Using metabolic medical imaging to model tumour growth and response to therapy

The number of cancer related deaths is predicted to reach over 13.1 million in 2030. Understanding the spatio-temporal evolution of tumours and their response to therapy is crucial to thwart this gloomy prognosis. Mathematical models have been used to elucidate the biological processes involved in tumour growth, but their complexity severely limits their clinical value. Eective models should base predictions on patient-specic data that can be obtained and tested in the clinic. This thesis aims to develop novel clinically-relevant methodologies to model solid tumour growth and reaction to therapy. Models are developed that predict the macroscopic tumour evolution as a result of complex biological processes happening at the microscopic scale (such as cell proliferation, hypoxia, necrosis and oxygenation) while retaining simplicity in terms of the number of parameters that require calibration. This facilitates integration of medical imaging-derived tumour descriptor maps (e.g. for cellularity and nutrient availability) to initialise, calibrate and verify the models. In addition to the models themselves, this thesis contributes to the literature by introducing novel frameworks to 1) derive maps of proliferative, hypoxic and necrotic cells using only Dynamic Contrast Enhanced Magnetic Resonance Imaging data; 2) apply these maps to constrain the models and obtain subject-specic predictions of tumour evolution; 3) account for tumour vascularisation, focusing on the angiogenesis-driven macroscopic changes observed longitudinally on medical imaging scans; and 4) produce a proof-of-concept updated model that could support medical decision in the treatment and management of cancer patients. The results demonstrate the feasibility of using information derived from imaging data to quantify and predict subject-specic global and local tumour evolution and reaction to therapy. Experimentally validated comparison between model predictions and tumour imaging data obtained at later time points of the tumours' evolution show noticeable improvements in prediction accuracy, model complexity and clinical value over state-of-the-art methods. Nevertheless, additional imaging information on complementary aspects of tumour growth and response to therapy could render our models even more relevant to clinical practice.