Radiograph-based organ dose reconstruction for childhood cancer survivors with long-term follow-up

The motivation of the research presented in this thesis was to provide solutions for accurate organ dose reconstruction based on the clinical records of childhood cancer survivors treated in the 2D radiation treatment (RT) planning era. The reconstructed dose information can be used to investigate dose-effect relationships in childhood cancer survivors with long-term follow-up. We need this knowledge to, for example, determine the dose that can be maximally delivered to specific organs so that children can receive better treatment. In this thesis, we investigated the impact of anatomical discrepancies between surrogate anatomies and the patient’s true anatomy on dose reconstruction quality, and formulated solutions to minimize this impact. To achieve this, we first assessed the feasibility of using patient CT scans as surrogate anatomies for organ dose reconstruction using a common literature standard (i.e., age and sex) for selecting a surrogate anatomy. Thereafter, we investigated the correlations between the deviations of a set of patient characteristics / anatomical features and deviations in organ dose reconstruction. Further, we proposed two novel approaches utilizing big data and machine learning (ML) techniques, one creating individualized surrogate anatomies, and the other one uses no surrogate anatomies at all. Finally, we validated and compared our approaches with the current state-of-the-art approaches on a common data set. Our approaches achieved an equally good (if not better) quality of dose reconstruction results with better efficiency and robustness compared to the current state-of-the-art approaches