Background and purpose: One of the main challenges in external beam radiotherapy treatment of locally advanced cervical cancer patients is dealing with bladder and rectum filling. Organ filling causes inter­fraction motion of the uterus, requiring large treatment planning volumes, or a plan library. Current assessment of tumor position is mainly done by visual inspection of a Cone Beam Computed Tomography (CBCT) scan. Eventually, this can lead to inter-­ and intra-­observer variability when choosing the best treatment plan from the plan library based on bladder filling. The incoming introduction of auto­contouring tools to obtain automatically-­generated (AG) contours of the bladder and the rectum on CBCT scans, allows the easier identification of these organs at risk and consequently, faster localization of the tumor region. However, to rely on these AG contours in the decision of plan selection, it is necessary to know if they have been reliably segmented. The goal of this project is to develop a strategy based on quantitative image features, to evaluate the quality of the AG contours to know if they are suitable for plan selection assessment. Materials & Methods: 140 LACC patients from Erasmus MC were included. For each patient, bladder and rectum contours were obtained from each of the CBCT scans done throughout the treatment (five fractions (CBCT scans) per patient). These contours were automatically­generated using a deep learning­-based auto­segmentation algorithm. Gold­-standard contours were manually delineated in some CBCT scans, but the rest of the automatically­generated contours did not have the corresponding ground-­truth contour, hence they were labeled with a score between 1 (bad quality) and 5 (good quality). For consistency, gold­-standard contours were included in the dataset with the class label 5. The contours were relabeled to have a binary classification problem, and those with label 3 were removed. Each contour volume was divided into three subregions: core region, inner and outer shell. This contour data was used for a comparison study between two supervised machine learning (ML) methodologies: Random forest (RF) networks and Logistic Regression (LR). For both strategies, feature extraction and selection were implemented. In RF methodology, a prior step of dimensionality reduction using principal component analysis (PCA) was performed. In LR, univariate feature selection followed by a multivariate logistic regression analysis was done. Before implementing the classifiers, the dataset was split into a training set and a test set. The ML models were trained using the training set, and they were tested on new unseen data. Predictions on the test data were obtained and used for evaluation of the model's performance using evaluation metrics: accuracy, sensitivity, specificity, confusion matrix, ROC curve, and AUC. Results: The RF classifier performed on the bladder test data with an AUC value of 0.87, while for the LR model, the value obtained was 0.77. The trained RF model identified the accurate and inaccurate bladder contours with a sensitivity of 94% and a specificity of 54%. The trained LR model resulted in a sensitivity of 91% and a specificity of 42%. In the case of the rectum, the RF classifier performance is indicated with the AUC value of 0.89, while the LR model obtained a value of 0.84. In the case of sensitivity and specificity, the RF model got 96% and 38%, and the LR classifier 95% and 38%, respectively. Conclusion: Random forest classifiers give the best results in terms of performance and classification skills for the OARs considered, especially for the bladder. It has been demonstrated that quantitative image features, paired with the corresponding contour class label, can be used for deriving statistical relationships from the data. This allows the identification of contouring errors and classifying the contours based on their quality. With the increasing automation of different steps in the radiotherapy treatment workflows, the automatic contour QA tool developed would be a key step in the process to ensure a faster, more feasible, and consistent plan selection. The tool could act as a support tool for radiotherapy technicians when choosing the plan from the plan library that best fits the daily anatomy of the patient.

Purpose: Proton therapy has been proposed as an alternative to conventional photon therapy for the treatment of locally advanced cervical cancer (LACC) since these patients experience toxicities. Proton therapy may allow for significant sparing of the organs at risk, reducing the incidence of treatment-related morbidities. The aim of this study is to develop a treatment strategy that is robust to motion and uncertainties in intensity-modulated proton therapy (IMPT) for the treatment of LACC. Materials and methods: Data from 14 LACC patients was included in this study. For each patient, a full and empty bladder planning CT (pCT) scan before treatment and four weekly repeat CT (reCT) scans after daily fraction were available. The full and empty pCT scans were used to create the patient-specific motion model of the cervix-uterus. An anisotropic CTV-to-ITV margin to expand this motion model was explored to account for uterine interfraction target motion. Subsequently, the motion model was divided into subranges to create a library of 1 to 4 plans, depending on the uterine motion due to bladder filling. Range and geometric uncertainties in the treatment of LACC are accounted for by robust optimization and evaluation. For each plan in the plan library, a treatment plan is created using the Erasmus-iCycle treatment planning system, taking into account EMBRACE-II constraints. To investigate whether the combination of margins, plan library, and robustness recipe is safe considering geometric and range uncertainties, ten treatments for each of the fourteen patients were simulated. These simulations were performed by recalculating the optimized treatment plans on the reCT scans with added uncertainties. We assumed that the target coverage was sufficient if the D95 of the target volumes was greater than or equal to 95% in at least 90% of the patients. Results: Of the 3430 margin recipes that were tested, the margin recipe with 95% cervix-uterus overlap and the smallest target volume was 1, 5, 7, 3, and 3 mm in the left/right, posterior, anterior, cranial, and caudal directions, respectively. The subranges of the motion model were expanded with the anisotropic margin recipe, after which robust optimization (setup robustness 5 mm, range robustness 3%) and evaluation (32 scenarios) of the treatment plans were performed. The treatment simulations showed that the D95 was greater than 42.75 for 99% and 92% of the patients for the cervix-uterus target volume and nodal target volume, respectively. Conclusion: The anisotropic margin and robustness recipe was robust to motion, geometric uncertainties, and range uncertainties when treating LACC patients with IMPT. Both values comfortably met the delivered dose criterion, indicating the strategy can be further improved.

Introduction Holland PTC (Delft, The Netherlands) is an independent outpatient clinic where ten different types of cancer are treated with Pencil Beam Scanning (PBS). Some of these cancer types are situated in the thoracic region and therefore exposed to breathing motion. The interplay between the dynamic pencil beam delivered in PBS and the target motion in the thoracic region can lead to severe dose inhomogeneities. While interplay mitigation tactics, such as repainting and using larger spot sizes, are available, they can come at a dosimetric cost. Simulating the interplay effect aids in formulating clinical protocols for managing interplay-related challenges during treatment planning, as well as assessing the robustness of treatment plans both retrospectively and prospectively. Goals The primary goal is to perform clinical evaluation of the treatment beams of 18 patients’ treatment plans. Clinical evaluation can be achieved using an interplay simulation script. The secondary goal is to assess which patient-specific factors affect the interplay effect in these patients. Methods A derivative script, based on a clinically validated interplay simulation script, has been used to perform clinical evaluation. This derivative script allowed the input of 2-dimensional tumor motion rather than only superior-inferior (SI) tumor motion. 2- dimensional tumor motion information has been extracted from 18 lung cancer, esophagus cancer, and lymphoma patients. Quality Assurance (QA) dosimetric measurements were available of all the patients’ beams, 161 in total. The static QA dose measurements are compared to the same measurements with a two-dimensional sinusoidal motion in an in-house developed interplay simulation script. Simulations of each treatment beam consisted of a number of patient-specific fractions, with 100 treatment simulations performed for each beam. Gamma acceptance criteria are set at 2% dose difference and 2mm distance to agreement. The last fraction where a treatment plan is below the predefined robustness threshold, known as the Last Fraction below Threshold (LFBT), is the metric on which clinical evaluation is based. If the 90th percentile LFBT was equal to or below the number of prescribed treatment fractions, a treatment beam is considered robust against the interplay effect. Near the end of thesis completion, the discovery was made that the interplay simulation script used throughout the majority of this thesis yielded results that differed notably and clinically from those generated by the validated script. Due to the work and time needed to repeat the simulations with the correct script, and the time necessary to analyze and interpret the new results, the decision was made not to repeat the simulations for all beams with the validated script. The simulations are repeated for nine out of 161 beams to compare the outcomes of both scripts. Results Using the unvalidated script,tumor-type specific interplay analysis shows interplay robustness in tumors with motion amplitude of up to 9 mm in lymphomas, 5 mm in lung cancer patients, and 6 mm in esophageal cancer patients. A high correlation (up to r=0.95) between 2D tumor motion amplitude and interplay robustness was found. The clinically validated script’s results show that all nine treatment beams are robust against the interplay effect, even when the amplitude is 11.22 mm in SI direction. All the patient beams with a tumor motion amplitude greater than 8.45 mm in SI direction are not robust according to the unvalidated script. This suggests that the unvalidated script underestimated the interplay robustness of the treatment beams. Implications/Conclusion Since only 9 out of 161 beams have been simulated using this script, limited general conclusions can be drawn from this data. The results of the clinically validated script suggest all beams up to 11.22 mm tumor motion amplitude are robust against the interplay.

Each year, more than 800 women in the Netherlands are diagnosed with cervical cancer, of which approximately half have locally advanced disease. The standard treatment for locally advanced cervical cancer (LACC) is external beam radiotherapy (EBRT) with concurrent chemotherapy followed by brachytherapy. Many patients experience some degree of treatment-related toxicity, mainly concerning the bowel, urinary tract, or vagina. Another important morbidity is hematologic toxicity (HT) due to bone marrow suppression, which might negatively impact the efficacy of adjuvant therapies. Treatment-related morbidity has a profound impact on patients’ quality of life. Proton therapy allows easier sparing of organs at risk (OARs), which might result in a decrease of treatment-related morbidities. One initiative to reduce treatment morbidities for LACC-patients was proposed in a collaborative project between Erasmus Medical Center (Erasmus MC), Leiden University Medical Center (LUMC), and Holland Proton Therapy Center (HollandPTC): the PROTECT-project. A clinical pilot study will be conducted to determine differences in dose to OARs and in morbidity outcomes when comparing state-of-the-art photon therapy with adaptive intensity-modulated proton therapy (aIMPT) for patients with LACC. Additionally, bone marrow sparing capabilities of both delivery techniques will be evaluated. The focus of this thesis is on the clinical implementation of aIMPT to facilitate the pilot-study and consists of three parts. The thesis started with a systematic review of the literature about the relationship between bone marrow dose and HT in LACC-patients treated with primary chemoradiation. The review has been submitted for publication. There was a scarcity of studies investigating the relationship between bone marrow dose and HT and clinically useful prediction models were not available yet. The majority of the studies defining bone marrow as the whole pelvic bone found a significant association between bone marrow and HT, in contrast to studies evaluating lower density marrow spaces or active bone marrow. Future studies may use whole pelvic bone contouring to develop normal tissue complication probability models. Secondly, the development and implementation of the treatment planning strategy for LACC-patients in HollandPTC were proposed. Uncertainties arising from proton therapy delivery were identified and strategies to address these uncertainties were determined. The proposed aIMPT-strategy for LACC-treatment consisted of a plan-of-the-day-approach with margins and robust planning. Further work includes establishing the balance between robustness settings and margins and optimizing the clinical implementation of the plan-of-the-day-strategy. Lastly, the workflow for LACC-patients in Erasmus MC was translated into the HollandPTC environment. Requirements for the new workflow were generated with input from both the investigator and a risk evaluation with the users. A workflow was designed and translated into HollandPTC's situation. Secondly, the implementation of the plan-of-the-day-strategy in the treatment management process of the oncology information system ARIA was investigated. Recommendations were made to finalize and validate the implementation of the workflows. This thesis provides a base for the clinical implementation of aIMPT for LACC in HollandPTC. The systematic review gives guidance for bone marrow sparing techniques in Erasmus MC and HollandPTC. Additionally, a combination of a plan-of-the-day-approach with margins and robust planning was identified as the most optimal treatment planning strategy. Lastly, an implementation strategy for the clinical workflow and treatment management process in HollandPTC was determined. Further work should focus on finalizing and validating the clinical implementation to facilitate PROTECT's clinical pilot study in HollandPTC.