: the absolute component where a point dose is measured, and a relative component where all points evaluated against each other to ensure they have the correct exposure relative to each other with the film. These had some obvious disadvantages. First, in the case of using an ion chamber, was that the ion chamber had to be put in an area of low dose gradient or its size would prohibit an accurate measurement. Another important disadvantage was that this method only allowed for the absolute dosimetry of one point. Even if one has film with the means for an absolute measurement the film used for this type of QA needed to sit for 12-24 hours before being developed, so there is a delay that occurs before one knows if there is an issue with the plan. Another way to QA IMRT plans is to use matrixes. These typically include either ion chamber arrays or diode arrays. (Li, 2009) MapCheck is one example of a diode array that is in wide use. (Sun Nuclear Corp, 2010) The mounted arrays are calibrated to the treatment machine by the physicist. The goal is to compare a planar fluence for each beam in the plan generated by the treatment planning system to the planar fluence from 4
the machine as measured by the array. Typically there is a tolerance set that each individual measured beam must meat. This method is a step up above the film method in that it gives a result quickly. Additionally, it is easily calibrated for absolute dosimetry so some hassle from film may be eliminated. It does however suffer from a lack of resolution when compared to film, but this is typically strongly overshadowed by its strengths of ease of use and speed. However this method still lacks in that it only looks at a two dimensional fluence, not a three-dimensional dose distribution. Additionally, it looks at a QA plan where the beams are copied over to a plan meant to duplicate the conditions of the phantom for direct dosimetric measurement; hence the exact patient plan is not used but the leaf motions that is used in patient plan are not disturbed in QA plan. One way of evaluating the cumulative effects of the treatment in 3-D space is to use a 3-D phantom. (Khan, 2003) This phantom may employ a 3-D array of ion chambers or diodes or may use film. These phantoms produce the cumulative volumetric picture that had been lacking previously. Yet still there are issues, namely that this lacks direct patient information to relate to the machine output. In using these phantoms, the plan is projected onto a CT of the phantom to determine the expected doses or fluences. This plan though is no longer a plan on the patient. The information gathered is a comparison of machine output on the phantom to the expected machine output on the phantom. Information of how the machine output looks on the patient is still lacking. Furthering the exploration of 3-D cumulative effects are systems that measure the actual machine output of the plan and use it to reconstruct the dose to the patient CT. Two commonly used systems that do this are COMPASS and Dosimetry Check. (IBA 5
Dosimetry, 2009; Renner et al., 2005) COMPASS uses an array attachment to the gantry head to measure the output of the machine. It then uploads the data from each beam into its software which then projects the dose onto the patient s CT data set. Dosimetry Check does the same thing but uses the Electronic Portal Imaging Device (EPID, see appendix A for basic description of its design and operation) that is already attached to the accelerator, to measure the machine output. (Renner et al., 2005) This device was originally designed to take images of the patient for the purpose of verifying the setup of the patient prior to treatment but it has been shown to have applications in measuring machine output and 3-D dose reconstruction on patient CT data for the purposes of IMRT QA. (Steciw, 2005; Renner, 2003; Warkentin et al., 2003) EPID images are taken and then imported to the program, deconvolved to fluence maps, and the dose is reconstructed on the patient s CT to give a clinical view of the effects of the machine output (this will be discussed in more detail later). The use of the EPID in this instance offers the advantages of accessibility and resolution over then need to purchase an array attachment. Even these volumetric evaluations of the treatment plans lack the ability to check a few parameters that are very important to the delivery of IMRT plans. Two that immediately come to mind are the patient setup and the internal patient anatomy. Being able to do in vivo dosimetry with an EPID by using exit images (also known as transit images) would have advantages that would include verification of patient setup and various methods of some implementation of this have been reported by numerous authors. (Kroonwijk, 1998; McNutt et al., 1996; McNutt et al., 1996; Pasma et al., 1999; Broggi, 2002; Wendling et al., 2006; McDermott et al., 2007; Mans et al., 2010) There 6
have even been reports that in vivo dosimetry with an EPID can detect internal anatomical movements. (Kroonwijk, 1998) The project that this paper encompasses deals with the extension of the method used for the Dosimetry Check program to use of exit images to reconstruct the dose on a patient CT for the purpose of comprehensive patient IMRT QA. 1.2 Useful Analytic Quantities There are a number of analytic tools created to quantify the measured dose when evaluating an IMRT plan. These help physicists assess whether the QA for a patient specific IMRT plan passes or fails, the most basic of these analytical tools being a dose difference criterion. (Fraass et al., 1994; Fraass, 1996) In this method measured doses are simply compared directly to calculated doses with a set tolerance (a percentage of the calculated dose) for pass or fail. This was useful in low dose gradient regions but was lacking when applied to high dose gradient regions. In high dose gradient regions a quantity called Distance To Agreement (DTA) was used. (Harms et al., submitted to Med Phys; ICRU Report 42, 1987) This is the distance between a measured data point and the nearest point in the calculated dose distribution that exhibits the same dose. (Low et al., 1998) If there is no point that is within the specified distance then it is said to fail. A composite analysis of these two has also been developed. (Harms et al., 1994; Cheng, 1996) Each measured point is evaluated to determine if both the dose difference and DTA exceed the selected tolerances (e.g., 3% and 3mm, respectively), if either one of the criteria are not met then the point is said to fail. This incorporates the tolerances 7
sequentially or separately, that is, the tolerances have no bearing on each other. It also does not have a good way of specifying how well a point passes or fails. Another quantity was developed that is able to simultaneously incorporate the dose difference criteria and the distance to agreement criteria, this is the gamma value. (Low et al., 1998) This method sets up an ellipsoidal surface based on the desired acceptance criteria and sets that surface equal to a value of one (this is called the gamma value). If the surface made by the point measured intersects the calculated point then the point passes. One can further use the relation between the ellipsoidal surface and the point to determine the gamma value of the point which shows how well a point passes or fails. The passing criterion is a gamma value of 1 or below. Another advantage is that it is able to organize and display points in terms of an iso-gamma distribution. In this paper we employ the gamma value by using a Gamma Volume Histogram. That is a gamma assessment of the points in a particular region of interest organized into a histogram display. 8
Chapter 2 Theory 2.1 Pre-Treatment Image Deconvolution To start, it will be helpful to go through a brief discussion of the present process used for the reconstruction of dose with in air pre-treatment images. The basic process is the gathering of the images, the deconvolution of the images to fluence, and then the construction of the dose using a pencil beam algorithm. Typically in a treatment planning system a virtual source model is used to aid in computing beam intensity downstream from the collimators. To do this properly all objects involved in collimation must be taken into account; from the flattening filter to the MLCs (Multi-leaf Collimators) anything that is in the path of the beam must be taken into account to properly compute the dose no matter what computation method is used. Dosimetry Check does not use a virtual source model but rather a measured source model. (Renner, 2010) The EPID images that are taken include the effects of all of these objects, so instead of having a very good estimation of the objects for a particular field you have the direct measurement of the effects of the objects. For the EPID images to be useful we need to get to a quantity of fluence, in other words we need to convert EPID pixel values to a unit that will aid in computing dose. This is where the signal to RMU calibration curve comes in. This curve is based on EPID 9
measurements from the calibration field of the linear accelerator (typically 10x10) at various MUs. For the EPID image being used in calculating patient dose it maps each spot on an image to the same monitor units needed to produce the same signal along the central axis in a 10x10 field. This value is called the Relative Monitor Unit (RMU). (Renner, 2010) However this RMU value is not the end of the story, it needs to be further manipulated. Studies have shown that this RMU value is as much as 12.5% low for small fields like 2x2 cm and 6.3% high for large fields like 25x25cm. (Renner et al., 2005) Internal scatter (Renner, 2010) and dose deposition in the scintillator screen as well as optical photon spreading from the scintillator to the photodiodes must be accounted for. The former is achieved via a dose kernel and the later via a glare kernel. (Renner et al., 2005) The composite kernel that incorporates each of these can be represented by a sum of exponentials. The point spread function can be represented by the sum of exponentials. (Steciw et al., 2005)Thus a composite point spread kernel can be created of the form: n bir aie i ( = (1) k r) Where k(r) is the composite point spread kernel, r is the radius from the central axis with 100cm SAD and a and b are parameters for the exponentials. (Renner et al., 2005) Since this is circularly symmetric its frequency transform is also circularly symmetric and can take the form: n 2πbi K( q) = ai 2 2 2 3 / 2 (4π q + b ) i i (2) Eq. (2) is the Hankel transform of the 1D circularly symmetric point spread kernel with q in cycles/cm and a limitation on a and b to being greater than 0. (Renner et al., 10
2005) To achieve incident EPID fluence a deconvolution process is necessary. A 2D fast Fourrier transform of the EPID image is taken after converting it to RMUs then each frequency component is divided by the corresponding value in eq. (2) and transformed back. Each RMU value is then multiplied by its in air Off Center Ratio (OCR) to restore the flatness and symmetry of the beam in the converted RMU data set, Now we have arrived at the final RMU value for the fluence picture that Dosimetry Check will use. A pencil beam algorithm is used to project the dose on the patient CT (Figure 2.1.1) using the deconvolved image and a CT number to density calibration curve that must be based on data from each institution s particular CT machine. The weighting of each pencil beam is determined by each pixel s RMU value. (Renner et al., 2005) Thus the dose projected onto the CT is dependent on the output of the actual treatment fields of the machine. Figure 2.1.1 Shows a projection of pencil beams through an axial patient CT slice. Every other pencil beam is shaded. (Renner, 2010) 11
2.2 Exit Image Deconvolution The basic process for exit image reconstruction will be the same but there will need to be alterations to the deconvolution. One such characteristic that needs to be accounted for is EPID response to the scatter from a patient of variable thickness being in the beam. From the wedge measurements taken in Renner et al (Renner et al., 2005) we expect that the EPID will over respond to scatter radiation with an object in the beam. Preliminary measurements show this to be the case: Figure 2.2.1 Field size response of deconvolution kernel with phantom in the beam (dotted) and without the phantom (solid), courtesy of Wendel Dean Renner President of Math Resolutions. The graph is an assessment of the required correction of the deconvolution kernel with field size, normalized to the output of a 10x10 field. The smaller field sizes match more closely as they produce less phantom scatter. The EPID sees 12
predominantly the primary component of the beam. As the field size increases more phantom scatter is produced and it has been found that the EPID over-responds requiring the kernel to compensate. The deconvolution kernel then needs to characterize EPID response for variable patient thickness and field sizes. This can be done by creating separate kernels at specified depths of a water phantom in the beam and interpolating the results between depths. Another problem to be taken into account is that the transmission of the primary field changes as one moves radially outward from the central axis. This requires the characterization of narrow beams as they move off axis along the diagonal at regular intervals. This will be discussed in further detail later. It is not a serious problem and the process can be performed without this measurement but this refinement does increase the accuracy of the off axis dose reconstruction. Another issue that needs to be addressed is a serious one. The kernel now being employed is a function of radius as indexed by thickness. (Renner et al., submitted to Med Phys, 2010) Interpolation between thicknesses allows for a continuous function of thicknesses. (Renner et al., submitted to Med Phys, 2010) Each pixel in an EPID image is assigned a thickness via a ray trace through the patient s CT data set. The thickness traversed for each pixel is used to reference the point spread kernel. (Renner et al., submitted to Med Phys, 2010) This variant kernel depends then upon the position relative to the image which means the convolution theorem will not apply and the deconvolution cannot be performed by simple multiplication in the frequency domain. (Renner et al., submitted to Med Phys, 2010) The fast Fourrier transform cannot be used which in turn seriously inhibits computation time. (Renner et al., submitted to Med Phys, 2010) One 13
way to alleviate this is to reduce the resolution of the EPID images to 1 mm. This is sufficiently smaller than the pencil sizes of 2 to 5 mm and still retains the high resolution advantage of the EPID, but the image sizes are still of the order of 400 x 300. (Renner et al., submitted to Med Phys, 2010) This discrete deconvolution will still take a presently impractical amount of time to transform, about 1 hour for a 7 field IMRT case. (Renner et al., submitted to Med Phys, 2010) However, an approximation for a discrete decovolution can be done using multiple fast Fourrier transforms of the particular EPID image in question. Since there was a direct kernel made for discreet water thicknesses we can deconvolve the image once for each thickness yielding numerous deconvolved images. To obtain the proper deconvolved pixel, each pixel thickness traversed for an image is identified and an interpolation is performed between the two deconvoled images adjacent to that thickness. (Renner et al., submitted to Med Phys, 2010) A pixel of thickness 42 cm would reference the deconvolved images from thicknesses 40 and 45 cm, for example, and interpolate between the two. (Renner et al., submitted to Med Phys, 2010) On one computer this reduced the computation time of a single in air fluence image from 9 minutes (with the discrete deconvolution) to 1.7 seconds (with the interpolated fast Fourrier deconvolution). One final issue will be left for further study. The present process assumes constant scatter from the phantom across the EPID. (Renner et al., submitted to Med Phys, 2010) Fortunately the discrepancy is only of significance for large field sizes so it should not seriously affect most cases. To achieve the use of this we must first commission Dosimetry Check for use with in air images. This will require some machine specific beam measurements as well as the 14
fit of various other components such as a CT number to density curve. After this is done two primary measurements (alluded to previously) must be made for the proper characterization of the Exit Kernel. One involves the EPID images for various field sizes and water depths and the other is off axis narrow beam transmissions through various water depths. 15
Chapter 3 Commissioning Dosimetry Check for in Air Measurements The Dosimetry Check program requires various measurements and data to install and use properly. The amount will vary based on the setup of the clinic. To commission the program a deconvolution kernel, an MU to signal calibration curve, and a CT # to density conversion were all fit. The process of fitting a kernel, calibration curve and density curve will be described below. For the theory on how the kernel works see the previous chapter. Additionally while some LINACs may be able to have their beam characteristics sufficiently matched to a predetermined estimate of various LINACs by each manufacturer, the LINACs used at the University of Toledo are unique and therefore require data such as depth dose curves and off center ratios to be input into the program. The process of gathering and inputting all of this data will be described below. 3.1 Beam Data 3.1.1 Issues with Beam Characteristics There are two Elekta SL Precise series LINACs at the University of Toledo; one designated the SL15 and the other the SL25. The SL15 has two photon energies, 6MV and 10MV. The SL25 has three photon energies, 6MV, 10MV, and 18MV. The two energies used for IMRT are 6MV and 10MV. These LINACs were setup up upon their 16
installation to be sufficiently alike that patients could be swapped between machines with relative ease. This required a few necessary changes to the original specifications of the machines. Originally the SL15 had photon energies of 6MV and 15MV while the SL25 had energies of 6MV, 10MV, and 25MV. The energies on the SL15 were adjusted to 6MV, and 10 MV to match the two lower energies of the SL25. Though irrelevant to the study, the SL25 maximum energy was also lowered to 18MV. Because the SL15 was originally intended for 15MV the depth dose profiles and flatness of each machine had to be tweaked at 10MV in order to match the machines. It was decided that the depth dose profiles should be ensured to match as closely as possible. The end result was that the energy of the 10MV beam of each machine was in fact slightly less than 10MV and that neither machine had a flat 10MV beam. However, the characteristics of the machines matched well enough to allow patients to be swapped in the event of necessary maintenance of one machine. This posed a slight problem though. Dosimetry Check had to be made to account for this change in energy and non-flatness. In fact the program does have the ability to account for non-flat beams. It is not able to characterize non-symmetric beams but those are rarely, if ever, in use. To account for non-flatness Dosimetry Check eliminates the non-flatness it sees in the EPID images and then multiplies by the In Air Off Center Ratios (OCRs) to restore the non-flatness of the beam. To account for the energy of the beams Dosimetry Check uses depth dose profiles at varying field sizes. This gives it the characteristics of the beam profiles it needs to do its IMRT calculations. 17
3.1.2 Equipment and Setup Depth Dose Profiles These were easily obtained from the annual machine QA data. The measurements required a Welhoffer Scanning Water Phantom with the necessary robotics and programs and the Welhoffer ion chamber. Depth dose profiles for field sizes of 5x5, 10x10 and 20x20 cm were obtained during the annual QA calibrations. These were the ones used for input into Dosimetry Check. Figure 3.1.1 A picture of the Blue Phantom 2, very similar to the Blue Phantom that was employed for the annual QA and many other measurements delineated in this section. In Air Off Center Ratios Obtaining the OCRs also required the use of the Welhoffer Scanning Water Phantom. We employed a 0.6cc Farmer chamber, a CNMC Electrometer, and buildup caps appropriate to the energy of the beam being measured. 18
Figure 3.1.2 In air OCR data taken on the SL15 at the University of Toledo. Note that Figure 3.1.2 is not a direct measurement of the flatness of the beams as flatness is defined at 10 cm water depth for photon beams. To measure the ratios properly the tank had to be carefully leveled so the robotics would move evenly throughout the beam profile being taken. Additionally the tank had to be rotated 45 degrees. This was because the profile being taken in this instance was along the diagonal of the square field in order to travel out the farthest distance possible. This would allow for characterization of the beam at large distances from the central axis should they be needed for treatment. The orientation of the Welhoffer had to be noted to maintain consistency in processing the data. The Welhoffer was also set to a reasonable resolution as the highest resolution would have yielded far more data points than are necessary. 19
3.1.3 Data Processing With the Welhoffer program there was much more data in the files than needed for the Dosimetry Check program. Also, the data had to be put in a readable format for Dosimetry Check. This required putting the data to be entered in text documents with a particular layout. The data had to be accessed from the Welhoffer scanning files and sorted using Excel. It is important to note the orientation of the Welhoffer coordinate system. In the end, Excel documents in the format seen in Appendix B are most useful for creating the text documents. The text documents follow a layout and format that is detailed on the Math Resolutions main website. We reviewed the previously created documents carefully to ensure consistency throughout each document. See Appendix C for an example of an in air OCR text file. Depth dose curves (Appendix D) and output factors (Appendix E) also had to be put into these text documents and labeled properly. Along with these measured data files there are a number of other files that were necessary to edit. There is a file that labels dmax of the beam in question and this must be entered properly. There is a file that delineates the dose rate in cgy/mu at the calibration depth and calibration field size of the machine. There is also a list file that must be edited properly to instruct the program as to how many depth dose curves there are and the names of the files that contain them. All files were checked for consistency in calibration depths and measured depths. Additionally, the program must generate some beam characteristics from the data that is entered. For example, from the output factors (measured at a specified depth) it will generate a file that will tell you the Sc and Sp factor separate from Scp. This is done 20
by the Generate Beam Data utility within the program. If there are any errors in the documents this will be brought to your attention later in a command prompt when this process is begun. 3.2 Necessary Curve Fits 3.2.1 MU to Signal Calibration Curve As discussed earlier, each beam needed an MU to signal calibration curve. This was achieved by taking a number of EPID images with the calibration field size (typically 10x10) of the machine at different MU s. We used a 10x10 field with MU s of 5, 15, 25, 50, 75, 100, and 200 MU s. The images then had to be imported to the Dosimetry Check program. In the Convert Images utility all of the images were selected and then an option to create a calibration curve was selected. Each point was labeled and associated with the right image file. After the points and files were input fit was performed and the curve was named. See Appendix F for and example text file of a signal to MU calibration curve. 21
Figure 3.2.1 An example fit of a signal to MU calibration curve. For an EPID the response should be linear and intersect at or near the origin. 3.2.2 The Deconvolution Kernel As noted before, this is one of the most important parts of Dosimetry Check. To fit the kernel EPID images were taken at a set MU for various field sizes corresponding to the field sizes whose output factors were input into Dosimetry Check. To keep track of those input field sizes the files Output6 or Output10 text file in the Beam Data directory were used as they contained the data. All of the EPID images were shot under a known directory in the Iview program. The easiest way was to set up a single patient in IMPAC with the necessary fields pre-loaded. A 10x10 centering and calibration field also had to be taken. After the images were collected they had to be imported into Dosimetry Check. Under the Convert Images utility EPID images were converted initially without using a deconvolution kernel, only an MU to signal calibration curve was used. Then the deconvolution kernel could be fit (note that this was an older way to fit the kernel, a newer way is described later in the exit dose section). All of the output points 22
(corresponding to the field size) and the RMU files had to be associated, first an output point was selected, then the corresponding RMU file was associated to the point right afterwards. For each point the program had to be told how many MU s were shot in that particular EPID image (it is important that the MUs are consistent). An initial guess then had to be read into the program (a basic one is provided for you) to provide a decent starting point for the fit, then the fit was performed. The goal of the kernel is to match the calculated central axis reading of the EPID images to the measured central axis reading. After fitting the data a report on the goodness of the fit will be shown. This is actually the text from a text file that details the fit. A full text file of an in air deconvolution kernel can be seen in Appendix G. In evaluating the fit of the kernel the primary section to look at is the last section in the text file where it delineates in water values (see Table 3.2.1 for quick reference). The last column will give a percent error between the measured cgy and the computer cgy from the deconvolution kernel. The goal is that all of the percentages are below 1% but having one or two from 2-3% is acceptable. Table 3.2.1 In air deconvolution kernel evaluation. Field Size Depth Measured Computed % Diff 2x2 2.1 21.75 22.03 1.30% 3x3 2.1 22.8 22.92 0.56% 4x4 2.1 23.3 23.36 0.27% 5x5 2.1 23.66 23.52-0.59% 6x6 2.1 24.07 23.89-0.71% 8x8 2.1 24.53 24.47-0.27% 10x10 2.1 25 24.92-0.32% 12x12 2.1 25.35 25.33-0.10% 14x14 2.1 25.7 25.7 0.02% 16x16 2.1 26.01 26.07 0.21% 18x18 2.1 26.27 26.4 0.50% 20x20 2.1 26.47 26.6 0.51% 25x25 2.1 26.74 26.6-0.54% 23
The example in Table 3.2.1, a kernel used for the SL25, is an acceptable kernel. One thing to note, just because a fit has low % differences does not automatically mean it is a useable fit. The low percentages will not necessarily indicate if there are issues with various measurements; for example, entering an incorrect depth at which the output factors were taken. However, if there is a significant deviation in a particular field size or range of field sizes then it may strongly indicate an issue with those measurements. For example, a 2x2 field size that has a large (say 13%) deviation may indicate an incorrectly measure output factor. 3.2.3 CT Number to Density Curve CT numbers are a scaled attenuation value that is a function of physical density and effective atomic number of the material. In the biological range of densities and materials it is easy to relate electron density to physical density and thus we can directly relate CT number to physical density. From this approximation we can set up a CT number to density curve that allows us to identify the density at any point in the body. Historically the unit of CT number has been in Hounsfield Units (HU) where water is set to a CT number of 0 with a minimum value of -1000 (indicating vacuum or air) and a maximum depending on the particular CT machine; typically 3096. However, we use a system that shifts water back to 1000 and leaves air at 0 so the maximum number would be 4096. This is done partially for ease of clipping data that would indicate a density of less than 0. To properly reconstruct dose on a CT a curve needed to be fit. To do this a phantom with various materials of known density was scanned with the CT scanner being 24
used. The image was then sent to the treatment planning system for statistical analysis. This gave points of CT number to density that allowed a curve to be fit. Below is a table of CT number to density data used for our curve. The data is a compilation of old data from the treatment planning system, filled in with new data from a scanned phantom. Table 3.2.2 Data for the CT number to density curve. Signal 0 46 500 946 1020 1122 1144 1903 6000 Density 0 0* 0.500 0.950 1.000 1.19* 1.200 2.00* 6.000 The numbers in Table 3.2.2 with an asterisk come from the scanned water phantom and were used to help the Dosimetry Check curve reflect the results from Pinnacle. There are a number of ways to actually fit the data; linear or higher order polynomial curves may be used but they suffer when a single large point that is outside the biological range is used to fill out the upper end of the curve. Instead, a point-to-point fit was employed where by the system in effect drew a straight line between each subsequent point. (Renner, 2010) This helped keep the one large point from throwing off the fit at the lower more used range and helped keep the data more reflective of the actual results from the scan. 25
Figure 3.2.2 CT number to density curve employed by our Dosimetry Check program. The fit was performed by linear interpolation point to point. 26
Chapter 4 Experimental Setup for Exit Dose Measurements After the in air pre-treatment process is up and running we can move onto taking measurements for Exit Dose Reconstruction. The goal of these measurements is to provide the necessary data for characterizing the EPID s response to images taken with an object in the beam. The two primary measurements, alluded to earlier, are the EPID measurements, which are similar to the EPID measurements required for the in air kernel, and the narrow beam transmission measurements, which help characterize the primary beam transmission off axis through a phantom. 4.1 Narrow Beam Transmission The narrow beam transmission measurement required an ion chamber, an electrometer, a buildup cap with suitable buildup for both energies being measured, a phantom, a treatment plan with a series of 2x2 square fields, and some tape to help mark the positions for the ion chamber on the ground. To get an idea of the setup for further discussion a diagram is shown below: 27
Figure 4.1.1 A diagram of the setup used to measure the narrow beam transmission. The red dots are locations for the placement of the ion chamber. The treatment plan was rather tricky; it required spacing for the jaw and leaf movements to allow for a 2x2 field that could be moved in increments along the diagonal from the central axis. Ideally, increments of 2 cm would be used but it made the planning much simpler to use increments of about 2.8 cm. The plan then had to be loaded into IMPAC so the fields could easily be loaded into the LINACS. After it was loaded into the LINACS it was discovered that the light from the field was not bright enough to reach the ground through the water phantom. Also, the refraction of light through the phantom at large water depths would make the field light on the ground an unreliable locator for the ion chamber. Using masking tape and a red permanent marker we loaded the fields individually on the machine and marked the ground where the light beam landed. The ion chamber was placed on the corresponding dot for each field being shot. The MU for each field had to be consistent and had to be large enough to register a reading for the large water depths. Knowing the specific MU value however was not 28
critical as the measurements were going to be used in a manner relative to the central axis. Table 4.1.1 Raw data for the 10x beam on the SL15. Depth 0 2.83 5.66 8.49 11.31 14.14 16.97 18.39 0 0.354 0.372 0.3835 0.393 0.404 0.4115 0.4165 0.4045 5 0.2885 0.3015 0.312 0.315 0.3215 0.328 0.3285 0.3135 10 0.235 0.245 0.257 0.2575 0.258 0.263 0.262 0.2505 15 0.195 0.205 0.211 0.211 0.215 0.214 0.212 0.207 20 0.1625 0.1695 0.173 0.173 0.176 0.175 0.172 0.1675 25 0.1345 0.1385 0.1455 0.145 0.138 0.145 0.1395 0.1345 30 0.113 0.117 0.121 0.1205 0.119 0.117 0.1155 0.11 35 0.0955 0.0985 0.1005 0.0995 0.0965 0.1 0.097 0.089 39 0.083 0.087 0.088 0.0875 0.0845 0.085 0.081 0.0765 The top row in Table 4.1.1 delineates the radius off axis while the first column is the depth of water. The measurements are in nc. Eight measurements were taken for each water depth and for each energy; the water depths ranged from 0 to about 40 cm at 5 cm increments. In all, 144 ion chamber measurements were made. The ion chamber had to be moved every other measurement and the water depth changed every 16 measurements. Not counting the preparation and setup time, these measurements alone took over 8 hours. 4.2 EPID Image Gathering The field sizes had to be selected and the MU selected prior to shooting to remain consistent throughout the measurements. Again, the phantom was placed on the table. Since the kernel we are reporting was measured on the SL25, which employs a non-radio transparent carbon fiber table, a water equivalent depth of the table had to be obtained. This was done by relating the table correction factor used in patient treatment planning to TMR values taken for the machine at the specified energy. The water equivalent depth was estimated to be around 0.7 cm. Later, for the kernel fit, this was added to each water 29
depth with the exception of 0 cm depth as the table was moved out of the way. In all 54 EPID images had to be taken. While shooting the fields, after each depth was completed the patient in the Iview program (which was reflective of the patient name in IMPAC) was renamed to correspond with that depth. This prevented confusion when the next set of EPID images was shot. Later the EPID images were converted in batches according to depth to RMU s using the calibration curve only (not deconvolved as the kernel had not been fit yet). Each individual RMU files was labeled separately for each depth and all of them placed in the same file to be read into the kernel fit program. A bulk rename utility was used to do this. The middle section in Appendix J (the kernel read file) depicts the organization of the files. It shows the number of files organized. Each section is divided by water depth; the first column in each section shows the file name while the second column shows the MU (which is necessary as the program needs to be told what the MU was for each file). 30
Chapter 5 Fit of the Exit Kernel and Results 5.1 Fit of the Narrow Beam Transmission and Exit Kernel 5.1.1 Narrow Beam Transmission The data for the narrow beam transmission was organized properly into a text file of a particular format. The Fit Transmission Data program in Dosimetry Check reads the file of organized data and does an exponential type fit to the data at each off axis radius through each water thickness. The fit interpolates between each water thickness. See Appendix H for an example text file of narrow beam data to be read into the Dosimetry Check program prior to performing the narrow beam transmission fit. After the text file is created the fit can be performed. The fit employs four exponentials for each radius to characterize the transmission through various thicknesses at that radius. The program is fitting the coefficient and the exponent of the exponentials in question. The Dosimetry Check program will create another text file with the results of the fit to be used in further calculations (Appendix I). It will also give you the option of looking at a graph of the exponentials that it fit. 31
Figure 5.1.1 Graph of exponential fit by Dosimetry Check for the narrow beam transmission. Table 5.1.1 Results taken from the last section in the text file delineating the fit. The data is for 30 cm water depth. Radius at 100cm Attenuation % Difference 0 0.3208 0 2.8 0.3172-1.13 5.7 0.316-1.49 8.5 0.3075-4.14 11.3 0.2931-8.64 14.1 0.2927-8.75 17 0.2821-12.07 18.4 0.2763-13.87 This fit file data was checked. This was done by observing the results at 30 cm thickness of water which is given in the last section of the text fit file. It was determined that the fit yielded numbers that were similar to fits of the same type and that there was nothing apparently wrong. 32
This particular data was gathered originally from the SL15 but was used in conjunction with a kernel from the SL25 for image being taken on the SL25. Though the non-flatness of the SL15 might be a source of error, and the deconvolution process does not technically require this data, it was estimated that the machines were similar enough that it would be helpful to include this fit in the deconvolution process. 5.1.2 Exit Dose Kernel Fit To perform an exit fit the computer must be given certain parameters. It must be told what fields sizes were used, the number of thicknesses, which EPID images were taken at which thickness, limits for the fit variables, and an initial kernel guess by which it can base its search for the fit. See for Appendix J for the exit kernel read file that was used. It incorporates all of the necessary parameters for the program and directs how to find the data, organize it, and provide it with a means of initiating the fit. The completed fit file is then read by the program which reads an initial guess. You can observe various results of the initial guess in a viewing window to ensure that nothing is seriously out of order. One can also use previously fit kernels as initial guesses. It may be possible to yield a better fit for the kernel by fitting it once then using that first fit as the initial guess for the subsequent fit. One can observe the graphs of the first fit of the kernel by reading it as an initial guess and viewing that guess. 33
Figure 5.1.2 Example screen plot of an evaluation of an initial guess for a deconvolution kernel. Each field size in Figure 5.1.4 has a pair of windows, one with the picture and one with plots of various data. The straight solid line at the top is the expected signal, the dotted lines represent the x and y deconvolved signal and the solid line represents the signal through the profile of the EPID image. This particular set of plots was for an equivalent water thickness of 20.7 cm. Figure 5.1.4 also helps show what happens before and after the deconvolution. The straight line at the top is the Sc factor multiplied by the in air off center ratio. In the case of our Elekta machine, the OCR is multiplied in later so the non-flatness of the beam is not seen here. The dotted lines are the deconvolved signal and should be equal with the 34
top line at the central axis (if the top line reflected the beam non-flatness it should be roughly parallel with it as well). (Renner, 2010) It is important to note an off axis discrepancy in certain situations. For large field sizes and water depths the deconvolved image and the actual fluence will disgree. This is due to the approximation that scatter from the phantom is uniform across the field when in fact it is not. (Renner, 2010) However the majority of field sizes used in IMRT are small enough that this approximation works. Further characterization of the phantom scatter may be addressed some time in the future. After reviewing this, the kernel may be computed; this takes about a day on a computer with a dual core processor. In effect the computer generates an individual kernel for each thickness at which the EPID images were taken and interpolates (in the method delineated in the chapter on theory) for water thicknesses that are between the thicknesses measured. Again the fit for the kernel is stored in a text file which the program can read for deconvolution of future images. The text file also contains some helpful analysis of the goodness of the fit of the kernel. It should be noted that these analyses will not necessarily indicate an error. The program fits the data that you tell it to fit and if that data is wrong it might not show up here. However the program should be able to fit the data reasonably well. If there is a significant problem with a particular field size it may indicate that an output measurement, for example, may be significantly off. Table 5.1.2 A sample of the results section from the Exit Kernel. This particular result is for a depth of 15.7 cm. Field Size cm Raw c.a. Signal After c.a. Sc % diff Ratio 3x3 12.0894 0.9668 0.9530 1.44 0.0800 5x5 12.6470 0.9644 0.9740-0.99 0.0763 10x10 14.0074 0.9907 1.0000-0.93 0.0707 15x15 15.2128 1.0179 1.0180-0.01 0.0669 20x20 16.3030 1.0317 1.0280 0.36 0.0633 25x25 17.2627 1.0292 1.0300-0.08 0.0596 35
See Appendix K for the full fit file of the deconvolution kernel. Like the in air kernel results we are looking for a % difference less than 1%. Ideally every field size would be off by no more than 1% and apart from a few instances this kernel meets that ideal. Additionally, there are no differences exceeding 3%. This would indicate that if our output factors and images are reliable then the kernel is a usable kernel. 5.2 Results of Deconvolution of Exit Images After the images are gathered they are put through the deconvolution process in a similar manner to the pre-treatment in air images. In the case of Exit images however we must ray trace through the patient s CT in order to identify the distance of the patient each pixel of the EPID image had traversed. The idea is that if there is significantly improper delivery of the treatment it should be shown in the dose reconstruction from the deconvolved EPID images due to the alteration in patient thickness and composition through which the beam passed. If the treatment was delivered in a consistent fashion then it should be reasonably close to the pre-treatment results and if it was delivered as accurately and consistently then it should be reasonably close to the TPS plan computation. One goal of this type of QA is to ensure that there is no misadministration of dose occurs. A definition of a misadministration of dose would be a 10% variance in the prescribed dose to the region. As a check on this one may look at some point doses (in Table 6.1) as well as a gamma volume histogram to ensure that the plan is being delivered as expected. It would be ideal then to test the Exit Kernel on a plan that is reasonably confidently deliverable (using exit images) and see how well it compares to the reconstruction of dose using in air images. The idea is that the in air reconstruction 36
will be a good base of reference for what ever measurement the exit reconstruction gives. Below are figures detailing the output of the reconstruction with the in air images next to the reconstruction with the exit images for comparison. Figure 5.2.1 Dose profile in the "X" direction through the calculation point. Figure 5.2.2 Dose profile in the "Y" direction through the calculation point. 37
Figure 5.2.3 Dose profile in the "Z" direction through the calculation point. In Figure 5.2.1, 5.2.2, and 5.2.3 the dotted line is the calculated treatment planning system dose and the solid line is the measured and reconstructed dose. The profiles on the left are from the pre-treatment image reconstruction, the ones on the right are from the exit image reconstruction. Figure 5.2.4 Overlay of two isodose lines from both the treatment planning system and from Dosimetry Check. 38