Title: Quantification of Brain Magnetic Susceptibility using Deep Learning in Susceptibility Tensor Imaging

Quantification of anisotropic tissue magnetic susceptibility allows tracking of neural fiber pathways at high spatial resolution and detection of myelin changes. Separation of para- and dia-magnetic susceptibility sources can further provide precise tissue information that facilitates diagnosis of neurological diseases such as multiple sclerosis and Alzheimer's. We present three algorithms for susceptibility quantification and separation using data acquired from magnetic resonance imaging. DeepSTI is a data-driven deep learning algorithm for anisotropic susceptibility tensor reconstruction that requires significantly fewer measurements than previously needed, closing the gap of susceptibility tensor imaging (STI) to clinical application. WaveSep is a flexible Wavelet-based tool for susceptibility source separation. DeepSepSTI is an algorithm that jointly estimates para- and dia-magnetic susceptibility maps from phase and R2 measurements using a learned proximal approach. We show how these algorithms can be useful for quantifying brain magnetic susceptibility in both healthy and multiple sclerosis subjects.