Quantification and management of uncertainties in radiation therapy

For the therapeutic success of radiotherapy treatment it is crucial that the actual dose distribution delivered to a patient coincides as precisely as possible with the planned distribution, since deviations might lead to unexpected treatment toxicities or tumor recurrences. In this cumulative dissertation, an approach for patient positioning to manage the impact of anatomical changes in photon therapy is evaluated, and a framework for quantitative analyses of various types of uncertainty in proton therapy is developed. In the first project, an existing algorithm for scatter correction of cone beam computed tomography (CBCT) images is used to enable a dose recalculation on the changed patient anatomy. This is combined with a software tool for multi-criterial optimization of the patient position based on dosimetric parameters. The feasibility of the approach is shown for two clinical datasets of head and neck cancer patients, and potential benefits in comparison with the clinical standard - a rigid registration of the CBCT to the planning images - are evaluated. It is concluded that the approach offers increased control over target coverage and organ-at-risk (OAR) doses, since in many cases target coverage or OAR dose could be improved compared to the rigid image registration approach. However, for pronounced anatomical changes, both approaches were unable to restore an acceptable target coverage. In the second project of this thesis, a framework for variance-based sensitivity analysis of uncertainties in proton therapy was developed. With a fast, GPU-based pencil beam algorithm, a large number of error scenarios for a proton therapy treatment plan can be calculated. In these scenarios, patient position, proton range and relative biological effectiveness (RBE) model parameters are varied simultaneously and independently within their assumed uncertainty distributions. With this Monte Carlo approach, also interactions between multiple types of uncertainty are taken into account. For the dose distribution and dosimetric parameters such as dose volume histogram (DVH) quantiles for target structures and OARs the overall uncertainty as well as sensitivity indices are calculated. These indices allow for a ranking of the individual input uncertainties with respect to their impact on the overall uncertainty. The feasibility and the capabilities of the framework are shown with two clinical patient datasets. In a further step, the framework is extended to include inter observer variability (IOV) in target definition. In a study with ten patients and 10 observers, the impact of IOV in comparison with setup and range uncertainty on clinical target volume (CTV) coverage is evaluated. For two out of ten patients, a relevant impact of IOV was found. In future studies, this framework might help to determine which types of uncertainty are driving the overall uncertainty of clinically relevant dosimetric parameters and might help to prioritize research attempts aiming at the reduction of uncertainty.