Physics-informed Machine Learning for Segmentation of Low Resolution Diffusion Magnetic Resonance Images

Abstract

Clinically-derived brain tissue classification on MRI is highly challenging due to the existence of apparent artefacts such as non-homogeneity, noise, aberrant tissue with a range of signal intensities, and the intricate anatomical structure of interest . This problem becomes worse in low-resource settings (such as in the global south) because of the low-field scanners. Physics machine learning (ML) seamlessly combines data and mathematical physics models, even in partially understood, uncertain and high-dimensional contexts. In recent research efforts, the need for developing new frameworks and standardized benchmarks as well as new mathematics for scalable, robust and rigorous next-generation physics-informed learning machines has taken center stage. This study explores Physics-based models and traditional machine learning methods for automated image segmentation using a low-resolution diffusion magnetic resonance images.