A data-driven approach to optimise parameters of a computational digital twin model in response to SBRT on MR-Linac

New hypofractionated schedules, such as stereotactic body radiation therapy (SBRT), have been proposed to adapt and personalise radiation therapy (RT) for prostate cancer (PCa) patients. The ultimate goal is to reduce recurrence risk by shortening the total treatment duration and increasing the fraction of doses. SBRT, not previously used due to limited precisions of linacs, is now used in clinics, alongside new precision systems such as the MR-Linac in adaptive RT online workflows. However the impact of such treatments on individual patients with MR-Linac remains unknown. As a result, there is an increased interest on numerically simulating SBRT response using computational digital twin models that allow to simulate tumour response to different fractionations. However, these models require the optimisation of several radiobiological parameters, which are often difficult to adapt to patients, with some being unobservable. This paper proposes a data-driven pipeline based on optimisation algorithms to estimate the parameters of a computational digital twin model in response to SBRT by identifying patient-specific tumour evolution during treatment on MR-Linac. To that end, we rely on longitudinal follow-up MRI data, acquired on an Elekta 1.5T MR-Linac at each session during SBRT treatment for 3 patients. Simulation results were in accordance with patient data. This methodology opens the road to personalised therapy for PCa patients by numerically simulating tumour evolution during RT.