Proposal Topic:Deep learning for outcome prediction after pelvic radiotherapy. Project description: Modern radiotherapy is highly optimized with respect to individual patient anatomy, utilising 3D anatomical imaging for treatment planning and guidance. This optimization is, however, fundamentally based on underlying assumptions about the relationships between the radiation dose delivered to specific anatomical structures (tumours and normal tissue) and tumour control and/or treatment toxicity – relationships which are still not well understood. Outcome modelling – relating radiation dose to early and long-term patient outcomes – is consequently an extremely active field of research. In this project, we use machine learning to predict toxicity and tumour control after pelvic radiotherapy in Cross-sectional data from a population of patients. We will construct a probabilistic statistical atlas describing the spatial patterns of radiosensitivity across the whole population. We will also create patient-specific sensitivity maps to feed into treatment plan optimisation. To alleviate the problem of missing outcome classification data, we will machine learning, e.g. semi-supervised models and cycle GANs.