Federated AI Radiotherapy Data Standardisation

Learning from large, diverse, international radiotherapy imaging, structure and radiation dose datasets enables us to understand variations in practice and develop AI models for personalised treatments to improve cancer outcomes and reduce side effects. This requires collaboration and standardisation of data across institutions. Federated learning is a machine learning approach where raw data remains at local institutions and only model parameters are exchanged to develop a combined model.

This PhD will focus on head and neck radiotherapy datasets using federated learning. Work will be undertaken with the Australian Cancer Data Network and in collaboration with Odense University Hospital, Denmark, strong partners of the internationally leading DAHANCA head and neck radiotherapy group.

Federated learning approaches will be developed and used to establish models to standardise retrospective datasets and then assess variation across institutions and patients. Auto-segmentation models will be adapted to incorporate accepted variation enabling robust modelling to improve patient outcomes.