Closing the loop: fully automated radiotherapy planning for head and neck cancer

Abstract

Head and neck cancer is a prevalent cancer worldwide, with radiation therapy being one of the main treatment modalities. The head-and-neck region is particularly challenging due to the large number of surrounding organs-at-risk (OARs) and the need for steep dose gradients between the tumor and nearby OARs. Deep learning methods are playing an increasingly important role in treatment planning by reducing manual effort; however, they have predominantly been applied to auto-segmentation. In this thesis, we first develop a foundation model for image segmentation and subsequently focus on radiotherapy dose prediction using deep learning methods. We analyze factors influencing dose prediction, develop a novel loss function to better align predictions with clinical needs, and demonstrate that a clinically deliverable dose can be automatically derived through integration with a commercial Treatment Planning System (TPS). Collectively, these methods enable a fully automated radiotherapy workflow for a challenging treatment site. The resulting dose distributions are of high quality, with promising potential for clinical use.