Background: While intensity modulated proton therapy (IMPT) plans allow for increased normal tissue sparing in comparison to photon radiotherapy, IMPT is also more sensitive to treatment uncertainties, including geometrical shifts in patient anatomy and range errors. Additionally, for thoracic treatment sites, e.g. the oesophagus, interference between breathing motion and IMPT treatment delivery, ‘the interplay effect’, can significantly degrade the quality of the dose distribution. To ensure IMPT plans are robust against treatment uncertainties, the dose distribution should ideally be evaluated in a high number of error scenarios. In this way, probabilistic robustness evaluation (PRE) enables the derivation of the probability distribution of relevant dose distribution parameters. Previous research has shown that polynomial chaos expansion (PCE) methods can efficiently perform PRE for head-and-neck and neurological IMPT plans, accounting for geometrical shifts and range errors. The goal of this thesis was to extend the PCE methodology for PRE to include the interplay effect, in addition to geometrical shifts and range errors, for evaluation of clinical IMPT plans of oesophageal cancer patients.
Methods: To include the interplay effect in the PCE methodology, we simulated 4D dynamic dose distributions with 1 set of PCE models per breathing signal. Sinusoidal breathing signals with a period between 3 s and 7 s were assumed. For 1 patient, we studied the effect of PCE parameters – Monte Carlo (MC) noise level, grid order (GO), polynomial order (PO), and PCE coefficient calculation methods – on PCE model accuracy. PCE model parameters and PRE parameters were selected with cost-accuracy analysis and validated for 2 patients. To evaluate the performance of the clinical protocol that was used to create the IMPT plans, PRE was performed including geometrical shifts, range errors, and different breathing signals for 20 clinical IMPT plans of oesophageal cancer patients treated at Holland PTC. We evaluated the probability of adequate CTV dose and scenario distributions of OAR (heart) dose and normal tissue complication probability (NTCP). Finally, we analysed the effect of several factors on PRE outcomes for 1 patient, including presence/absence of fractionation and repainting, extreme breathing signals, and presence/absence of geometrical shifts and range errors.
Results: Lowering the MC noise level improved PCE model accuracy. Selection of the GO3E1PO4 MC1.0% regression model resulted in accurate models for acceptable computational costs. Cost-accuracy analysis showed that 10 000 fully fractionated scenarios and 5 included breathing signals were statistically adequately for PRE, resulting in a total computation time of < 1 day per patient. For all patients, we found a probability of adequate clinical target volume dose (CTV D99.8% > 0.95 Dprescribed) larger than the criterium of 90%, with a population mean 99.56% (minimum – maximum: 98.75% – 99.96%). If fractionation and/or repainting are not considered, ≥ 20 breathing signals could be required for statistically adequate PRE.
Conclusion: We established and validated a pipeline for extending the PCE methodology for PRE to account for breathing motion. Our selected PCE models were as accurate as previously reported models. However, the number of included breathing signals required for statistically adequate PRE could vary depending on the IMPT treatment planning protocol. Our results show that the protocol used to create the included IMPT plans leads to safe, but conservative treatment plans. A lower prescription dose level could be used to significantly reduce the heart dose and NTCP, while still ensuring adequate dose to the CTV.

Background: Histopathological examination in the diagnostic workflow of oropharyngeal squamous cell carcinoma (OPSCC) is essential. We aimed to develop a machine learning pipeline to predict human papillomavirus (HPV) status in OPSCC patients based on clinical variables and multiparametric magnetic resonance imaging (MRI).
Methods: In a dataset of OPSCC patients (n=59), we extracted features from three categories: clinical variables; histogram parameters from diffusion weighted imaging (DWI)-MRI model; radiomics based on T2-weighted and DWI-MRI. We performed ten-times repeated stratified five-fold cross-validation and divided each outer training set (80%) into an inner training set (80% of outer training set) and validation set (20% of outer training set) using five-fold stratified cross-validation. We performed three types of feature selection methods (LASSO, statistical analysis and manual selection), tuned and trained seven classifiers (logistic regression, k-nearest-neighbours, naive bayes, random forest, support vector machine, XGBoost and LightGBM) to find the optimal combination of features, hyperparameters and classifier on each inner training set. We ensembled the inner fold models to fit on the outer training set and tested on the outer test set (20%). We constructed additional models with subsets of the features.
Results: the combined model area under the curve was 0.793±0.136. Models including clinical features outperformed models without clinical features (p<0.001). Features from all feature categories were selected for the combined model.
Conclusion: we were able to predict HPV status in OPSCC patients using multiparametric MRI and clinical variables with reasonable accuracy, though retraining and validating on larger, external datasets is needed before implementation in clinic.

The goal of radiotherapy is to maximize the dose to the target while minimizing the dose to normal tissue. Treatment plans are optimized for this goal and dose delivery is improved in accuracy and precision. The optimization is based on the delineations of the volumes of interest on the medical images of the patient. The problem is that these delineations contain uncertainties and the effect of these uncertainties become more pronounced as the accuracy and precision of dose delivery improves. The delineation uncertainties are caused by various factors such as knowledge and experience of the observers, guidelines, and image quality and modality. The goal of this thesis is to research the effects of delineation uncertainties. Uncertainties are characterized using a rolling ball algorithm (RBA) to modify the delineation. The radius of the ball represents the uncertainty and ranges from -5 mm to 5 mm for the clinical target volume (CTV) and -2 mm to 2 mm for the brain stem. The effect of these uncertainties will be researched for both fixed and re-optimized dose distributions. These are then used as input to create a model that will simulate the dose distributions for different uncertainties using Polynomial Chaos Expansion (PCE). PCE will be used to model dose volume histograms (DVH) of the CTV and the brain stem. The dosimetric effect will be determined by setting confidence intervals in D98% of the CTV and D2% of the brain stem. The widths of the confidence intervals in these two metrics represent the dose uncertainty. This thesis uses three sets of patient data. Each patient had a CTV close to the brain stem. Dose uncertainty in the two metrics was found to increase as the uncertainty of CTV and brain stem delineation increases. For a fixed dose distribution, a delineation uncertainty of 0.75 mm to 1.25 mm in the CTV would lead to a dose uncertainty of 2 Gy in the D98% of the CTV. A delineation uncertainty of 0.25 mm to 0.5 mm in the brain stem would lead to the same dose uncertainty in D2% of the brain stem. For a re-optimized dose distribution, a combined delineation uncertainty of 1.25 mm for two patients and 4.5 mm for one patient would lead to a dose uncertainty of 2 Gy in D98% of the CTV. For the same dose uncertainty in the D2% of the brain stem, there would be a combined delineation uncertainty between 1.0 mm and 4.5 mm. For all three patients, the dose uncertainty in the D2% was more sensitive to delineation uncertainties in the brain stem. The effects of these uncertainties will depend on the nominal situation. For all patients a 2 Gy dose uncertainty corresponds to 2.85% of the prescribed dose to the CTV and 3.33% of the maximum dose constraint of the brain stem. One patient was on the boundary of underdosing the CTV in the nominal situation and a 2 Gy uncertainty would be sufficient to underdose the CTV. The nominal D2% for all three patients was sufficiently low that it is unlikely it would exceed the maximum dose constraint.

Radiotherapy (RT) is an important modality in treatment against cancer. Developments in this modality are necessary for improvement of patient complications and survival. Aside from fractionation and precise irradiation, FLASH radiotherapy could be a third way to significantly increase the therapeutic window of radiotherapy.The FLASH effect is a biological, healthy tissue sparing effect that is found in tissue that receive a high dose (>10 Gy) within a very-short irradiation time (<100 ms). Although the mechanics behind the FLASH effect remain unknown, it is found to benefit a large variety of tissues and organs. Nevertheless, FLASH-RT still lacks clinical translation and implementation of FLASH irradiation is limited by current technology. However, the extremely high dose rates required for FLASH can readily be achieved using cyclotron accelerated proton transmission beams. Cancer patients with small lung tumours are a preferred starting point for clinical translation of FLASH because (i) transmission beams are of special benefit for lung tumours since they mitigate for range uncertainties and contain a sharp lateral penumbra, (ii) high doses are required for the FLASH effect to occur and hypofractionation is not uncommon for lung patients, and (iii) the very short irradiation times and high doses in combination with gating allow smaller treatment margins for moving tumours.In this study, we took a first step towards intensity modulated proton therapy (IMPT) with FLASH treatment plan optimisation by (i) investigating the most delivery-time effective way to irradiate a field, (ii) proposing an optimisation method to maximise FLASH within a treatment plan, and (iii)analysing and evaluating FLASH treatment plans. Because increased pencil-beam (spot) separation showed to benefit the pencil-beam scanning(PBS) delivery time of a field, spot overlap minimisation was proposed as optimisation method for FLASH. Treatment plans using 3, 5 and 7 transmission beams were generated with and without FLASH optimisation for 6 small lung tumour patients. Although spot overlap minimisation doesoccasionally improve delivery time for a limited number of beam angles, the overall treatment plan does not benefit from the optimisation. Therefore, spot overlap minimisation cannot be used for FLASH optimisation in clinical IMPT treatment planning. Recommendations for future research include optimisation on PBS delivery time and PBS pattern optimisation.