Learning-Based Methods for Improved Simultaneous PET-MRI

This thesis deals with developing robust learning-based algorithms that leverage the underlying imaging physics for improving image quality in simultaneous positron emission tomography (PET) and magnetic resonance imaging (MRI) systems. The thesis focusses on developing algorithms for improving image quality arising from low-dose PET and accelerated MRI scenarios and validates the innovated algorithms on simulated and in vivo data. Further, the thesis validates the robustness of the algorithms on out of distribution data arising from physiological and scanner related perturbations. Further, the thesis also proposes a framework to use the enhanced PET images for a novel imaging application called functional PET imaging using MRI information, for improved estimations of brain functions at higher temporal resolution.