Brachytherapy. Linda Poplawski. Graduate Program of Medical Physics Duke University. Date: Approved: Oana Craciunescu, Supervisor.

Transcription

1 Evaluating Dose Summation in Gynecological Brachytherapy by Linda Poplawski Graduate Program of Medical Physics Duke University Date: Approved: Oana Craciunescu, Supervisor Junzo Chino Joseph Lo Thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in the Graduate Program in Medical Physics in the Graduate School of Duke University 2015

2 Abstract Evaluating Dose Summation in Gynecological Brachytherapy by Linda Poplawski Graduate Program in Medical Physics Duke University Date: Approved: Oana Craciunescu, Supervisor Junzo Chino Joseph Lo An abstract of a thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in the Graduate Program of Medical Physics in the Graduate School of Duke University 2015

3 Copyright c 2015 by Linda Poplawski All rights reserved except the rights granted by the Creative Commons Attribution-Noncommercial Licence

4 Abstract Purpose Gynecological malignancies present challenges in determining an appropriate volumetric dose due to the highly variable physiologic activity of the surrounding tissue. Because of the high doses used in brachytherapy, surrounding structures have the potential to move around in the dose region and receive unknown amounts of radiation. Deformable image registration could overcome challenges in determining the true delivered dose through a dose accumulation process. This study uses two dose summation techniques to determine the efficacy of a deformation registration for the Syed-Neblett template applicators and cylinder applicators. Methods and Materials Data for patients treated with a vaginal cylinder or Syed-Neblett template were imported into the MIM software (Cleveland, OH). The bladder, rectum and applicator were contoured on each computed tomography (CT) scan. The deformable registration was applied to structures only by masking other image data to a single intensity with the purpose of focusing the registration on the high dose area, as well as to minimize any uncertainty from the CT data. The deformable registration flow consisted of the following steps: 1) Using a different Hounsfield Unit, the CTs were masked so that each of the structures-of-interest (SOIs) had one unique intensity value; 2) Perform a rigid registration between two image sets with alignment based on the applicator position; 3) Perform a deformable registration; 4) Refine registration by using local manual alignment in area with large contour changes; 5) Repeat steps 1 to 3 to register the desired structure from iv

5 all the subsequent fractions to the first fraction structure; 5) Transfer each deformed contoured to the first fraction CT. The deformed structure accuracy was measured by a comparison to the first fraction using the dice similarity coefficient (DSC). Two dose summation techniques were investigated: a) deform the desired structure to one fraction and determine the fractional dose to these new structures, and b) deform the fractional dose to one CT and accumulate to find the total dose. Point doses, D 2cm 3, were used as a comparison value for each method. Results The cylinder set of patients had a DSC ranging from 0.82 to 0.96 for bladder and 0.82 to 0.94 for rectum. The contour deformation addition method has variations up to 35% from the initial clinical point dose for the cylinder applicators. The dose deformation accumulation method gave up to a 15% difference from the clinical point dose. The Syed-Neblett template applicator patient set has DSC ranging from 0.53 to 0.97 for the bladder and 0.75 and 0.95 for the rectum. These registrations dose additions varied up to 35% and the dose deformation accumulation varied up to 58%. Conclusions With the changing anatomy in brachytherapy, deforming the dose with the end point of dose summation leads to different volumetric doses then when dose is recalculated on deformed structures, raising concerns about the accuracy of the deformed dose. Dice Similarity Coefficients alone cannot be used to establish the accuracy of a deformation for brachytherapy dose summation purpose. v

6 Contents Abstract List of Tables List of Figures List of Abbreviations and Symbols Acknowledgements iv ix x xiii xiv 1 Introduction Treating gynecological malignancies with radiation therapy External beam Brachytherapy Dose Additions GEC-ESTRO Guidelines Adding HDR Brachytherapy Fractions Combining External Beam Radiation Therapy and Brachytherapy Deformable Image Registration Deformation Algorithm Limitations Deformable Image Registration Phantoms Materials and Methods Clinical Case Selection vi

7 2.1.1 HDR Cylinder Cases HDR Syed-Neblett Template Applicator Deformable Registration Workflow MIM Software Deformation Process Hot Spot Location: D 2cm 3 Structure Dose Addition Methods First Method of Dose Summation: Contour Deformation Followed by Point Dose Addition (CDA) Second Method of Dose Summation: Dose Deformation Accumulation (DDA) Results Clinical Case Selection Deformable Registration Workflow Clinical Results Cylinder Applicator Patients Bladder Dose Rectal Dose D 2cm 3 Structures for the Bladder and Rectum Dice Similarity Coefficients of Deformed Structures Syed-Neblett Template Applicator Patients Bladder Dose Rectal Doses D 2cm 3 Structures for the Bladder and Rectum Dice Similarity Coefficients of Deformed Structures Discussion 65 vii

8 5 Conclusions 69 A Dose Addition Tables 70 Bibliography 74 viii

9 List of Tables 3.1 Characteristics of Cylinder Applicator Patients Characteristics of Syed-Neblett Applicator Patients Bladder Volume and DSC Table; Cylinder Cases Rectum Volume and DSC Table; Cylinder Cases Bladder Volume and DSC Table; Syed-Neblett Cases Rectum Volume and DSC Table; Syed-Neblett Cases A.1 Cylinder Applicator Dose Addition and Comparison Calculations.. 71 A.2 Syed-Neblett Template Applicator Dose Addition and Comparison Calculations for Rectum A.3 Syed-Neblett Template Applicator Dose Addition and Comparison Calculations for Bladder ix

10 List of Figures 2.1 Cylinder Applicator Syed Template Dice Similarity Coefficient Contour Deformation Followed by Point Addition Workflow Dose Deformation Accumulation Workflow Standard Imported Image Contour Masked Image Initial Rigid Registration Performed by MIM Software Primary and Deformed Image RegRefine Tool Example Contours Transferred to One Fraction CT Scout Images For Cylinder Applicator Scout Images For Syed-Neblett Applicator DVH Comparisons For Syed-Neblett Applicator Dose Distributions Before Deformation; Cylinder Applicator Dose Distributions Before Deformation; Cylinder Applicator Dose Distribution After Accumulation; CylinderApplicator CDA Example for Bladder; Cylinder Applicator Dose Difference for the Bladder; Cylinder Applicator Dose Difference for the Rectum with Cylinder Applicator x

11 3.16 Patient Image for Rectum Dose Deformation and Accumulation Outlier of 15.27% D 2cm 3 Location Example For Cylinder Applicator without Foley D 2cm 3 Location Example For Cylinder Applicator with Foley D 2cm 3 Location Example For Rectum for Cylinder Applicator D 2cm 3 Location Rendering for Cylinder Applicator Dice Metric as a Function of the Change in Bladder Size; Cylinder Applicators DSC vs Vol Patient Outlier Image for Cylinder Bladder with Large Volume Change Dice Metric as a Function of the Change in Rectum Size; Cylinder Applicators Dose Distributions Before Deformation; Syed-Neblett Applicator Dose Distribution After Accumulation; Syed-Neblett Applicator Contour Deformation and Addition Example for Bladder; Syed-Neblett Applicator Dose Difference for the Bladder; Syed-Neblett Applicator Patient Image for Bladder CDA Outlier Before Deformation; Syed- Neblett Patient Image for Bladder DDA Outlier of 37.87% Patient Image for Bladder DDA Outlier of % Dose Difference for the Rectum; Syed-Neblett Applicator Patient Image for Rectum CDA Outlier Before and After Deformation; Syed-Neblett Patient Image for Rectum DDA Outlier of 20.53% Patient Image for Rectum DDA Outlier of 46.88% D 2cm 3 Location Example For Bladder; Syed-Neblett Applicator D 2cm 3 Location Example For Rectum;Syed-Neblett Applicator xi

12 3.37 D 2cm 3 Location Rendering for Syed-Neblett Applicator Cases Dice Metric as a Function of the Change in Bladder Size; Syed-Neblett Applicators DSC vs Vol Outlier Patient Image for Bladder with Large Volume Change; Syed-Neblett Applicator DSC vs Vol Patient Outlier Image for Bladder with Small Volume Change; Syed-Neblett Applicator Dice Metric as a Function of the Change in Rectum Size; Syed-Neblett Applicators Dose Difference Comparison xii

13 List of Abbreviations and Symbols CT DSC D 2cm 3 CTV HR-CTV IR-CTV MR BED EQD2 DDA CDA Computed Tomography Dice Similarity Coefficient Dose to 2 Cubic Centemeters Clinical Target Volume High Risk Clinical Target Volume Intermediate Risk Clinical Target Volume Magnetic Resonance Bioloigical Equivalent Dose Equivalent Dose to 2 Gray Dose Deformation Accumulation Contour Dose Addition xiii

14 Acknowledgements I would like to thank my advisor, Dr. Oana Craciunescu for working with me on this project. Thank you to Dr. Joseph Lo and Dr. Junzo Chino for serving on my committee. Thank you to Dr. Taoran Li for guiding me through the beginning steps of this process. Thank you to the Duke Medical Physics Graduate Program for letting me be a part of this talented group of people, where I have learned so much more than physics. Lastly, thank you to my family for supporting me through these past 2 years. xiv

15 1 Introduction 95,000 women were estimated to be diagnosed with gynecological malignancies for 2014 alone. About 12,900 of these cases will be invasive cervical cancer [1]. Approximately 54,870 new cases of cancer of the uterus will be diagnosed [2] and 1,600 of these will be uterine sarcomas [2]. A rare gynecological cancer is located in the vagina and about 1 of every 1,100 women will develop it [3]. Gynecological malignancies are the leading cause of cancer related death in developing countries, with 500,000 new cases and 233,000 deaths per year, worldwide, due to cervical cancer. These cancers are treated with surgery, chemotherapy, radiation therapy or a combination of the three. 1.1 Treating gynecological malignancies with radiation therapy External beam The current standard of care for gynecological malignancies is radiation therapy [4]. This consists of external beam radiation therapy with a brachytherapy boost. The external beam radiation therapy technique uses two sets of parallel-opposed beams, which treats the entire pelvic region [5]. This 4-field box technique gives an equal 1

16 dosage to all structures in the pelvis. Additionally, new techniques such as intensity modulated radiation therapy (IMRT) or arc rotation can help with tumor control [4]. However, even with these techniques, a brachytherapy boost can allow for greater tumor control Brachytherapy For gynecological malignancies, brachytherapy can be used as a boost to external beam radiation therapy, or as the sole radiation treatment. Brachytherapy allows high doses of radiation to the primary tumor and reasonable sparing to the organs at risk [6]. An applicator is inserted into the patient and the radioactive source is placed within the applicator at several positions to create a dose distribution. When used as concurrent therapies, the dose delivered to the patient must be calculated by adding the dose from both external beam therapy and brachytherapy. However, these two therapies have different methods of delivery and require the patient in different positions. It is hence unclear exactly how much dose is delivered to the patient because these therapies do not have the same biological effect and because there is no accepted method to sum doses volumetrically. There are two categories of brachytherapy procedures: High Dose Rate (HDR) and Low Dose Rate (LDR). HDR brachytherapy refers to sources with a dose rate of 20 cgy/min or higher [7]. HDR treatments for gynecological tumors are traditionally intracavitary or interstitial techniques and require an after-loader. LDR treatments are at dose rate in which the source can be permanently implanted into the patient and will deliver dose over the entire lifetime of the source. Each of these techniques have different advantages depending on the malignancy location. In order to standardize all gynecological brachytherapy treatments, GEC-ESTRO created guidelines for target delineation and planning on different imaging modalities. Two different clinical target volumes (CTV) were created: high risk CTV and 2

17 intermediate risk CTV [8]. The HR-CTV is defined as the CTV with a large risk of local reoccurrence from residual disease. This volume should receive the highest possible total dose, in order to ensure all macroscopic disease in this area has been treated. The IR-CTV is defined by GEC-ESTRO as an area of microscopic disease at the time of brachytherapy treatments. This target volume should receive a curative dose for the microscopic disease, which defined as 60 Gy for cervical cancer. These target volumes are constrained by the natural anatomical borders, such as the bladder and rectum and are based on the target location and volume at diagnosis and at each brachytherapy fraction. The tumor volumes are ideally viewed on magnetic resonance (MR) images with the brachytherapy applicator in place. MR images have the best definition of the tumor volume [8]. CT images provide comparable definition of the bladder, rectum, sigmoid, bowel and vagina, but not the target volume form normal tissue [9]. 1.2 Dose Additions GEC-ESTRO Guidelines Point doses were used to define the normal tissue effects because brachytherapy was traditionally planned on film radiographs. ICRU Report 38 describes a point at the bottom of the Foley catheter for the bladder dose calculation point. The rectum dose calculation point is 5 mm posterior to the vaginal wall at the level of the mid-ovoid. However, dose volume relations better describe the volumetric dose effects in an organ. Unlike external beam radiation therapy, brachytherapy has large dose gradients occurring in the organs at risk and therefore give heterogeneous dose distributions within those organs. Due to the known inverse square effect, areas closer to the applicator receive a higher dose than those farther away, which is seen when comparing the anterior rectum wall to the posterior rectum wall, for example. Dose volume points assume that the region of interest is the same each time that 3

18 it is measured. Because this is not true, due to tumor shrinkage and normal tissue movement, dose volume relations are considered a worse case assumption. [9]. Dose parameters are obtained from the dose volume histogram (DVH) of the organ. The dose to the organs at risk is described by small volumes, because of the rapid dose fall off in the area in nearest the applicator. The minimum dose to 1 and 2 cm 3 (D 1cm 3 and D 2cm 3) should be recorded and correspond to biological endpoints, such as organ toxicity. [9] Adding HDR Brachytherapy Fractions The location of the high dose region is not always the same for each fraction. This is due to tumor shrinkage, as well as the change in normal tissue topography. The brachytherapy applicator also impacts the normal tissue distribution because it is not inserted identically each time. Point doses are used to calculate the total dose from the brachytherapy treatment. Biological effects of radiation therapy depend not only on the dose delivered, but the dose distribution, treatment volume, dose rate and treatment time [9]. These effects must be taken into account when adding doses between fractions and between therapies. The biological equivalent dose (BED) uses a biological damage model to convert all of the different doses into a comparable form. The Equivalent Dose to 2 Gray (EQD2) is used to determine the dose that would give the same biological damage as an external beam treatment delivered in 2 Gy fractions. This EQD2 factor allows the dose from each brachytherapy fraction to be summed, as well as the external beam doses, to create a total dose from the treatment Combining External Beam Radiation Therapy and Brachytherapy When combining the doses delivered to tissues from external beam radiation therapy and brachytherapy, each tissue volume element from external beam must be matched 4

19 exactly to the same tissue volume element in brachytherapy. Matching these voxels requires complex calculations and some assumptions are created instead. Additionally, the insertion of an applicator for brachytherapy changes the normal tissue location. When combining external beam therapy and brachytherapy dose distributions, both must be represented as the biologically weighted dose. If the external beam fractionation schedule differs from 2 Gy per day, the EQD2 must be calculated. The summations of distributions give a cumulative biologically weighted dose for the volume of interest. The dose to the tumor and organs at risk are calculated by adding the D 90 or D 98 and D 2cm 3, respectively, from the two treatments. Currently, rigid registration of the fractions gives a rough estimate of the changes in dose distribution, from fraction to fraction. This type of registration only has the ability to translate or rotate the moving image. It cannot adjust for any changes such as organ filling, organ emptying, or insertion of an applicator. It is suggested that the use of deformable image registration would allow for a better volumetric dose summation that will better assess the correlation between certain dose metrics and treatment endpoints, like local control, or normal tissue toxicities [10]. This would give volumetric dose distribution to better assess the potential biological endpoints, due to change in anatomical topography changes, as well as patient position and the introduction of applicators in the brachytherapy case. 1.3 Deformable Image Registration Deformable image registration is used to take a source image and warp it to match a target image. There are two types of deformation algorithms: intensity and landmark based. Landmark based registrations match the same landmarks on the source image to the target image and interpolates the points in between the landmarks [11]. The landmarks are matched exactly or approximately, depending on how well the landmarks are known. Landmark based methods are generally faster than intensity 5

20 based, but it is difficult to know if the deformed points on a feature correspond to the same exact point. Intensity based deformable registration uses all of the data stored in the CT images. The points are transformed from the source CT to the target CT by matching the intensity values of voxels and a smoothing factor is used to constrain points nearby each other. Intensity based methods are subject to image artifacts. Each voxel in the source image can move in 3 dimensions in order to match the target image. The images are measured to determine the difference between the intensity of each pixel using a metric such as sum of squared differences, cross correlation, or mutual information. There are three groups of methods to model a deformation transformation: global modeling, semi-local, and local [12]. The accuracy of these algorithms is a constant point of discussion Deformation Algorithm Limitations A multi-institutional study assessing model based deformable registration shows that the deformable image registration can have registration errors of up to 0.05 cm for the prostate [13]. This method used point based registration methods and surface mesh defined by the contour segmentations between two MR images. This study focused on one algorithm but used various anatomical sites and image types. A separate multi-institutional study looked at various sites from external beam therapy and different image types, and also found results for many deformable image registration algorithms. The duration of the registrations showed potential for real time deformable registration[14]. However, some algorithms took more than 10 hours to complete. Three out of the 11 algorithms studied had a maximum error of 5mm and 7 had a maximum error of 7mm. The images used for registration were specified by anatomical site and the deformation algorithm was described its similarity metric, regularization method, and optimization type. The study found that deformable registration algorithms overall perform well in areas with strong contrast, but are 6

21 much less accurate for areas of low contrast, such as prostate, or between modalities. An area of low contrast that is of great concern for gynecological brachytherapy is the bladder. Andersen et al compared the DVH parameter additions for the bladder to a dose accumulation process using an in house-created deformable registration algorithm for brachytherapy [6]. The study used a contour-based deformable registration algorithm, which was inspected after each registration to ensure it created a feasible deformation. The DVH parameter system tended to overestimate the doses for 2cm 3, 1cm 3 and 0.1cm 3 compared to the dose calculated from the deformable image registration. Only 2% of the patients had a deviation greater than 5% when comparing the DVH addition to the deformable dose. Because the dose deformation addition and DVH parameter addition were closely related, it is believed that the hot spot location is in the same region of the bladder for both brachytherapy fractions. If the hot spot location varied more, the DVH parameter would not match the deformable dose addition as well. Sabater et al also looked at dose addition for gynecological brachytherapy and compared the dose to the rectum and the bladder with 3 dose accumulation processes for the vaginal cuff[15]. The study accumulated dose by a single plan, a rigid registration, and a deformable registration using pelvic CTs. The deformable registration was performed using the 3d-SLICER General BRAINSFit Registration module and a b-spline deformation method. For the rectum, the study found that there was no statistically significant differences between a single plan distribution and either the rigid registration or deformable registration. The deformable registration gave values closer the single plan dose accumulation. For the bladder, there were significant differences between the dose deformation and the single plan dose accumulation. This study presents small dosimetric differences between the single plan addition and deformation addition. 7

22 1.3.2 Deformable Image Registration Phantoms A deformable phantom allows the quality of the deformable registration algorithm to be quantified. Two commercially available systems are MIM Software and Velocity. These systems have been evaluated by several deformable phantoms to describe the accuracy the each deformation process. The difference between the prediction and the true deformation can range from 0 to 9.0 mm for a Presage phantom [16]. The regions of maximum deformation had the most inaccuracy. This phantom has the ability to determine the dose at each state and determine the difference between the deformed and total delivered dose. Although the contours could be deformed well, the dose deformation by the algorithm was less accurate with only 60% of the points having a gamma-index passing rate. When focusing on anatomical site with a virtual anatomically defined phantom, Nie et al found that MIM software showed the greatest accuracy for low-contrast small regions that underwent significant deformation, such as bladder expansion for prostate when using a physical deformable phantom[17]. This would infer that the deformation would accurately deform our region of interest. However, it is important to note that the same study also found that for large field deformations with strong contrast, this approach may increase errors. [17]. Any strong contrast changes in our method may decrease the accuracy that the system would normally show. A 3-dimensional head and neck deformable phantom evaluated the accuracy of several deformable registration algorithms [18]. Images of the phantom were taken with kv and MV imaging before and after deformation. Error for this study was considered the 3D distance to agreement between the ground truths, the known deformation in the phantom, and the deformation created by the algorithm. For the MIM system, the mean error was 1.8 mm and the maximum error was 11.5 mm for a kv-to-kv image registration. The Velocity system had a mean error of 2.4 8

23 mm and a maximum of 7.1 mm. Rigid registration had a mean error of 3.5 mm and maximum error of 8.2 mm. The largest errors for the MIM system are found in a homogeneous region, due to its focus on image similarity. This system may produce nonphysical deformation fields because of this focus. Velocity also has large errors in the homogenous region, but uses a different similarity metric to create the deformation. In another study, a deformable pelvis phantom was used to measure the accuracy of 11 different deformable registration algorithms [19]. For the MIM system, 9.8% of points had an error greater than 3 mm and 3.7% of points had and error greater than 7mm. When using the MIM deformation algorithm, the dice similarity coefficient was the highest of the study at This system had the lowest mean absolute difference between images, but the largest percentage of errors over 7mm. With the Velocity system, 3.0% of all the point had an error greater than 3 mm and 2.0% of the point had an error greater than 7mm. The dice similarity coefficient for this system was This system has the lowest mean error of the study. MIM has less of an effect from noise in areas of high contrast, then other algorithms. However, it in regions of low contrast it is greatly affected by the noise. The low mean absolute difference means that the moving image was very similar to the fixed image, despite the large errors. Given the importance of dose summation in radiation therapy for gynecological cancers, and the known issue with limitations of the commercially available deformation systems, this study examines the impact that deformable image registration can have on a brachytherapy dose distribution, with the intent of reporting the total dose delivered to the two organs at risk: bladder and rectum. Two different point dose addition methods will be compared to the current clinical standard for dose calculation. The following chapters will discuss the clinical workflow for brachytherapy, as well as the flow to add doses from different brachytherapy fractions to determine 9

24 a total dose delivered to the organs at risk. These chapters will show examples of images for the deformation process, as well as comparisons to the current clinical dose calculation standard. Patient cases in which the deformed dose addition methods vary greatly from the current clinical standard will be discussed to determine any potential causes for the large deviation. Lastly, there will be discussion on how these results compare to previous studies and any potential improvements to deformation process from the clinical setting. 10

25 2 Materials and Methods 2.1 Clinical Case Selection Image deformation for HDR brachytherapy is particularly difficult because of the high dose gradients achieved from the sources. For this reason, this study chose two types of treatments, in which there is minimal change between fractions: 1) HDR interstitial treatments using a Syed-Neblett template; 2) HDR treatment using stump cylinders. The interstitial technique was chosen because the Syed-Neblett template is sutured to the perineum, the needles provide stability to the implant, and the position of the applicator is checked daily with imaging and depth measurements. Cylinder applicator cases were selected as they have a simple geometry; they are inserted to same depth, and imaged every fraction. These characteristics make the cylinder applicator also ideal to examine deformable registration for HDR brachytherapy. For each of these cases, only bladder and rectum are considered in the dose summation study. We excluded sigmoid and bowel from this analysis due to the variation at contouring between fractions as well as the large variability in the position at each treatment. 11

26 2.1.1 HDR Cylinder Cases HDR Cylinder applicators are one example of an intracavitary treatment. The Duke clinic uses stump cylinders manufactured by Varian (Varian Medical Systems, Inc., Palo Alto, CA). Examples are shown in Figure 2.1. These applicators are cylindrical in shape and have a single channel. They vary in size from 2 to 3.5 cm diameter. Typical prescriptions vary, but they can be between 4 to 7Gy/Fx for 3 to 5 fractions and dose prescribed 0.5 cm in tissue or on the surface of the cylinder. The cases considered in this study had each five fractions, each of 500 cgy prescribed at 0.5 cm in tissue. We have selected patients that had a Foley catheter placed for each fraction, patients that were treated without a catheter and patients with catheter only in the first fraction, to understand the effects of the Foley on the bladder shape and proximity to the cylinder. At the first fraction, following a clinical exam, the physician places a gold marker that defines the distal extent of the disease. For each fraction, the cylinder applicator is inserted to the level of the gold marker and the position verified with imaging. As such, the cylinder can serve as a confident marker for the deformable registration. The patients are treated with the same plan each fraction and the D 2cm 3 metrics for OAR are extracted from the DVH and converted to EQD2 dose. The clinical standard for dose summation for brachytherapy consists of adding the EQD2 D 2cm 3 point doses from each of the fractions. 12

27 Figure 2.1: Cylinder Applicators HDR Syed-Neblett Template Applicator The Syed-Neblett template applicator has a plastic cylinder attached to a base that is sutured to the perineum. The cylinder is inserted in the vagina and interstitial guide needles can be guided through holes in the base into the tumor. A typical clinical Syed-Neblett template is shown in Figure 2.2. The cervix is dilated and then the uterine catheter is inserted. Markers are placed at the cervix and the initial needle is placed two to thee cm beyond the cervical os. This needle serves as a guide to the depth that all other needles are inserted. Needle placement is under transrectal ultrasound. Once all the needles are placed, the template is sutured to the perineum [9]. All patients had a Foley catheter to empty the bladder. The sutured template gives confidence for a registration point. Additionally, the needles are measured before treatment to ensure that they have not shifted. The Syed-Neblett template patients were treated with 5 fractions over 3 days, with images at the first fraction, 13

28 second fraction (day 2, morning) and fourth fraction (day 3, morning). Because anatomy may change after planning, using the images from fraction 2 or 4 a post plan can be created to show the dose delivered with the new anatomical topography. If necessary, the plan can be altered to achieve the desired dose distribution. D 2cm 3 metrics for OAR are extracted from the DVH of each fraction and converted to EQD2 dose. The clinical standard for dose summation for brachytherapy consists of adding the EQD2 D 2cm 3 point doses from each of the fractions. Figure 2.2: Syed Template (Best Medical International, Inc, Springfield, VA). 2.2 Deformable Registration Workflow MIM Software The deformations were created in MIM Maestro, by MIM Software (Cleveland, OH), referred to as MIM. The software uses an intensity-based algorithm with a free-form regularization and focused for CT-to-CT registration. A gradient steepest descent 14

29 algorithm determines the optimization of the deformation. It has essentially limitless degrees of freedom for each voxel to allow for global and local deformations [20]. The quality of deformation is determined by a squared differences metric. This compares the difference in intensities of the moving and target image. The MIM deformation software has a feature that allows the user to see the rigid deformation at each point [21]. Reg-Refine, as this feature is called, allows the user to see the where the deformation is incorrect and adjust these points. The deformation can be performed iteratively and continuously to correct places where the deformation is unsatisfactory until a optimal deformation is achieved Deformation Process Cylinder and Syed-Neblett template applicators both had the same workflow. They differed only in how many image registrations were required. Each image is imported into the MIM software, along with its corresponding dose and structure set. The body, rectum, bladder, and applicator were contoured for each image. If clinical contours were available, these contours were used. The rectum clinical contours may have variation in the superior start and inferior end locations. The structures of interest must be masked, in order to have the registration focus on no other image data. The structures of interest include the body, applicator, and bladder or rectum. Masking the applicator gives confidence in the dose deformation because we know that the location of the sources is relatively correct when the applicators are matched. The body contour must be set to an intensity value of 0, before setting the applicator and bladder to 1000 and -200 Hounsfield Units, respectively. These values are applied to all of the images for each brachytherapy fraction of the patient. The newly masked images are registered together using a rigid registration. We elected to use the first fraction as the primary image for all registrations. Because 15

30 there is so little image information, we direct the software to realign the registration based on a volume of interest (VOI). The VOI is placed around the applicator and part of the high dose region for the structure of interest. The new alignment registers the applicator. Caution must be taken to ensure that the applicator has not shifted. Large differences between the location of the body contours relative to each other would hint to this issue. Once the applicators are rigidly registered, the deformable registration is performed. The rigid registration of the applicators serves as a guide or the deformable image registration. The field of interest for the registration is a volumetric box containing all image data. The primary image is the image of the first fraction. The RegRefine tool shows the rigid registrations at each point that makes up the deformations. This function also allows the user to create new alignments for points that the software did not deform well. These alignments are based on a boxed based rigid registration with emphasis on the center point. The user can edit the box-based registration to achieve the best registration at the center point. When the user agrees with an alignment, it is saved as a lock in the software. A few points that the deformation algorithm ran well should also be locked to prevent those points from becoming misaligned. Once the user has placed several locks, the deformation algorithm on the image can be run again, with these them as guidelines. The user can edit and view the newest deformation with the RegRefine function and create new locks at unfavorable registration points. The deformation algorithm can be run once again and will use the previous locks, as well as the newly created locks. This iterative process can continue until an acceptable deformation has been achieved. This deformation workflow is repeated to deform each remaining fraction to the initial fraction image. The deformations are also carried out with the second organ at risk, the rectum. The initial fraction image is the primary image and the image is 16

31 masked with the same intensity values as the bladder: 0 for the entire body, 1000 for the applicator. To focus on the rectum for these registrations, the bladder structure is not masked and the rectum structure is given an intensity of -200 Hounsfield Units. The deformation process is run for all fractional images with the rectum as the structure of interest. The organ of interest (either bladder or rectum) from the secondary image is transferred to the first fraction using the deformable image registration. The same organ of interest is transferred from any remaining fraction images to the first fraction image. To determine how well the transferred contours match the first fraction contour, the dice similarity coefficient is calculated comparing each transferred contour to the contour of the first fraction. The dice similarity coefficient (DSC) is calculated by: DSC A X B A ` B (2.1) The DSC is a number between 0 and 1, where 0 represents no overlap, 1 represents total overlap, and anything between represents a partial overlap. 17

32 Figure 2.3: Dice Similarity Coefficient Calculation Hot Spot Location: D 2cm 3 Structure The hot spot in each organ at risk may not always be in the same location on the structure for each fraction. If the hot spot moves location between fractions then the structure may not receive the dose that is estimated in the clinical point dose calculation. For each fraction, the dose level corresponding to the D 2cm 3 value for both bladder and rectum were converted into a structure with the purpose of identifying the location of the hottest D 2cm 3 area within the organ. To create the D 2cm 3 structure, the dose to 2cm 3 is determined by the dose volume histogram (DVH). An isodose line is created for this dose value and a new structure is created from the isodose line. The D 2cm 3 structure is created by using a boolean operator to find the 18

33 intersection of the structure of interest and isodose line. These structures can also be deformed using the appropriate deformation matrix for the main structure that contains this D 2cm 3 and compared to the D 2cm 3 of the first fraction also using a dice similarity coefficient. 2.3 Dose Addition Methods For clinics that do not image at every fraction, the dose distribution is determined by multiplying the number of fractions by the dose distribution (5 times Fraction 1 for this case). For clinics that image at each fraction, a rigid registration is performed to roughly estimate the dose delivered. Deformable registration may give a more realistic distribution of the dose delivered during treatment First Method of Dose Summation: Contour Deformation Followed by Point Dose Addition (CDA) Using the fractional dose, the point dose, D 2cm 3, is calculated on each deformed organ of interest structure and the first fraction s structure. These are added up to determine a total point dose to the organ. The first fractional dose is calculated on the deformed structure. To compare this to the clinical value, the number of fractions, in this case 5, is multiplied by the clinic point dose for the first fraction. Dose values should be calculated for both the bladder and the rectum using the deformations created for the respective organ. Figure 2.4: Contour Deformation Followed by Point Addition Workflow 19

34 2.3.2 Second Method of Dose Summation: Dose Deformation Accumulation (DDA) The deformation matrix used to deform the contours is also used to deform the dose from each fractional dose to the first fraction, using the deformation specific to the organ of interest. Once all of the dose distributions are transformed to the first fraction image, the dose is added using MIM s dose accumulation function. This function allows the user to input how many fractions should be accounted for each dose distribution. For the Syed-Neblett template applicator, the first fraction dose contributions 1 fraction, the second fraction dose contributes two fractions and the fourth fraction dose contributes two fractions. For the cylinder, each fraction contributes one fraction dose. The point dose to the first fraction contour is calculated from the total accumulated dose. The doses are compared to the clinical point dose by % difference Dose clinical Dose deformed Dose clinical ˆ 100 (2.2) 20

35 Figure 2.5: Dose Deformation and Accumulation Workflow 21

36 3 Results 3.1 Clinical Case Selection Five patients treated with a cylinder applicator and ten patients treated with a Syed-Neblett interstitial implant were considered in this study. Of the five cylinder patients, two had Foley catheter inserted at each fraction and three had no catheter. All Syed-Neblett patients had a Foley catheter inserted. One Syed- Neblett template patients also had a tandem, in addition to the interstitial needles. 3.2 Deformable Registration Workflow The deformation flow was similar for both cylinder and Syed-Neblett patients. For simplicity, the results exemplifying the deformable registration flow are going to use a Syed-Neblett example case only. Figures 3.1 to 3.6 show results of each step of the deformation workflow. Figure 3.1 shows an example of the Syed-Neblett template treatment images, which have been imported into the MIM system. 22

37 Figure 3.1: Image Imported into MIM Software with Structures Contoured. (Syed Applicator, Top: Fraction 1, Middle: Fraction 2, Bottom: Fraction 4) Bladder contoured in yellow, Syed-Neblett applicator in cyan. Figure 3.2 shows the same Syed-Neblett template patient with CT information removed and the structures of interest masked with specific intensity values. Rigid registration is performed first, concentrating on the applicator and one normal structure at a time, in this case, bladder. 23

38 Figure 3.2: Patient Image Showing the Masking of Body, Applicator, and Bladder.(Syed Applicator, Top: Fraction 1, Middle: Fraction 2, Bottom: Fraction 4) To ensure a good registration in areas of expected high dose gradients, Figure 3.3 shows the use of a volume of interest around the applicator and parts of the bladder volume to achieve the registration. 24

39 Figure 3.3: Top: Initial Rigid Registration Using MIM Software on the Masked Images Before Using a Volume of Interest (VOI) (2D Box in Figure) Bottom: Registration After Focusing on the Applicator and High Dose Regions of the Bladder With VOI Figure 3.4 shows the primary image and the secondary image after the first iteration of deformable registration. To better see the points of registration, Figure 3.5 is an example of how the RegRefine tool looks at the point-wise rigid registration. 25

40 Figure 3.4: An Example of a Primary and Deformed Image. Left: Primary Image, Right: moving image deformed to the primary image. Figure 3.5: RegRefine tool shows the Rigid Registration at a Single Point. Multiple such points are selected and used to guide the deformable registration. Figure 3.6 shows an example for Syed-Neblett template applicator patients in which all of the contours from subsequent imaged fractions are transferred to the 26

41 first fraction image. Figure 3.6: Syed Applicator Patient With Fraction 2 and Fraction 4 Bladder Contours Deformed onto Fraction 1 Image.(Fraction 1 -Yellow; Fraction 2-Green; Fraction 4- Purple) 3.3 Clinical Results Figure 3.7 shows the scout images from each fraction of a representative HDR Cylinder patient. These images show that the cylinder applicator is inserted to the gold marker for each fraction. Table 3.1 gives a description of the target volumes, number of needles, and source dwell times used to treat the Cylinder applicator cases. 27

42 Figure 3.7: Example cylinder applicator patient: Scout Images for Fractions 1 through 5 (Cylinder size = 3.0 cm diameter, Foley). Table 3.1: Characteristics of the Treatment Information for Cylinder Applicator Cases. Total Dwell Time is Calculated with a 10 Ci Nominal Source. Figure 3.8 shows the scout images for a representative Syed-Neblett interstitial patient. These images, together with the depth check performed pre imaging ensure that the needles have not shifted from treatment depth for each day of treatment. 28

43 Figure 3.8: Example Syed-Neblett applicator patient: Scout Images for Fraction 1 (Day 1), Fraction2 (day 2 am), and Fraction 4 (Day 3 am) for the Syed- Neblett Template Applicator. Table 3.2 gives a description of the target volumes, number of needles, and source dwell times used to treat the Syed-Neblett template cases. Table 3.2: Characteristics of the Treatment Information for Syed-Neblett Template Applicator Cases. Total Dwell Time is Calculated with a 10 Ci Nominal Source For a typical Syed-Neblett template case, Figure 3.9 shows the DVH comparisons between these different methods of dose addition for the target (the high risk CTV defined by the GEC-ESTRO guidelines, 2005) and the normal tissue structures considered in his study bladder, and rectum. 29

44 Figure 3.9: Comparison Between Dose Volume Histograms For Different Clinical Addition Techniques. The DVHs displayed represent the dose to the bladder, rectum, and HR-CTV. The dose addition methods for each structure are rigid registration, deformable registration, and 5 Times Fraction 1 Distribution. 30

45 3.4 Cylinder Applicator Patients Bladder Dose Figure 3.10 shows a typical cylinder applicator patient with the corresponding dose distribution and bladder contour for each fraction. Note that this particular patient had a Foley inserted at each fraction, whereas Figure 3.11 also shows an example of a patient without a Foley in the bladder. Figure 3.12 shows the dose distribution after deforming and accumulating all doses to the first fraction image. Figure 3.13 shows a typical cylinder patient with the contours from subsequent fractions deformed onto the image of the first fraction. The first fraction dose distribution is shown with this images. Figure 3.10: The Dose Distributions Before the Deformation for the Bladder; Top to Bottom: Fraction 1 to 5. Notice Foley in bladder for each fraction. Variations in bladder volume of 3.9% to 72% from the first fraction bladder contour. Variations in rectum volume of -22.9% to 61.0% from the first fraction rectum contour. 31

46 Figure 3.11: The Dose Distributions Before the Deformation for the Bladder; Top to Bottom: Fraction 1 to 5. Notice the lack of Foley in bladder for each fraction. Variations in bladder volume of -39.4% to 51.1% from the first fraction bladder contour. Variations in rectum volume of -6.82% to 61.6% from the first fraction rectum contour. Figure 3.12: Total Dose Distribution After All Fractional Doses Have Been Deformed and Accumulated onto Fraction 1 Image for Bladder. An example of DDA 32

47 Figure 3.13: Dose Distribution For Fraction 1 After All Contours Have Been Deformed onto Fraction 1 Image for Bladder. An example of CDA. Figure 3.14: Dose Difference Expressed as a Percentage between the Dose Addition Techniques and the Clinical Standard for D 2cm 3 Bladder Metric. 33

48 Figure 3.14 shows the percent dose difference between the clinical standard of D 2cm 3 point dose summation and each of the CDAs and the DDAs for the bladder. The CDA showed a mean dose difference of 2.46% 8.93% for the bladder from the current clinical standard. The differences range from 7.15% to 16.98%. The DDA has a mean dose change of 1.66% 6.12%. The dose difference ranges between -7.76% and 8.33% Rectal Dose Figure 3.15: Dose Difference Expressed as a Percentage between the Dose Addition Techniques and the Clinical Standard for D 2cm 3 Rectum Metric. The distribution of dose variations between the addition methods and the clinical point dose standards is seen in Figure The CDA differences for the rectum have a mean change of 11.70% 16.85% from the current clinical standard. The dose differences for the rectum range between 34

49 2.22% and % for the cases when the clinical standard D 2cm 3 point summation was compared to CDA. These values are not equally spread about the median and are highly skewed to the positive side. This shows that the clinical point dose method overestimated the dose compared to this dose deformation technique. This could be due to variations in the hotspot location throughout the treatment. The DDA, for the rectum and cylinder applicator, results in a mean dose difference of 1.96% 7.92% from the clinical standard. The range of dose differences is 5.96% to 15.27%. This distribution has an outlier at 15.27% difference from the clinical standard. The patient images for this outlier are shown in Figure In this case, the clinical point dose standard estimated more dose than the DDA method. This may be an outlier due to the bend of the superior region of the rectum. The clinical point dose may overestimate because of the change in hot spot location. Additionally, some of the rectum contours have lower end positions and therefore the dose may be shifted upward, moving the high dose region away from the first fraction contour. 35

50 Figure 3.16: Dose Distribution for the Rectum Dose Deformation and Accumulation (DDA) Method Outlier of 15.27% Difference From Clinical Standard D 2cm 3 Structures for the Bladder and Rectum An example of the locations for the D 2cm 3 structure for the bladder without a Foley catheter is shown in Figure Alternatively, Figure, 3.18 shows the location of the D 2cm 3 structure for the bladder when the treatment used a Foley catheter. Figure 3.19 shows the location of the D 2cm 3 contour for each fraction for the rectum. Figure 3.20 shows a rendering perspective of the D 2cm 3 locations for the rectum and bladder respective to the cylinder applicator. 36

51 Figure 3.17: Cylinder Applicator Example for Location of the D 2cm 3 Structure for Bladder without a Foley Catheter Inserted. Top to Bottom: Fraction 1, Fraction 2, Fraction 3, Fraction 4, Fraction 5. 37

52 Figure 3.18: Cylinder Applicator Example for Location of the D 2cm 3 Structure for Bladder with a Foley Catheter Inserted. Top to Bottom: Fraction 1, Fraction 2, Fraction 3, Fraction 4, Fraction 5. 38

53 Figure 3.19: Cylinder Applicator Example for Location of the D 2cm 3 Structure for the Rectum. Top to Bottom: Fraction 1, Fraction 2, Fraction 3, Fraction 4, Fraction 5. Figure 3.20: Cylinder Applicator Example of D 2cm 3 Locations Rendering. D 2cm 3 locations for bladder and rectum are shown as white or pink lines in the organ at risk 39

54 3.4.4 Dice Similarity Coefficients of Deformed Structures All of the deformations for the cylinder applicator patients have a dice similarity coefficient of at least 0.82 for the bladder or rectum contours when they are deformed to the first fraction. The maximum dice similarity coefficient for the bladder and cylinder is 0.96, while the maximum for the rectum is Figures 3.21 and 3.23 show the variation of the dice values with the change in size to the original fraction volume for cylinder applicators. For the bladder, the dice values decrease when the registration requires large changes in volume, such as shrinking to nearly -300% of the original size. Patient A from Figure 3.21 requires a large change in bladder size and therefore this dice value is considerably low. The patient images for this deformation is shown in Figure 3.22 Figure 3.21: Relationship between the Dice Similarity Coefficient and the Change in Bladder Volume when Deforming to the First Fraction. Each point on the graph represents a fraction that was deformed. Table 3.3 shows the bladder volume for each fraction before and after deformation and the corresponding dice similarity coefficient for each Cylinder case. 40

55 Table 3.3: Bladder Volume and DSC Table; Cylinder Cases. 41

56 Figure 3.22: Dice Similarity Coefficient vs Volume Outlier Patient Image for Bladder with Larger Volume Change. (Top: Fraction 1 Image; Bottom: Fraction 2 Image) In this case, the patient had a foley catheter for the first fraction but did not for subsequent fractions. Figure 3.23: Relationship between the Dice Similarity Coefficient and the Change in Rectum Volume when Deforming to the First Fraction. Each point on the graph represents a fraction that was deformed. 42

57 Table 3.4 shows the rectum volume for each fraction before and after deformation and the corresponding dice similarity coefficient for each Cylinder case. Table 3.4: Rectum Volume and DSC Table; Cylinder Cases. 3.5 Syed-Neblett Template Applicator Patients Bladder Dose Figure 3.24 shows a Syed-Neblett applicator patient with the corresponding dose distribution for each fraction. Figure 3.25 shows the dose distribution after deforming and accumulating all doses to the first fraction image. 43

58 Figure 3.24: The Dose Distributions Before the Deformation; Top to Bottom: Fraction 1, Fraction 2(morning day 2), Fraction 4 (morning day 3) for the bladder. Bladder size varies from -47.5% to 35%. Rectum size varies from 43.5% to 35%. Figure 3.25: Total Dose Distribution After All Fractional Doses Have Been Deformed and Accumulated onto Fraction 1 Image for Bladder. A DDA Example Figure 3.26 is an example of a Syed-Neblett template applicator in which the contours from fraction 2 and fraction 4 images are deformed to the first fraction image. These contours are shown with the dose distribution planned for fraction 1. 44

59 Figure 3.26: Dose Distribution For Fraction 1 After All Contours Have Been Deformed onto Fraction 1 Image for Bladder. An example of CDA. The Deformed Fraction 2 Bladder had a 2.15 mm maximum difference from the Fraction 1 bladder in the high dose region. The Deformed Fraction 4 bladder had a 3.2 mm maximum difference from the Fraction 1 bladder in the high dose region. 45

60 Figure 3.27: Dose Difference Expressed as a Percentage between the Dose Addition Techniques and the Clinical Standard for D 2cm 3 Bladder Metric. The distribution of the bladder dose differences for Syed applicator patients are seen in Figure The dose differences for bladder CDA from the clinical standards had a mean of 4.65% 8.52%. The difference ranges between % and 3.27% for CDA. Figure 3.28 shows the patient images for the CDA method s outlier of %. The CDA method estimates a larger dose delivered to the bladder than the clinical point dose method. This could be due to the bladder shifting away from the applicator for fractions 2 and 4. The CDA method would then potentially give a higher dose than actually received by the bladder when deforming the contours to be near the applicator and then recalculating the dose from fraction 1. The deformed fraction 2 bladder had a 1.24 mm maximum difference from the fraction 1 bladder in the high dose region. The deformed fraction 4 bladder had a 1.85 mm maximum difference 46

61 from the fraction 1 bladder in the high dose region. Figure 3.28: Patient Images for the Bladder CDA Method Outlier of %. Before and After Deformation. (Top to Bottom: Fraction 1, Fraction 2, Fraction 4, Bladder Contours From Fraction 2 and Fraction 4 Deformed onto the First Fraction Image with Fraction 1 Dose Distribution Displayed). The deformed fraction 2 bladder had a 1.24 mm maximum difference from the fraction 1 bladder in the high dose region. The deformed fraction 4 bladder had a 1.85 mm maximum difference from the fraction 1 bladder. 47

62 The DDA technique gives a mean of 1.88% 23.63% from the clinical standard. For these patients, the dose difference ranged between % and 37.87%. The dose distribution has an outlier on either side of the distribution for DDA. Images for the outlier of 37.87% difference is shown in Figure In this outlier, the clinical point dose estimated a larger dose delivered than the DDA method. This could be contributed to the large change in bladder size. The change in bladder size could cause different regions of high dose for fractions 2 and 4. Also, when deforming the dose, the dose may be shifted away from the bladder, in order to compensate for the shrinking of the dose in the bladder region. The images for the outlier of % difference from the clinical standard are shown in Figure In this case, the DDA method estimated a larger dose than the clinical point dose calculated. This could be contributed to the need for the fraction 2 and 4 bladder contours to expand anteriorly to match the fraction 1 bladder contour. This would shift some of the high dose region of the dose distribution farther into the bladder. This shift would create a higher point dose than calculated clinically. 48

63 Figure 3.29: Dose Distribution For the Bladder DDA Method Outlier of 37.87% Difference From Clinical Standard. (Top to Bottom: Fraction 1, Fraction 2, Fraction 4, Total Deformed Dose on Fraction 1 Image) 49

64 Figure 3.30: Dose Distribution for the Bladder DDA Method Outlier of % Difference From Clinical Standard. (Top to Bottom: Fraction 1, Fraction 2, Fraction 4, Total Deformed Dose on Fraction 1 Image) 50

65 3.5.2 Rectal Doses Figure 3.31: Dose Difference Expressed as a Percentage between the Dose Addition Techniques and the Clinical Standard for D 2cm 3 Rectum Metric. When deforming the rectum for the Syed-Neblett template applicator patients, the CDA showed an average dose difference of 2.67% 13.20%. The range from -5.71% to 35.92% is seen in Figure The patient images and structures before and after deformation for the outlier of 35.92% for CDA are shown in Figure This dose calculated by CDA method was less than the clinical point dose method. The deformed fraction 2 rectum had a 4.82 mm maximum difference from the fraction 1 rectum in the high dose region. The deformed fraction 4 rectum had a 5.11 mm maximum difference from the fraction 1 rectum in the high dose region. The deformed contours laid closer to the high dose than some parts of the fraction 1. The CDA method could underestimate the dose because of the change in the hot spot location. 51

66 The shape of the rectum varies greatly from fraction to fraction, due to differential filling, which could contribute to the difference from the clinical value. 52

67 Figure 3.32: Patient Images for the Rectum CDA Method Outlier of 35.92%. Before Deformation. (Top to Bottom: Fraction 1, Fraction 2, Fraction 4, Rectum Contours From Fraction 2 and Fraction 4 Deformed onto the First Fraction Image with Fraction 1 Dose Distribution Displayed). The deformed fraction 2 rectum had a 4.82 mm maximum difference from the fraction 1 rectum in the high dose region. The deformed fraction 4 rectum had a 5.11 mm maximum difference from the fraction 1 rectum in the high dose region. The DDA method has an average of 1.30% 18.24%. This method showed 53

68 dose differences ranging from % to 20.53%. Figure 3.31 shows that there are two outliers on the DDA method, but the main distribution appears to be settled around small dose differences form the clinical standard. The patient images for the positive outlier of 20.53% are shown in Figure3.33. This patient case had a dose less than the current clinical standard when using DDA. This may be due to the superior portion of the rectum located at different points for each fraction. The fraction 2 and fraction 4 contours have superior locations much higher than the fraction 1 contour. Additionally, the fraction 1 contour is thinner in the anterior to posterior direction compared to fraction 2. This could shift the dose anteriorly, which would move the high dose away from the rectum. The images for the rectum dose difference from the dose deformation and addition method outlier of % is shown in Figure This means that the clinical point dose was less than the deformed and accumulated dose calculation. This overestimation by the DDA method could be due to the deformation matrix transitioning the superior portion of fraction 2 and 4 contours to the location of the fraction 1 contour. This cause the dose to spread superiorly as well and then shift more dose into the rectum when accumulating. 54

69 Figure 3.33: Dose Distribution For the Rectum DDA Method Outlier of 20.53%. (Top to bottom: Fraction 1, Fraction 2, Fraction 4) 55

70 Figure 3.34: Dose Distribution For the Rectum DDA Method Outlier of 46.88% Difference from Clinical Standard. (Top to Bottom: Fraction 1, Fraction 2, Fraction 4, Total Deformed Dose on Fraction 1 Image) D 2cm 3 Structures for the Bladder and Rectum The location of the D 2cm 3 structure for the bladder for one Syed-Neblett applicator patient is shown in Figure An example of the D 2cm 3 locations for the Syed- Neblett template applicator on the rectum is shown in Figure These D 2cm 3 locations respective to the Syed-Neblett applicator without the interstitial needles are shown in Figure

71 Figure 3.35: Syed-Neblett Template Applicator Example for Location of the D 2cm 3 Structure for Bladder. 57

72 Figure 3.36: Syed-Neblett Template Applicator Example for Location of the D 2cm 3 Structure for Rectum. Figure 3.37: Syed-Neblett Applicator Example of D 2cm 3 Locations Rendering. D 2cm 3 locations for bladder and rectum are shown as white lines in the organ at risk. The interstitial needles are not shown. 58

73 3.5.4 Dice Similarity Coefficients of Deformed Structures The dice similarity coefficient for the bladder deformations, in Syed-Neblett template applicator patients, range from 0.53 to As seen in Figure 3.38, when the bladder size must change drastically, such as -400% of its original size, the bladder contours do not match the first fraction as well as those that had smaller changes and thus giving a lower dice similarity coefficient (Figure 3.39). However, the figure also shows that not all of the small volume changes are guaranteed to have a high dice similarity coefficient (Figure 3.40). For the rectum, the dice similarity coefficients vary between 0.75 and These do not appear to have any dependence on the change in volume size, which is shown in Figure Figure 3.38: Relationship between the Dice Similarity Coefficient and the Change in Bladder Volume when Deforming to the First Fraction. Table 3.5 shows the bladder volume for each fraction before and after deformation and the corresponding dice similarity coefficient for each Syed-Neblett template case. 59

74 Table 3.5: Bladder Volume and DSC Table; Syed-Neblett Cases. 60

75 Figure 3.39: Dice Similarity Coefficient vs Volume Outlier Patient Image for Bladder with Large Volume Change. (Top: Fraction 1 Image; Bottom: Fraction 4 Image) 61

76 Figure 3.40: Dice Similarity Coefficient vs Volume Outlier Patient Image for Bladder with Small Volume Change (Top: Fraction 1 Image; Bottom: Fraction 2 Image) Figure 3.41: Relationship between the Dice Similarity Coefficient and the Change in Rectum Volume when Deforming to the First Fraction. Table 3.6 shows the rectum volume for each fraction before and after deformation and the corresponding dice similarity coefficient for each Syed-Neblett template case. 62

77 Table 3.6: Rectum Volume and DSC Table; Syed-Neblett Cases. A comparison of all the dose differences for the Cylinder and Syed-Neblett applicators are presented in Figure The Syed-Neblett applicators have more outliers than the cylinder applicators. This figure shows that the CDA method for the rectum with the Cylinder applicator had the largest differences from the standard clinical point dose calculation. This could be contributed to using the clinical contours which were not consistent between fractions and therefore the contour deformation was challenged to overcome this inconsistency. 63