A fundamental challenge for personalised radiotherapy (RT) is enabling regular in-treatment RT plan adaption to account for organ motions, e.g. due to respiration, peristalsis, and bladder/bowel filling. These motions diminish the accuracy of the RT delivery to target regions (tumours), meaning that significant treatment margins must be incorporated in the plans, with the certainty that normal tissues also will then receive elevated radiation exposure.
This project’s aim is to develop image-driven modelling approaches that can reliably predict tissue motion states from sparse, and in some cases low quality, images acquired during and between treatments. Such predictions could enable responsive adaption of RT plans, potentially allowing dose escalation in target regions, with simultaneous toxicity reduction in sensitive nearby tissues.
We envisage a hybrid modelling approach for this purpose, that integrates biomechanical models of the relevant organs with AI techniques that can learn mappings between model configurations and corresponding images. This will ensure motion predictions are based on reliable physical and physiological principles while enabling robust and efficient exploitation of information derived from images. The work will thus span computational biomechanics and machine learning/ AI, while drawing heavily on medical image computing techniques.