Distributed learning and prediction modelling in radiation oncology

Artificial intelligence (AI) research has the potential to improve the treatment of cancer but requires access to massive amounts of patient data. Exchanging patient data between clinics and researchers always bears the risk of unintended data leaks and patient privacy violations. This thesis describes a computer network that has been implemented in oncology clinics in the Netherlands and around the globe, which allows conducting AI research without the need to exchange patient data: patient data stays secure in each clinic. This thesis additionally compares existing AI algorithms for radiation oncology and proposes a new AI methodology.