Browsing by Subject "Radiation therapy"

Introduction: RECA (Radiotherapy Enhanced with Cherenkov photo- Activation) is a novel treatment that induces a synergistic therapeutic effect by combining conventional radiation therapy with phototherapy using the anti-cancer and potentially immunogenic drug, psoralen. This work presents a novel method to investigate the therapeutic effect of RECA using rat brain slices and the agarose- based tissue equivalent material. Methods: 4T1 mCherry Firefly Luciferase mouse breast cancer cells are placed on the brain slice after exposed to psoralen solution. Taking fluorescent imaging of the brain slices every day after irradiation, an independent luciferase imaging was taken after the fifth fluorescence imaging. Using different imaging processing and analysis method to identify the cells. Result: Four analyzing method give different result about the fluorescence signal or luminescence signal. The overall trend of the fluorescence signal is rising over day, reaches the lowest point at 48 hours after irradiation. Control group (no radiation and no Cherenkov lights) has the lowest signal compared with other groups. The signal of brain slices with 4T1 cells exposed to psoralen solution is lower than that of brain slices without psoralen exposition. Conclusion: This work shows that rat brain slice can be used to simulate in vivo environment in exploring the therapeutic effect of RECA. Future work should focus on improving the image analyze method to better identify cells and noises.

In external beam radiation therapy (EBRT) for cancer patients, it is highly desirable to completely eradicate the cancerous cells for the purpose of improving the patient’s quality of life and increasing the patient’s likelihood of survival. However, there can be significant side effects when large regions of healthy cells are irradiated during EBRT, particularly for organs-at-risk (OARs). Due to the juxtaposition of the cancerous and non-cancerous tissue, trade-offs need to be made between target coverage and OAR sparing during treatment planning. For this reason, the treatment planning process can be posed as a multi-criterial optimization (MCO) problem, which has previously been studied extensively with several exact solutions existing specifically for radiation therapy. Typical MCO implementations for EBRT involve creating, optimizing, and calculating many treatment plans to infer the set of feasible best radiation doses, or the Pareto surface. However, each optimization and calculation can take 10-30 minutes per plan. As a result, generating enough plans to attain an accurate representation of the Pareto surface can be very time-consuming, particularly in higher-dimensions with many possible trade-offs.

The purpose of this study is to streamline the MCO workflow by using a machine-learning model to quickly predict the Pareto surface plan doses, rather than exactly computing them. The primary focus of this study focuses on the development and analysis of the dose prediction model. The secondary focus of this study is to develop new metrics for analyzing the similarity between different Pareto surface interpolations. The tertiary focus of this study is to investigate the feasibility of deliberately irradiating the epidural space in spine stereotactic radiosurgery (SRS), as well as estimate its potential effect on preventing tumor recurrence.

For the primary focus of this study, the model’s architecture proceeds as follows. The model begins by creating an initial dose distribution via an inverse fit of inter-slice and intra-slice PTV distance maps on a voxel-wise basis. The model proceeds by extracting three sets of transverse patches from all structure maps and the initialized dose map at each voxel. The model then uses the patch vectors as inputs for a neural network which updates and refines the dose initialization to achieve a final dose prediction. The primary motivation behind our model is to use our understanding of the general shape of dose distributions to remove much of the nonlinearity of the dose prediction problem, decreasing the difficulty of subsequent network predictions. Our model is able to take the optimization priorities into account during dose prediction and infer feasible dose distributions across a range of optimization priority combinations, allowing for indirect Pareto surface inference.

The model’s performance was analyzed on conventional prostate volumetric modulated arc therapy (VMAT), pancreas stereotactic body radiation therapy (SBRT), and spine stereotactic radiosurgery (SRS) with epidural space irradiation. For each of these treatment paradigms, the Pareto surfaces of many patients were thoroughly sampled to train and test the model. On all of these cases, our model achieved good performance in terms of speed and accuracy. Overfitting was shown to be minimal in all cases, and dose distribution slices and dose-volume histograms (DVHs) were shown for comparison, confirming the proficiency of our model. This model is relatively fast (0.05-0.20 seconds per plan), and it is capable of sampling the entire Pareto surface much faster than commercial dose optimization and calculation engines.

While these results were generally promising, the model achieved lower error on the prostate VMAT treatment plans compared to the pancreas SBRT and spine SRS treatment plans. This is likely due to the existence of heavier beam streaks in the stereotactic treatment plans which are generated by a sharper control of the delivered dose distribution. However, the Pareto surface errors were similar across all three cases, so these dose distribution errors did not propagate to the Pareto objective space.

The secondary focus of this study is the development and analysis of Pareto surface similarity metrics. The dose prediction model can be used to rapidly estimate many Pareto-optimal plans for quick Pareto surface inference. This could allow for a potentially significant increase in the speed at which Pareto surfaces are inferenced to provide treatment planning assistance and acceleration. However, previous investigations into Pareto surface analysis typically do not compare a ground truth Pareto surface with a Pareto surface prediction. Therefore, there is a need to develop a Pareto surface metric in order to evaluate the ability of the model to generate accurate Pareto surfaces in addition to accurate dose distributions.

To address these needs, we developed four Pareto surface similarity metrics, emphasizing the ability to represent distances between the interpolations rather than the sampled points. The most straightforward metric is the root-mean-square error (RMSE) evaluated between matched, sampled points on the Pareto surfaces, augmented by intra-simplex upsampling of the barycentric dimensions of each simplex. The second metric is the Hausdorff distance, which evaluates the maximum closest distance between the sets of sampled points. The third metric is the average projected distance (APD), which evaluates the displacements between the sampled points and evaluates their projections along the mean displacement. The fourth metric is the average nearest-point distance (ANPD), which numerically integrates point-to-simplex distances over the upsampled simplices of the Pareto surfaces. These metrics are compared by their convergence rates as a function of intra-simplex upsampling, the calculation times required to achieve convergence, and their qualitative meaningfulness in representing the underlying interpolated surfaces. For testing, several simplex pairs were constructed abstractly, and Pareto surfaces were constructed using inverse optimization and our dose prediction model applied to conventional prostate VMAT, pancreas SBRT, and spine SRS with epidural irradiation.

For the abstract simplex pairs, convergence within 1% was typically achieved at approximately 50 and 100 samples per barycentric dimension for the ANPD and the RMSE, respectively. The RMSE and the ANPD required approximately 50 milliseconds and 3 seconds to calculate to these sampling rates, respectively, while the APD and HD required much less than 1 millisecond. Additionally, the APD values closely resembled the ANPD limits, while the RMSE limits and HD tended to be more different. The ANPD is likely more meaningful than the RMSE and APD, as the ANPD’s point-to-simplex distance functions more closely represent the dissimilarity between the underlying interpolated surfaces rather than the sampling points on the surfaces. However, in situations requiring high-speed evaluations, the APD may be more desirable due to its speed, lack of subjective specification of intra-simplex upsampling rates, and similarity to the ANPD limits.

The tertiary focus of this study is the analysis of the feasibility of epidural space irradiation in spine SRS. The epidural space is a frequent site of cancer recurrence after spine SRS. This may be due to microscopic disease in the epidural space which is under-dosed to obey strict spinal cord dose constraints. We hypothesized that the epidural space could be purposefully irradiated to prescription dose levels, potentially reducing the risk of recurrence in the epidural space without increasing toxicity. To address this, we sought to analyze the feasibility of irradiating the epidural space in spine SRS. Analyzing the data associated with this study is synergistic to our MCO acceleration study, since the range of trade-offs between epidural space irradiation and spinal cord sparing represents an MCO problem which our dose prediction model may quickly solve.

Spine SRS clinical treatment plans with associated spinal PTV (PTVspine) and spinal cord contours, and prior delivered dose distributions were identified retrospectively. An epidural space PTV (PTVepidural) was contoured to avoid the spinal cord and focus on regions near the PTVspine. Clinical plan constraints included PTVspine constraints (D95% = 1800 cGy, D5% < 1950 cGy) and spinal cord constraints (Dmax < 1300 cGy, D10% < 1000 cGy). Prior clinical plan doses were mapped onto the new PTVepidural contour for analysis. Plans were copied and revised to additionally target the PTVepidural, optimizing PTVepidural D95% after meeting clinical plan constraints. Tumor control probabilities (TCPs) were estimated for the PTVepidural using a radiobiological linear-quadratic model of cell survival for both clinical and revised plans. Clinical and revised plans were compared according to their PTVepidural DVH distributions, D95% distributions, and TCPs.

Seventeen SSRS plans were identified and included in this study. Revised plan DVHs demonstrated higher doses to the epidural low-dose regions, with D95% improving from 10.96 Gy ± 1.76 Gy to 16.84 Gy ± 0.87 Gy (p < 10-5). Our TCP modeling set the clinical plan TCP average to 85%, while revised plan TCPs were all greater than 99.99%. Therefore, irradiating the epidural space in spine SRS is likely feasible, and purposefully targeting the epidural space in SSRS should increase control in the epidural space without significantly increasing the risk of spinal cord toxicity.

Purpose: GRID therapy (Spatially Fractionated RT) has the potential to amplify systemic anti-tumor immune effect. The optimal GRID design, radiation dosage and combination with immunotherapies are not well understood. In this work, we characterized two novel, high-resolution GRIDs of smaller width and spacing had been was previously employed at Duke University. By combining these GRIDs with anti-PD-1 immune checkpoint blockade, we investigated the efficacy of this combination therapy in a preclinical mouse model. This work has two main aims. First, to observe any anti-tumor response from GRID therapy that arises from sparing T cell lymphocytes in the valleys adjacent to high peak doses of radiation facilitating tumor antigen presentation. Second, to investigate the robustness and replicability of a previously published influential work (Markovsky et al., 2019) which demonstrated that hemi-irradiation can produce similar levels of tumor control as conventional radiation therapy [1]. This is one of the first studies we are aware of that combines mini-GRID treatment with immunotherapies. Methods: Prior to in vivo studies, two novel, high-resolution in-house mini-GRIDs were characterized using the Small Animal Radiation Research Platform (SARRP). To perform this characterization, the SARRP (225 kV,13 mAs) irradiated EBT3 film. Using the Epson 11000XL scanner, EBT3 films were scanned prior to and post irradiation. Median filters were applied to avoid artificially suppressing peak and valley dose distributions. A calibration curve was generated using irradiations of a known dosage to determine what dose was delivered to the high and low-dose regions of the GRIDs. From this, peak-to-valley dose ratios as well as output factors were be calculated. Then, two pilot studies were performed using SARRP to deliver RT to C57BL/6J mice with subcutaneous LLC1 (Lewis Lung Carcinoma) flank tumors. The first study tested the therapeutic efficacy of single dose radiation while the second investigated fractionated radiation utilizing the newer, high-resolution GRIDs. In the first study, mice were randomized to four groups: 15 Gy to an open 20 mm x 20 mm field (n=5), 15 Gy to a GRID with 1mm width and spacing (n=5), and 24 Gy to a GRID with 1mm width and spacing (n=5). For the second study, mice were randomized to four groups: an open 20 mm x 20 mm field (n=6), the same field irradiating only half the tumor (n=6) (following Markovsky et al., 2019), a GRID with 1 mm width and spacing (n=6) and a GRID with 254 µm width and spacing (n=7). For both in vivo studies, all mice in this study were treated with 200 μg of anti-PD-1 antibody prior to 15 Gy of RT (single AP field) on days 0, 3, and 6. Anti-PD-1 was then administered weekly until mice reached humane endpoint (>15 mm in any dimension or ulceration). Tumor growth was measured thrice weekly using digital calipers. Results: • Film Characterizations: The peak to valley dose ratios for the 254 µm and 152 µm GRIDS were 19.8 ± 0.7 and 9.37 ± 0.33 respectively. The output factors for these GRIDs were 0.62 ± 0.09 and 0.59 ± 0.03. • First in vivo Study: Tumor quadrupling times (days, ± SD) were: 8.94 ± 1.17 (open field, 15 Gy), 7.75 ± 0.91 (1mm GRID, 15 Gy) and 7.98 ± 1.08 (1mm GRID, 24 Gy). Mean survival times (days, ± SD) were: 16.00 ± 0.00 (open field, 15 Gy), 12.8 ± 1.09 (1mm GRID, 15 Gy and 24 Gy). None of these differences were statistically significant. The width of the valleys for the 254 µm GRID is 544 ± 33.94 µm and for the 152 µm GRID is 548 µm ± 31.57. Assuming a clinically that 100 cells with a diameter of 5µm represent a clinically relevant sample for irradiation, this is a sufficient area for irradiation. • Second in vivo Study: Tumor quadrupling times (days, ± SD) were: 12.8 ± 2.6 (open field), 8.4 ± 2.8 (hemi-irradiation), 9.7 ± 2.4 (1mm GRID), and 6.4 ± 4.4 (0.25 mm GRID). Mean survival times (days, ± SD) were: 14.2 ± 2.1 (open field), 12.2 ± 1.0 (hemi-irradiation), 11.3 ± 1.6 (1mm GRID), and 10.4 ± 2.2 (254 µm GRID). Compared to the open field, time to tumor quadrupling was lower in all groups, significantly so in the hemi-irradiated and 0.25 mm GRID groups (p<0.05). Both the hemi-irradiated and GRID groups showed significantly shorter mean survival times compared to conventional open-field treatment (p<0.05 for 1 mm GRID, p<0.01 for hemi-irradiation and 0.25 mm GRID). Conclusion: Two novel mini-GRIDs were successfully characterized using the SARRP for preclinical work, and sufficiently kept valley doses below 1.5 Gy for infiltrative T cell function [2] with peak doses greater than 15 Gy, thereby enabling tumor antigen presentation. However, neither single dose nor fractionated GRID therapy with anti-PD-1 improved tumor growth delay or survival in a preclinical LLC flank model. In contrast to published data with this model, hemi-irradiation worsened tumor control compared to conventional treatment. Our work, therefore, does support the conclusion drawn in the Markovsky paper that hem-irradiation provides comparable tumor control using hemi-irradiation to conventional treatment [1]. The development of new technologies such as FLASH radiotherapy may present new opportunities for future studies utilizing GRID therapy.

Purpose: To develop and implement novel methods of extreme spatially fractionated radiation therapy (including GRID therapy) and subsequent evaluation in pre-clinical mice trials investigating the potential of novel radiation treatments with potential for promoting anti-cancer immunogenic response.

Methods: Spatially fractionated GRIDs were designed and precision-milled from 3mm thick lead sheets compatible with mounting on a 225 kVp small animal irradiator (X-Rad). Three pencil-beam GRIDs created arrays of 1mm diameter beams, and three “bar” GRIDs created 1x20mm rectangular fields. GRIDs projected 20x20mm fields at isocenter and beamlets were spaced at 1, 1.25, and 1.5mm, respectively. Output factors, peak-to-valley ratios, and dose distributions were determined with Gafchromic film. The bar GRID with 1mm beamlet spacing (50:50 open:closed ratio) was selected for the pre-clinical trial. Soft-tissue sarcoma (p53/MCA) was transplanted into C57BL/6 mice’s flanks. Four treatment arms were compared: unirradiated control (n=18), conventional radiation therapy (n=16), GRID therapy (n=17), and hemi-irradiation (n=17) where one-half of the beam was blocked. All irradiated mice received a single fraction of 15 Gy to irradiated regions. To date, this is the first study to compare GRID treatment against conventional RT at the same dose.

Results: Very high peak-to-valley ratios were achieved (bar GRIDs: 11.9±0.9, 13.6±0.4, 13.8±0.5; pencil-beam GRIDs: 18.7±0.6, 26.3±1.5, 31.0±3.3). Pencil-beam GRIDs spared twice the number of intra-tumor immune cells as bar GRIDs but left more of the tumor untreated (2-3% vs 14-17% area receiving 95% prescription, respectively). Penumbra was halved when GRIDs were 50% closer to treatment isocenter. The GRID selected for mouse trials was capable of sparing approximately 15% of intra-tumor CD8+ and CD4+ T cells. Preliminary results indicate mean times to tumor quintupling were: 12, 13, 14, and 20 days for unirradiated, GRID, hemi-irradiated, and conventional treatment groups, respectively. To date, all tumors have quintupled except for nine in the AP control group.

Conclusions: Peak-to-valley ratios with kV grids were substantially superior to MV grids, which historically achieve ratios between 2.5 and 6.5. In data collected to date, GRID and hemi-irradiation did not significantly delay tumor growth as compared to an unirradiated control (P = 0.122 and P = 0.2437, respectively, P-values from logrank analysis). Differences between GRID and hemi-irradiation were not statistically significant (P = 0.5257). To date, the AP control group has performed significantly better than all other groups (P<0.001). These results do not corroborate the success of hemi-irradiation in Markovsky et al. 2019. GRID treatments may be more effective if a substantially higher dose and/or multiple fractions were employed.

Purpose: To investigate the inter-patient and inter-sequence variation in liver tumor contrast in MRI and the feasibility of improving the liver tumor contrast by using an in-house developed multi-source adaptive fusion method for use in MRI-based treatment planning.

Methods and Materials: MR-images from 29 patients were retrospectively reviewed in this study. The imaging sequences acquired by a 1.5T GE and 3T Siemens MR scanner consisted of T1-w, T1-w, Post C, T2-w, T2/T1-w, and DWI. Using an in-house developed MSAF algorithm, we created fused images for a smaller subset of 12 patients using T1-w, T2-w, T2/T1-w, and DWI as inputs. Two fusion-images were obtained for each patient by implementing either an input-driven or output-driven fusion optimization method. Once a fusion-image was obtained an analysis was performed on each original image, and the fusion-image for each patient to calculate the tumor-to-tissue contrast-to-noise ratio(CNR) by contouring the tumor and a liver background-region(BG) in a homogeneous region of the liver using this in-house algorithm. CNR was calculated by (Itum-IBG)/SDBG, where Itum and IBG are the mean values of the tumor and the BG respectively, and SDBG is the standard deviation of the BG. To assess variation in tumor to tissue CNR for each image type an inter-patient coefficient-of-variation(CV) was calculated across all patients, as well as an inter-sequence CV. CV was calculated using the following: CV = σ/µ, where σ and µ are the standard deviation, and mean CNR for a single image sequence, respectively. These values were calculated for the original sequence types and fusion-images and compared.

Results: Our results from the 29 patients showed large inter-patient and inter-sequence variability, ranging from 86.90% to 67.03%, and 134.67% to 1.22% respectively. The T1-w, T1-w, Post Contrast, T2-w, T2/T1-w, DWI, and CT CV was 85.25%, 84.11%, 67.03%, 81.78%, 86.90%, and 74.30% respectively. Tumor CNR ranged from 0.95 to 4.47 with mean (± SD) CNR for T1-w, T1-w, Post Contrast, T2-w, T2/T1-w, DWI, and CT of 1.90 (±1.60), 2.12 (±1.42), 3.59 (±2.94), 1.95 (±1.70), 4.47 (±3.32), and 0.95 (±0.81) respectively. In the smaller subset of 12 patients, our results show a reduction in the inter-patient CV when using the in-house algorithm to obtain a tumor enhanced – fusion image. The inter-patient CV for T1-w, T2-w, T2/T1-w, DWI, Balanced Anatomy – Fusion, and Tumor Enhanced – Fusion was 94.16%, 112.73%, 105.69%, 124.23%, and 67.94% respectively. Tumor-CNR was significantly enhanced for each patient when using the in-house algorithm to obtain a tumor-enhanced image. The mean (± SD) CNR for T1-w, T2-w, T2/T1-w, Balanced Anatomy – Fusion, and Tumor Enhanced – Fusion was 2.11 (±1.99), 3.89 (±4.38), 3.71 (±3.92), 5.73 (±7.12), and 17.01 (±11.55) respectively.

Conclusion: The in-house multi-source adaptive fusion algorithm has the potential to increase the liver tumor contrast, as well as, improve the consistency for use in MRI based radiation therapy treatment planning.

The use of positron emission tomography (PET) in radiation therapy has continued to grow, especially since the development of combined computed tomography (CT) and PET imaging system in the early 1990s. Today, the biggest use of PET-CT is in oncology, where a glucose analog radiotracer is rapidly incorporated into the metabolic pathways of a variety of cancers. Images representing the in-vivo distribution of this radiotracer are used for the staging, delineation and assessment of treatment response of patients undergoing chemotherapy or radiation therapy. While PET offers the ability to provide functional information, the imaging quality of PET is adversely affected by its lower spatial resolution. It also has unfavorable image noise characteristics due to radiation dose concerns and patient compliance. These factors result in PET images having less detail and lower signal-to-noise (SNR) properties compared to images produced by CT. This complicates the use of PET within many areas of radiation oncology, but particularly the delineation of targets for radiation therapy and the assessment of patient response to therapy. The development of segmentation methods that can provide accurate object identification in PET images under a variety of imaging conditions has been a goal of the imaging community for years. The goal of this thesis are to: (1) investigate the effect of filtering on segmentation methods; (2) investigate whether combining individual segmentation methods can improve segmentation accuracy; (3) investigate whether the consensus volumes can be useful in aiding physicians of different experience in defining gross tumor volumes (GTV) for head-and-neck cancer patients; and (4) to investigate whether consensus volumes can be useful in assessing early treatment response in head-and-neck cancer patients.

For this dissertation work, standard spherical objects of volumes ranging from 1.15 cc to 37 cc and two irregularly shaped objects of volume 16 cc and 32 cc formed by deforming high density plastic bottles were placed in a standardized image quality phantom and imaged at two contrasts (4:1 or 8:1 for spheres, and 4.5:1 and 9:1 for irregular) and three scan durations (1, 2 and 5 minutes). For the work carried out into the comparison of images filters, Gaussian and bilateral filters matched to produce similar image signal to noise (SNR) in background regions were applied to raw unfiltered images. Objects were segmented using thresholding at 40% of the maximum intensity within a region-of-interest (ROI), an adaptive thresholding method which accounts for the signal of the object as well as background, k-means clustering, and a seeded region-growing method adapted from the literature. Quality of the segmentations was assessed using the Dice Similarity Coefficient (DSC) and symmetric mean absolute surface distance (SMASD). Further, models describing how DSC varies with object size, contrast, scan duration, filter choice and segmentation method were fitted using generalized estimating equations (GEEs) and standard regression for comparison. GEEs accounted for the bounded, correlated and heteroscedastic nature of the DSC metric. Our analysis revealed that object size had the largest effect on DSC for spheres, followed by contrast and scan duration. In addition, compared to filtering images with a 5 mm full-width at half maximum (FWHM) Gaussian filter, a 7 mm bilateral filter with moderate pre-smoothing (3 mm Gaussian (G3B7)) produced significant improvements in 3 out of the 4 segmentation methods for spheres. For the irregular objects, time had the biggest effect on DSC values, followed by contrast.

For the study of applying consensus methods to PET segmentation, an additional gradient based method was included into the collection individual segmentation methods used for the filtering study. Objects in images acquired for 5 minute scan durations were filtered with a 5 mm FWHM Gaussian before being segmented by all individual methods. Two approaches of creating a volume reflecting the agreement between the individual methods were investigated. First, a simple majority voting scheme (MJV), where individual voxels segmented by three or more of the individual methods are included in the consensus volume, and second, the Simultaneous Truth and Performance Level Estimation (STAPLE) method which is a maximum likelihood methodology previously presented in the literature but never applied to PET segmentation. Improvements in accuracy to match or exceed the best performing individual method were observed, and importantly, both consensus methods provided robustness against poorly performing individual methods. In fact, the distributions of DSC and SMASD values for the MJV and STAPLE closely match the distribution that would result if the best individual method result were selected for all objects (the best individual method varies by objects). Given that the best individual method is dependent on object type, size, contrast, and image noise and the best individual method is not able to be known before segmentation, consensus methods offer a marked improvement over the current standard of using just one of the individual segmentation methods used in this dissertation.

To explore the potential application of consensus volumes to radiation therapy, the MJV consensus method was used to produce GTVs in a population of head and neck cancer patients. This GTV and one created using simple 40% thresholding were then available to be used as a guidance volume for an attending head and neck radiation oncologist and a resident who had completed their head and neck rotation. The task for each physician was to manually delineate GTVs using the CT and PET images. Each patient was contoured three times by each physician- without guidance and with guidance using either the MJV consensus volume or 40% thresholding. Differences in GTV volumes between physicians were not significant, nor were differences between the GTV volumes regardless of the guidance volume available to the physicians. However, on average, 15-20% of the provided guidance volume lay outside the final physician-defined contour.

In the final study, the MJV and STAPLE consensus volumes were used to extract maximum, peak and mean SUV measurements in two baseline PET scans and one PET scan taken during patients' prescribed radiation therapy treatments. Mean SUV values derived from consensus volumes showed smaller variability compared to maximum SUV values. Baseline and intratreatment variability was assessed using a Bland-Altman analysis which showed that baseline variability in SUV was lower than intratreatment changes in SUV.

The techniques developed and reported in this thesis demonstrate how filter choice affects segmentation accuracy, how the use of GEEs more appropriately account for the properties of a common segmentation quality metric, and how consensus volumes not only provide an accuracy on par with the single best performing individual method in a given activity distribution, but also exhibit a robustness against variable performance of individual segmentation methods that make up the consensus volume. These properties make the use of consensus volumes appealing for a variety of tasks in radiation oncology.

The purpose of this study was to create a DVH quality index that can be used as a comparison tool between two separate plans and as a clinical workflow tool to improve plan quality resulting in better patient outcome. To create this DVH quality index, physician preference was used as the basis of the dose-volume relationship quantification rather than biological models since physicians are the ones who drive plan quality within in a clinic. An intra-patient observership study was created to gather the qualitative and quantitative from radiation oncologists who ranked a set of plans of varying plan quality from a specific patient. The qualitative data gave rise to the formation of the algorithm to produce a DVH quality index while the quantitative data drove the weighting factors within the algorithm. The intra-patient study validated the algorithms ability to determine the best DVH among separate plans from a specific patient. An inter-patient study was then introduced to validate the DVH quality index across the spectrum of scores given by the algorithm by comparing the algorithm's rank list with the oncologists' rank lists. These studies used spearman rank correlation tests to compare the rank lists between the algorithm and the oncologists. The perfect index that the algorithm can calculate is 10. Subsequently any penalization that occurs within the DVH will be subtracted away from the score of 10 with no bottom limit. For the intra-patient study, the mean correlation coefficient of our group's algorithm with the oncologists is 0.726 and the mean correlation coefficient of the oncologists with each other oncologist is 0.564. In the inter-patient study, the correlations proved to be stronger where the mean correlation coefficient of the algorithm with the oncologists is 0.822 and the mean correlation coefficient of the oncologists with each other is 0.699. Since our mean correlation coefficients with the oncologists for the intra-patient and the inter-patient study is higher than the mean correlation coefficient of the oncologists with each other, we can state that we represent a general oncologist within the hospital system when ranking DVHs.

Approximately 50% of all patients with cancer receive radiation therapy at some point during the course of their illness. Despite advances in radiation delivery and treatment planning, normal tissue toxicity often limits the ability of radiation to eradicate tumors. The tumor microenvironment consists of tumor cells and stromal cells such as endothelial cells that contribute to tumor initiation, progression and response to therapy. Although endothelial cells can contribute to normal tissue injury following radiation, the contribution of stromal cells to tumor response to radiation therapy remains controversial. To investigate the contribution of endothelial cells to the radiation response of primary tumors, we have developed the technology to contemporaneously mutate different genes in the tumor cells and stromal cells of a genetically engineered mouse model of soft tissue sarcoma. Using this dual recombinase technology, we deleted the DNA damage response gene Atm in sarcoma and heart endothelial cells. Although deletion of Atm increased cell death of proliferating tumor endothelial cells, Atm deletion in quiescent endothelial cells of the heart did not sensitize mice to radiation-induced myocardial necrosis. In addition, the ATM inhibitor NVP-BEZ235 selectively radiosensitized primary sarcomas, demonstrating a therapeutic window for inhibiting ATM during radiation therapy. Sensitizing tumor endothelial cells to radiation by deleting Atm prolonged tumor growth delay following a non-curative dose of radiation, but failed to increase local control. In contrast, deletion of Atm in tumor parenchymal cells increased the probability of tumor eradication. These results demonstrate that tumor parenchymal cells rather than endothelial cells are the critical targets that regulate tumor eradicaiton by radiation therapy.

Oropharyngeal Squamous Cell Carcinoma (OPSCC) is by far the most predominant form of head and neck cancer in the United States. The survival rate for OPSCC is very high, which, while fortunate, yields many patients who are left with the late term toxicities consequent of their treatment. This project aimed to use patient-reported outcome (PRO) data from two sources – the PRO-CTCAE and the QLQ-C30 – along with the dosimetric data of patients that have already been treated, in order to characterize retrospectively a relationship between patient dosimetric data and the severity of response of PRO data. In particular, PRO data was used as a way to characterize the severity of patient-experienced xerostomia and dysphagia. Additionally, this data was used to fit the radiobiological parameters for two normal tissue complication probability (NTCP) models: the Lyman-Kutcher-Burman (LKB) model, and the Relative Seriality (RS) model. Overall, it was found that the PRO-CTCAE data was more robust than the QLQ-C30 data in its characterization. Based on the PRO-CTCAE data, the V52 (volume which receives at least 52 Gy) of the combined constrictors and the V59 of the superior pharyngeal constrictor show the strongest relationship with patient-reported dysphagia. Additionally, the V27 of the contralaterals and the V12 of the contralateral parotid show the strongest relationship with patient-reported xerostomia. Furthermore, it was found that the dose response curves for both NTCP models fit the data with similar accuracy.

Purpose: Radiotherapy Enhanced by Cherenkov photo-Activation (RECA) is a novel radiation treatment method that seeks an anti-cancer effect with the introduction of a psoralen compound administered for treatment. The goal of the RECA method is to enhance standard radiation therapy treatments with the addition of psoralen being photo-activated by Cherenkov radiation that is generated during radiotherapy. The purpose of this work is to investigate the effectiveness of RECA on 4T1 mCherry FLuc breast cancer cells seeded on a psoralen-baked-agarose-based rat brain slice.Methods: A previously established CellProfiler pipeline, developed in our lab by Holden et al., was used to assess tumor burden on rat brain slices used for a tissue-equivalent medium for cell culturing. The CellProfiler pipeline was implemented on images of 4T1 breast cancer cells growing over the course of four to five days post-treatment to measure the average intensity of fluorescing cells. Prior to the RECA experiment, multiple preparatory experiments were conducted to refine and optimize experimental techniques. The first preparatory experiment tested the possibility of a plate reader bias effect, i.e., signal from nearby wells contributing to signal of other wells, seen during measurements of cell luminescence within individual wells of a clear-bottom 96-well plate. A CellTiter-Glo endpoint readout was taken 48-hours post-treatment for an endpoint measure to assess the if there was any added signal from nearby wells in the clear-bottom plates. The next experiment tested whether fractionation of dose was feasible and preferrable to single dose treatment by irradiating 4T1 mCherry Fluc cells with 2 Gy and 4 Gy of kV radiation with and without fractionation. An endpoint CellTiter-Glo readout was conducted 72 hours post-treatment to assess cell viability between the treatment plans. Additional preparatory experiments investigated whether psoralen-doped agarose was an effective method for cell loading. A 30 µM AMT-baked agar base was placed in half of the wells in plates with 4T1 mCherry Fluc cells seeded on brain slices on top of the agar. One plate received no treatment and one plate received treatment of 365 nm UVA, and an endpoint Firefly Luciferase reporter assay was conducted 48 hours post-treatment to assess cell viability between the conditions. For the RECA experiment, five 12-well plates, each containing 1 cm of agar with a 400 µm thick coronal slice of rat brain tissue, were given one of five conditions of treatment: no treatment, 4.95 Gy of fractionated kV or MV treatment, or 4.95 Gy of whole kV of MV treatment. Each plate condition consisted of six wells containing AMT-baked agar and six wells containing a standard agar base. After irradiation, images were taken of each of the plates for each day over the course of five days five days with a Zeiss Lumar microscope. The microscope was equipped with a rhodamine filter to analyze the luminescence readings from each well for assessment of cell viability. Results: The preparatory experiments all yielded results that allowed for development of the RECA experiment procedure. Investigation of the plate reader effect showed that background signal from nearby wells was not leaking into well signal readout, with all wells having nearly consistent signal throughout all the wells. Fractionating the dose was found to be preferable because it decreased cell viability less than delivering all dose at once, which floored cell viability. Testing psoralen-doped agar demonstrated that this is an effective delivery method for psoralen to intercalate with cells. The RECA experiment utilizing kV and MV whole dose conditions allowed comparison between irradiations with and without a fractionation scheme. The Firefly Luciferase reporter assay signal for the MV treatment conditions showed less cell viability than the Dark control conditions for both AMT and DMSO. Additionally, the whole dose MV conditions demonstrated a more pronounced decrease in cell viability than the fractionated MV conditions, as expected. The CellProfiler analysis demonstrated the same trends with the whole dose MV AMT condition (8.54 ± 0.99-fold increase) and whole dose MV DMSO condition (11.80 ± 0.70-fold increase) demonstrating less cell viability than the Dark AMT (13.41 ± 0.83-fold increase) and Dark DMSO (14.11 ± 0.62-fold increase). Interestingly, there was not a significant difference in cell viability seen between the fractionated and whole dose conditions. Conclusions: The procedural techniques developed for the analysis of the RECA effect during the preparatory experiments ruled out a plate reader effect and demonstrated that introducing fractionation and psoralen-baked agar is effective. The testing of the fractionation scheme used for kV irradiations proved to be sufficient for decreasing cell viability without killing all the cells. Additionally, the testing of the psoralen-baked agarose slabs proved to be an adequate psoralen delivery method when compared to methods that used cells suspended in psoralen treated media in prior studies. When these changes to the procedure were introduced together during MV irradiations, the RECA effect did not clearly replicate the results demonstrated during kV irradiations in the preparatory experiments. Further investigation is required to confirm and validate the RECA effect generated during radiotherapy.

Mice are the predominant animal model used in radiation therapy research for investigating radiobiological kinetics and evaluating new therapeutics to achieve a higher therapeutic ratio in the clinic. A novel carcinogen-induced and genetically engineered head and neck squamous cell carcinoma mouse model was developed at Duke to study head and neck cancer, one of the most widely spread cancers in the world. However, platforms that are able to perform precise and reproducible radiation therapy on these mice to mimic human radiation therapy are lacking. To address this issue, a platform based on the X-RAD 225Cx orthovoltage irradiator was developed. 3D printing technique was used to generate imaging phantoms, immobilization devices, and blocks. A simulation was conducted to optimize imaging protocol. Results were verified on the measurement on both the 3D-printed phantom and the actual mouse. Prior to irradiation, mice were placed on the immobilization device in a supine position, and the isocenter was determined by the position of the device since the irradiator does not have a laser localizer system. The performance of the immobilization was obtained by scanning several mice separately at various time points, ranging from several hours post-imaging to two months post-imaging. In order to make up the deficiency that irradiator only have rectangular and circular collimators which cannot provide moderate protection for organs at risk. Blocks with 3% transmission were designed based on the contours of central nervous system by a state-of-art program, BlockGen.

A protocol was developed for immobilization and image acquisition. 60 kVp was found to give the highest contrast of iodine, so it was set as the tube voltage for image acquisition. The deviations of positioning, i.e. the same mouse in separate scanning, are measured as 0.22±0.44 mm in LR axis, 0.15±0.30 mm in PA axis, and -0.24±0.25 mm in IS axis. Blocks with a 1.5 mm margin which can shield brain and spinal cord even in the worst case, were printed for opposed lateral beams; they were verified on fluoroscopy.

The block system was modified to eliminate potential human errors. Comparison on brain and spinal cord among different mice showed the largest deviation in 2.6 mm, however, with manually selection of the middle one, 1.5 mm margin is enough to shield central nervous system. Indicating that a generic block could be used in the experiment that does not require a very accurate treatment. The generic block can significant save time and effort for preclinical radiation treatment experiment. In this study, a platform that is capable of enhancing contrast imaging and allowing precise radiation therapy to be performed on genetically-engineered mice with head and neck cancer has been developed. This paves the way for more accurate head and neck mice model radiation therapy studies. In addition, the platform could be used in other types of preclinical studies.

Radiation-induced pulmonary toxicity poses a serious challenge and limiting factor in delivering a sufficient amount of dose to eradicate thoracic tumors without compromising lung function. Functional avoidance radiation therapy (RT) using lung ventilation mapping techniques would allow for preferential avoidance of functional lung tissue during radiotherapy and potentially reduce RT-induced lung injuries. Additionally, lung ventilation is also a key metric to assess lung function in patients with pulmonary diseases such as asthma, pulmonary embolism, cystic fibrosis, and chronic obstructive pulmonary disease (COPD). In contrast to global pulmonary function tests such as spirometry, ventilation images provide a regional measure of pulmonary function. Conventional methods for lung ventilation imaging include gamma camera scintigraphy and positron emission tomography scan after inhaling a gaseous radionuclide, as well as hyperpolarized (HP) gas magnetic resonance imaging (MRI) using Helium-3 and Xenon-129 as imaging contrast. Recently, a new method has been proposed in which deformable image registration (DIR) is performed on a pair of anatomical lung images at different respiratory phases to obtain the displacement vector field (DVF) between both phases, and generate a lung ventilation map from the Jacobian Determinant of the DVF. This DIR-based method is advantageous in its high image resolutions and simpler imaging procedures making it a more feasible option for implementation into the clinical workflow. However, current DIR-based lung ventilation methods have been largely hampered due to two major deficiencies: 1) current DIR algorithms are morphologically based, lacking of sufficient physiological realism and thus resulting in erroneous calculations of lung ventilation; and 2) there is a lack of validation of the DIR-based lung ventilation calculation against clinical ground truth, as well as large uncertainties and variations among different DIR algorithms. The long-term goal of this proposal is to develop the necessary tools and metrics for validation and testing of DIR-based lung ventilation mapping techniques to contribute to their clinical implementation in advanced radiotherapy of lung cancer and diagnosis of obstructive pulmonary diseases. The objective of the proposed research is to develop digital thoracic phantoms from physiologically-plausible lung motion models as a valuable tool for validation of DIR algorithms, and evaluate deformation-based lung ventilation mapping techniques against reference HP gas MRI ventilation images. The specific aims of this dissertation are the following. (1) Develop digital thoracic phantoms based on physiological modeling of respiratory motion from hyperpolarized gas tagging MRI. (2) Investigate the differences between HP gas tagging-based, DIR-based, and HP MRI ventilation mapping methods. (3) Evaluate and compare the sensitivity to deformation changes of ventilation and strain as lung functionality metrics.

This study investigates a unique dataset which includes three types of MR images acquired using a novel hybrid technique in a single breath-hold maneuver including a HP Helium-3 ventilation image, a pair of proton MR images, and a pair of HP Helium-3 tagging images at end of inhalation (EOI) and end of exhalation (EOE).

In order to create a physiologically plausible lung motion model, we used the novel HP gas tagging MRI technique. The tagged elements in the 3-dimensional (3D) tagged grid pattern are essentially ~500 evenly distributed landmarks throughout the entire lung area. Therefore, the displacement vector field calculated by tracking their motion from the EOI to the EOE phases provided us with a true lung deformation model which is physiologically-based.

The respiratory motion model was utilized to evaluate DIR-based displacement vector fields. The mean absolute DVF differences were found to be 8.2 mm for Subject 1, 7.5 mm for Subject 2, 5.6 mm for Subject 3, and 8.8 mm for Subject 4. These results show that there can be significant differences in DVF when performing a DIR compared to the respiratory motion models created from the tagged elements’ displacement.

The thoracic motion model was then created through a combination of the DIR-based DVF to model the deformation outside the lungs from the registration of proton MR images, and the tagging-based DVF to model deformation inside the lungs using the manually measured DVF from the tagging MR images.

The next part of this dissertation focused on investigating a DIR- based lung ventilation mapping technique using proton MR images by evaluating its correlation with hyperpolarized Helium-3 gas ventilation MRI reference images which provide a ground-truth measure of lung ventilation. Correlation between the reference ventilation images and ventilation maps computed from HP gas tagging MRIs, which provide ground-truth lung deformation, was also investigated. This is the first study, to our knowledge, to investigate three types of ventilation maps that are all MR-based. Furthermore, all images/data used in our evaluation are acquired during one same breath hold maneuver, eliminating the uncertainties associated with reproducibility of the respiratory cycle, patient positioning, and finding the spatial correspondence between the ventilation maps being evaluated.

The results of the spatial comparison between the DIR-based and reference ventilation images showed moderate to strong spearman correlation coefficients which are higher than many previous ventilation evaluation studies in the literature This may also be due to the fact that the images in this study were acquired during the same breath-hold and therefore inherently co-registered. The tagging-based ventilation maps, which are independent of the accuracy of any DIR algorithms, showed very similar spatial correlations to the reference images compared to the DIR-based ventilation maps. This proves the potential of the Jacobian ventilation calculation method which assumes that local volume changes are an appropriate lung ventilation surrogate. As more RT clinics incorporate MR imaging for patient simulation, and contouring for treatment planning, this study shows the feasibility of utilizing MR images for DIR-based ventilation calculations.

In the final part of this dissertation, we investigate lung strain as an additional metric to assess respiratory mechanics. We evaluation the sensitivity to deformation changes of both ventilation and strain as lung functionality metrics by comparing both metrics’ sensitivity to changes in displacement vector fields using Hyperpolarized He-3 Tagging MRI data. This study utilized physiologically-based respiratory motion models from three subjects to assess the sensitivity of lung strain and lung ventilation by introducing a number of modifications to the DVF, generating new lung function maps, and investigating how much each of these lung function metrics were affected. Lung strain was computed voxel-wise from the gradient of the tagging-based displacement. Through this algorithm, we obtained a 3x3 tensor that directly measures both the magnitude and direction of the lung deformation and then determines the three principal strains. For the lung ventilation calculation, we used the previously described voxel-by-voxel algorithm which was based on computing the Jacobian Determinant of the tagging-based DVF to determine the local volume changes. These results show much larger mean absolute percent differences between original and modified ventilation maps compared to the principal strain maps for all tests performed in this study; ranging from an average of 49.5 to 2743.7% for ventilation and 30.6 to 650.0% for strain among the 3 subjects. The principal strain maps showed much smaller average standard deviations between subjects. We found that Tagging-based ventilation maps calculated through the Jacobian of the DVF might be more sensitive to deformation changes compared to the lung strain maps, showing much larger mean absolute percent differences between the original and modified maps. This could indicate two things, while ventilation might be more sensitive to smaller deformation changes which could be an advantage, this could also indicate that it is more sensitive to small errors or uncertainties in the DVF which could make the calculation more unstable compared to strain.

In conclusion, this dissertation utilized a unique hybrid MR image acquisition method to present: the development of valuable physiologically-based respiratory motion models and DIR validation tools using novel tagging MR images; the first all MR-based evaluation of deformation-based ventilation mapping techniques, against reference HP gas He-3 MR ventilation images with improved spatial correspondence; and the investigation of lung strain as an additional metric to assess lung function.

The need for clinically intuitive metrics for patient-specific quality assurance in radiation therapy has been well-documented (Zhen, Nelms et al. 2011). A novel transform method has shown to be effective at converting full-density 3D dose measurements made in a phantom to dose values in the patient geometry, enabling comparisons using clinically intuitive metrics such as dose-volume histograms (Oldham et al. 2011). This work investigates the transform method and compares its calculated dose-volume histograms (DVHs) to DVH values calculated by a Delta4 QA device (Scandidos), marking the first comparison of a true 3D system to a semi-3D device using clinical metrics. Measurements were made using Presage 3D dosimeters, which were readout by an in-house optical-CT scanner. Three patient cases were chosen for the study: one head-and-neck VMAT treatment and two spine IMRT treatments. The transform method showed good agreement with the planned dose values for all three cases. Furthermore, the transformed DVHs adhered to the planned dose with more accuracy than the Delta4 DVHs. The similarity between the Delta4 DVHs and the transformed DVHs, however, was greater for one of the spine cases than it was for the head-and-neck case, implying that the accuracy of the Delta4 Anatomy software may vary from one treatment site to another. Overall, the transform method, which incorporates data from full-density 3D dose measurements, provides clinically intuitive results that are more accurate and consistent than the corresponding results from a semi-3D Delta4 system.

Radiation has been utilized in medical domain for multiple purposes. Treating cancer using radiation has increasing popularity during the last century. Radiation beam is directed to the tumor cells while the surrounding healthy tissue is attempted to be avoided. Radiation therapy treatment planning serves the goal of delivering high concentrated radiation to the treatment volume while minimizing the normal tissue as much as possible. With the advent of more sophisticated delivery technology, treatment planning time increases over time. In addition, the treatment plan quality relies on the experience of the planner. Several computer assistance techniques emerged to help the treatment planning process, among which knowledge-based planning (KBP) has been successful in inverse planning IMRT. KBP falls under the umbrella of Knowledge Discovery in Databases (KDD) which originated in industry. The philosophy is to extract useful knowledge from previous application/data/observations to make predictions in the future practice. KBP reduces the iterative trial-and-error process in manual planning, and more importantly guarantees consistent plan quality. Despite the great potential of treatment planning KDD (TPKDD), three major challenges remain before TPKDD can be widely implemented in the clinical environment: 1. a good knowledge model asks for sufficient amount of training data to extract useful knowledge and is therefore less efficient; 2. a knowledge model is usually only applicable for the specific treatment site and treatment technique and is therefore less generalizable; 3. a knowledge model needs meticulous inspection before implementing in the clinic to verify the robustness.

This study aims at filling in the niche in TPKDD and improves current TPKDD workflow by tackling the aforementioned challenges. This study is divided into three parts. The first part of the study aims to improve the modeling efficiency by introducing an atlas-based treatment planning guidance. In the second part of the study, an automated treatment planning technique for whole breast radiation therapy (WBRT) is proposed to provide a solution for the area where TPKDD has not yet set foot on. In the third part of the study, several topics related to the knowledge model quality are addressed, including improving the model training workflow, identifying geometric novelty and dosimetric outlier case, building a global model and facilitating incremental learning.

I. Improvement of the modeling efficiency. First, a prostate cancer patient anatomy atlas was established to generate 3D dose distribution guidance for the new patient. The anatomy pattern of the prostate cancer patient was parameterized with two descriptors. Each training case was represented in 2D feature space. All training cases were clustered using the k-medoids algorithm. The optimal number of clusters was determined by the largest average silhouette width. For the new case, the most similar case in the atlas was identified and used to generate dose guidance. The anatomy of the atlas case and the query case was registered and the deformation field was applied to the 3D radiation dose of the atlas case. The deformed dose served as the goal dose for the query case. Dose volume objectives were then extracted from the goal dose to guide the inverse IMRT planning. Results showed that the plans generated with atlas guidance had similar dosimetric quality as compared to the clinical manual plans. The monitor units (MU) of the auto plan were also comparable with the clinical plan. Atlas-guided radiation therapy has proven to be effective and efficient in inverse IMRT planning.

II. Improvement of model generalization. An automatic WBRT treatment planning workflow was developed. First of all, an energy selection tool was developed based on previous single energy and dual energy WBRT plans. The DRR intensity histograms of training cases were collected and the principal component analysis (PCA) was performed to reduce the dimension of the histogram. First two components were used to represent each case and the classification was performed in the 2D space. This tool helps new patient to select appropriate energy based on the anatomy information. Secondly, an anatomy feature based random forest (RF) model was proposed to predict the fluence map for the patient. The model took the input of multiple anatomical features and output the fluence intensity of each pixel within the fluence map. Finally, a physics rule based method was proposed to further fine tune the fluence map to achieve optimal dose distribution within the irradiated volume. Extra validation cases were tested on the proposed workflow. Results showed similar dosimetric quality between auto plan and clinical manual plan. The treatment planning time was reduced from between 1-4 hours for the manual planning to within 1 minute for the auto planning. The proposed automatic WBRT planning technique has proven to be efficient.

III. Rapid learning of radiation therapy KBP. Several topics were analyzed in this part of the study. First of all, a systematic workflow was established to improve the KBP model quality. The workflow started with identifying geometric novelty case using the statistical metric “leverage”, followed by removing the novelty case. Then the dosimetric outlier was identified using studentized residual and then cleaned. The cleaned model was compared with the uncleaned model using the extra validation cases. This study used pelvic cases as an example. Results showed that the existence of novelty and outlier cases did degrade the model quality. The proposed statistical tools can effectively identify novelty and outlier cases. The workflow is able to improve the quality of the knowledge-based model.

Secondly, a clustering-based method was proposed to identify multiple geometric novelty cases and dosimetric outlier cases at the same time. One class support vector machine (OCSVM) was applied to the feature vectors of all training cases to generate one class of inliers while cases falling out of the frontier belonged to the novelty case group. Once the novelty cases were identified and cleaned, the robust regression followed by outlier identification (ROUT) was applied to all remaining cases to identify dosimetric outliers. A cleaned model was trained with the novelty and outlier free case pool and was tested using 10 fold cross validation. Initial training pool included intentionally added outlier cases to evaluate the efficacy of the proposed method. The model prediction on the inlier cases was compared with that of novelty and outlier cases. Results showed that the method can successfully identify geometric novelty and dosimetric outliers. The model prediction accuracy between the inliers and novelty/outliers was significantly different, indicating different dosimetric behavior between two groups. The proposed method proved to be effective in identifying multiple geometric novelty and dosimetric outliers.

Thirdly, a global model using the model tree and the clustering-based model was proposed to include cases with different clinical conditions and indications. The model tree is a combination of decision tree and linear regression, where all cases are branched into leaves and regression is performed within each leaf. A clustering-based model used k-means algorithm to segment all cases into more aggregated groups, and then the regression was performed within each small group. The overall philosophy of both the model tree and the clustering-based method is that cases with similar features have similar geometry-dosimetry relation. Training cases within small feature range gives better model accuracy. The proposed method proved to be effective in improving the model accuracy over the model trained with all cases without segmenting the cases.

At last, the incremental learning was analyzed in radiation therapy treatment planning model. This study tries to answer the question when model re-training should be invoked. In the clinical environment, it is often unnecessary to re-train the model whenever there is a new case. The scenario of incrementally adapting the model was simulated using the pelvic cases with different number of training cases and new incoming cases. The result showed that re-training was often necessary for small training dataset and as the number of cases increased, re-training became less frequent.

In summary, this study addressed three major challenges in TPKDD. In the first part, an atlas-guided treatment planning technique was proposed to improve the modeling efficiency. In the second part, an automatic whole breast radiation therapy treatment planning technique was proposed to tackle the issue where TPKDD has not yet resolved. In the final part, outlier analysis, global model training and incremental learning was further analyzed to facilitate rapid learning, which lay the foundation of future clinical implementation of radiation therapy knowledge models.

Purpose: Cherenkov light is a natural byproduct of MV radiotherapy; recent results demonstrate that it can activate the drug psoralen sufficiently to induce cytotoxicity and increase MHC1 signal in vitro. Here, we investigate Cherenkov radiation from common radiotherapy beams using Monte Carlo, as well as methods to maximize Cherenkov production per unit dose, using filters placed in the beam path.

Methods: GAMOS, a GEANT4-based framework for Monte Carlo simulations, was used to model primary photon beams using spectra from a Varian linear accelerator and mono-energetic electron beams. Cherenkov photon spectra and track length along with dose were scored when irradiating a sphere of water with radius 50cm and SSD=50cm. Further measurements were taken with photon beams irradiating a 17.8cm3 cubic water phantom at 1mm3 detectors with depths of 8 to 9cm; SSD was set to 94cm. Finally, measurements were taken with filters of varying material and thickness placed 15cm below a 10MV FFF beam source.

Results: Simulated Cherenkov spectra were found to have strong overlap with the psoralen absorbance spectrum; dose and Cherenkov photon track length measurements established that higher beam energies had greater Cherenkov production per unit dose, with 18MV providing greater Cherenkov/dose than 6MV by a factor of 4. Simulations with filters suggest that copper and iron filters increase Cherenkov per dose more than aluminum for a given filter thickness, but that aluminum yields a greater boost for a given dose rate.

Conclusion: Initial work has been completed to show that the Cherenkov spectrum produced by radiotherapy beams is well suited for activation of psoralen, and that higher energy photon beams will result in more psoralen activation due to greater Cherenkov radiation per unit dose. We have also demonstrated that significant boosts in Cherenkov/dose can be achieved with the use of filters without overly compromising dose rate. Future work should expand analysis to include optical properties of tissues as well as additional filter materials.

On-board imaging is useful for guiding radiation to patients in the treatment position; however, current treatment-room imaging modalities are not sensitive to physiology - features that may differentiate tumor from nearby tissue or identify biological targets, e.g., hypoxia, high tumor burden, or increased proliferation. Single photon emission computed tomography (SPECT) is sensitive to physiology. We propose on-board SPECT for biological target localization.

Localization performance was studied in computer-simulated and scanner-acquired parallel-hole SPECT images. Numerical observers were forced to localize hot targets in limited search volumes that account for uncertainties common to radiation therapy delivery. Localization performance was studied for spherical targets of various diameters, activity ratios, and anatomical locations. Also investigated were the effects of detector response function compensation (DRC) and observer normalization on target localization. Localization performance was optimized as a function of iteration number and degree of post-reconstruction smoothing. Localization error patterns were analyzed for directional dependencies and were related to the detector trajectory. Localization performance and the effect of the detector trajectory were investigated in a hardware study using a whole-body phantom.

Typically targets of 6:1 activity were localized as accurately using 4-minute scans as those of 3:1 activity using 20-minute scans. This trend is consistent with the relationship between contrast and noise in the contrast-to-noise ratio (CNR) and implies that higher contrast targets are better candidates for on-board SPECT because of time constraints in the treatment room. Using 4-minute scans, mean localization errors were within 2 mm for superficial targets of 6:1 activity that were proximal to the detector trajectory and of at least 14 mm in diameter. Localization was significantly better (p < 0.05, Wilcoxon signed-rank test) with than without observer normalization and DRC at 5 of 6 superficial tumor sites. Observer normalization improved localization substantially for a target proximal to the much hotter heart. Localization error patterns were shown to be anisotropic and dependent on target position relative to the detector trajectory. Detector views of close approach and of minimal attenuation were predictive of directions with the smallest (magnitude) localization bias and precision. The detector trajectory had a substantial effect on localization performance. In scanner-acquired SPECT images, mean localization errors of a 22-mm-diameter superficial target were 0.8, 1.5, and 6.9 mm respectively using proximal 180°, 360°, and distal 180° detector trajectories, thus demonstrating the benefits of using a proximal 180° detector trajectory.

In conclusion, the potential performance characteristics of on-board SPECT were investigated using computer-simulation and real-detector studies. Mean localization errors < 2 mm were obtained for proximal, superficial targets with diameters >14 mm and of 6:1 activity relative to background using scan times of approximately 5 minutes. The observed direction-dependent localization errors are related to the detector trajectory and have important implications for radiation therapy. This works shows that parallel-hole SPECT could be useful for localizing certain biological targets.

Purpose: The goal of this planning study is to determine which sectors of the gantry rotation are most and least important in the treatment of head-and-neck carcinomas with Intensity Modulated Arc Therapy, and then use this knowledge to optimize the arc arrangement by adding arcs to reinforce the sectors that are most significant. Materials and Methods: Ten patients with head-and-neck cancer involving bilateral lymph nodes were selected for this planning study. Baseline RapidArc plans comprising two full gantry rotation arcs (RA2) were generated. Avoidance sectors and partial gantry rotations were used to produce RapidArc plans with various sectors removed: posterior (RA_post-), anterior (RA_ant-), or lateral sections (RA_lat-). Based on the results of these two-arc plans, two different resulting three-arc plans were created, with the third arc used to reinforce the important sectors. Results:The posterior sector was the least important contributor to overall plan quality. Removal of the lateral sector increased the dose to all critical structures with a resultant decrease in the median dose to the parotids. Removal of the anterior portion increased the dose to the larynx and parotids. The first three-arc plan produced from these results removed the posterior and lateral section and reinforced the anterior sectors (RA3_ant+). The second three-arc plan removed the posterior and one lateral sector, while reinforcing the anterior sector (RA3_ant+lat+). Both three arc plans provided better sparing to the parotids and spinal cord over RA2. Doses to the oral cavity, larynx, and brainstem were larger than RA2. RapidArc always produced plans with lower MUs than the corresponding IMRT plans while integral dose was lower for IMRT. Conclusions: For the class of tumors investigated in this report, RA3_ant+lat+ produced the most optimal plan in terms of target coverage and critical structure sparing while also being the simplest to develop treatment plans for.

Organ and tumor positions are constantly subject to change due to involuntary movement from the gastrointestinal and respiratory systems. In radiation therapy, accurate and precise anatomical localization is critical for treatment planning and delivery. Localization, prior to and during treatment, is most significant in stereotactic body radiation therapy (SBRT), which aims to aggressively target tumors by delivering high fractional dose to tight planning target volumes (PTV). Inter-fraction uncertainties from therapy-responding anatomical change and or patient positioning errors can be mitigated with adaptive therapy and on-board four-dimensional (4D) imaging. On the other hand, intra-fraction uncertainties from involuntary movement must be minimized by using real-time imaging. Real-time imaging enables more advanced treatment delivery techniques such as respiratory-gating and target tracking. Currently, no real-time 3-dimensional (3D) MRI tracking exist for on-board MRI-guided radiotherapy. Present MRI-guided radiotherapy machines are only capable of on-board two-dimensional (2D) cine MRI. Improving to real-time 3D MRI would provide plane-to-plane information and greatly improve target localization. The purpose of this thesis is to develop real-time 3D deformation field map (DFM) predictions using volumetric cine MRI (VC-MRI) and adaptive boosting and multi-layer perceptron neural network (ADMLP-NN) for MRI-guided 4D target tracking.

On-board VC-MRI is considered as the deformation of a prior 4D-MRI phase, MRIprior, obtained during patient simulation. The DFM that best estimates VC-MRI is constructed from a weighted linear combination of three major respiratory deformation modes extracted from principal component analysis (PCA) of DFMs between MRIprior and its remaining phases. PCA weighting coefficients are solved by the data fidelity constraint using on-board 2D cine MRI. Optimized PCA coefficients are tracked and used to train the ADMLP-NN to estimate future PCA coefficients from previous ones. ADMLP-NN uses several identical multi-layer perceptron neural networks with an adaptive boosting decision algorithm to avoid local minimums. Predicted PCA coefficients are used to build 3D DFMs for VC-MRI prediction.

This method was evaluated using a 4D computerized extended-cardiac torso (XCAT) simulation of lung cancer patients. Motion was simulated in the anterior-posterior and superior-inferior direction based on patient-specific real-position management (RPM) curve. Predicted PCA coefficient accuracy was evaluated against estimated PCA coefficients using normalized cross-correlation (NCC) and normalized root-mean-squared error (NRMSE). Predicted VC-MRIs was evaluated against ground-truth VC-MRIs using Volume Percent Difference (VPD), Volume Dice Coefficient (VDC), and Center of Mass Shift (COMS). Effects of ADMLP-NN parameter variation (number of input neurons, number of hidden neurons, number of MLP-NN, cost function threshold, prediction step size) on VC-MRI prediction accuracy were evaluated. Additionally, breathing pattern change effects between 4D MRI simulation and on-board 2D cine MRI were also evaluated.

Among all RPM signals examined, when no breathing pattern change occurred between the prior 4D MRI and on-board 2D cine MRI, the average predicted VPD, VDC, and COMS was 17.50 ± 2.85%, .92 ± .02, and 1.08 ± .44 mm. Prediction accuracy decreased when the breathing amplitude increased, but remained the same or improved when the breathing amplitude decreased between prior 4D MRI and on-board 2D cines. The feasibility and robustness of using ADMLP-NN to predict deformation fields maps for VC-MRI predictions for on-board target localization during radiotherapy treatments was demonstrated.

Abstract

Background: Monte Carlo and square root are two commonly used calculation methods to predict electron cutout factors in clinic. Monte Carlo is accurate but time consuming, while square root is faster but less accurate for cutouts with highly irregular shapes.

Purpose: Simplify and develop an efficient residual neural network model to predict electron cutout factors accurately and instantly.

Methods: 281 clinical cutouts were screened, and 12 groups were designed combining four different electron energies (6, 9, 12 and 15 MeV) and three different cones (10, 14 and 20 cm). The cutout factors of 35 previously used electron cutouts were calculated by Monte Carlo simulation and also measured with a solid water phantom and an ion chamber for validation of Monte Carlo accuracy. To solve the issue of sparse training data, 600 cutout samples were created for each group based on modifications of previously screened clinical cutouts. Cutout factors of these 600 samples were calculated with Monte Carlo simulation. 400 samples were randomly selected as training data, 50 as validating and the remaining 150 as testing. 1D Distance histograms were calculated as model input instead of 2D images to accelerate the training process. 1D Residual neural network with four residual blocks and three linear blocks was used. Performance of the trained model was evaluated with testing data, and the accuracy of the model was compared with square root method for eight selected cutouts with highly irregular shapes.

Results: The Monte Carlo calculated cutout factor agreed with the measurement within 0.7±0.5% on average. During the training process, the model started to converge within 20 epochs with 30 seconds. For model prediction, mean errors and maximum discrepancies for each energy and cone combinations were all within 1%, and the average overall error was 0.2±0.16%. Compared to square root method, the trained model performed better for cutouts with highly irregular shapes.

Conclusion: An efficient residual neural network model was simplified and developed, which is capable of estimating electron cutout factors accurately and instantly.

Abstract

Intensity Modulated Radiation Therapy (IMRT) and Volumetric Modulated Arc Therapy (VMAT) have become effective tools for treating cancer with radiation. Designing a high quality IMRT/VMAT treatment plan is time consuming. Different kinds of knowledge-based methods are being developed to reduce planning time and improve the plan quality by extracting knowledge from previous expert plans to form knowledge models and applying such models to the new patient cases. Currently, these methods are mostly limited to a particular cancer type and therefore various diseases types require training of multiple knowledge models with a large number of cases.

To investigate the feasibility of knowledge modeling of IMRT/VMAT treatment planning for multiple cancer types, a progressive study is conducted with a treatment planning knowledge model that quantifies correlations between patient pelvic anatomical features and the OAR sparing features. Low risk prostate plans with relatively simpler PTV-OAR geometry, which is the most common geometry type in previous knowledge based studies, are used to train the model as the starting point of the progressive modeling process. Cases with more complex PTV-OAR anatomies (prostate cancer cases with lymph node irradiation, and anal rectal cancer cases) are added to the training dataset one by one until the model prediction accuracies reach plateau. The DVHs predicted by the knowledge model for bladder, femoral heads and rectum are validated by cases from all three types of cases. Dosimetric parameters are extracted from the predicted DVHs and the corresponding actual plan values measure the prediction accuracy of this multi-disease type model. Further, its accuracy was also compared with the models trained by single disease type cases (including low risk prostate cancer, or type 1, high risk prostate cancer with lymph nodes, or type 2 and anal rectal cancer, or type 3).

Prediction accuracy reaches plateau when 6 high risk prostate cancer with lymph nodes irradiation cases and 8 anal rectal cancer cases were added to the training dataset. The determination coefficients R2 for the OARs are: Bladder: 0.90, rectum: 0.64 and femoral heads: 0.82. The prediction accuracies by the multi-disease type model and single-disease type models have no significant differences by F-test (p-value: bladder: 0.58, femoral head: 0.44, rectum: 0.97).

Conclusion:

Progressive knowledge modeling of OAR sparing for multiple cancer types in in the pelvic region is feasible and has comparable accuracy to single-disease type modeling.