Advanced Radiotherapy Planning Based on Probabilistic Concepts: Understanding and modelling geometrical uncertainties in head and neck radiotherapy to evaluate plan robustness

Abstract

Geometrical uncertainties in radiotherapy such as differences in patient set-up positioning, and organ motion/deformations can cause the target to be underdosed or organs at risk to be overdosed. The aim of this thesis was to understand geometrical uncertainties in head and neck (H&N) radiotherapy; how to model them, how to evaluate their impact and how to make plans that are robust against them. One way of modelling anatomical deformations is by using principal component analysis (PCA). In this thesis, a method for evaluating how well PCA models represent unseen deformations within a patient or population was first developed. This evaluation scheme was demonstrated in H&N cancer patients for both patient specific and population-based models. In the studied cohorts, the largest residual errors were found around the oropharynx. Next, we developed a population-based time-dependent model for anatomical deformations in H&N using data from 30 H&N patients. This involved creating a PCA model for the systematic components of the deformations and weekly models for the random components. These models were then used to simulate many treatments and the effect of deformations on the delivered dose to the patient was evaluated alone and in combination with set-up uncertainties. The effect of anatomical deformations was found to be similar to or smaller than that of set-up uncertainties for all organs considered except the larynx and the primary clinical target volume (CTV). Considering this finding, we then investigated whether plans could be created using tools to account for set-up uncertainties that were robust to anatomical deformations. We compared plans created using margin, robust and probabilistic approaches for different uncertainty settings. Our results show that margin-based plans were the most robust to anatomical deformations using only methods to account for set-up uncertainties. Finally, we investigated whether the shape of the target affects the robustness of plans for set-up uncertainties. We compared margin, robust and probabilistic plans for different target shapes and found that as the CTV was less spherical, the plan robustness decreased. Margin-based plans were seen to be over-robust to set-up uncertainties and both robust and probabilistic planning approaches were seen to underdose voxels on the 'corners' of more complex CTV shapes. Collectively, these results show that while in many cases plans can be created that are robust to geometrical uncertainties, including deformations, robust and probabilistic planning approaches should be used with care to ensure adequate target coverage.

Date of Award	31 Dec 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Eliana Vasquez Osorio (Supervisor), Andrew Green (Supervisor) & Marcel Van Herk (Supervisor)