Learning-Based Motion Correction for High-Resolution 3D MRSI of the Brain without Water Suppression

Motivation: MRSI scans are susceptible to motion artifacts due to long acquisition times. The problem is more serious in non-water-suppressed MRSI experiments, where head motion makes the removal of water and lipid signals more difficult. Goal(s): To develop an effective method for correcting head motion in high-resolution, non-water-suppressed MRSI scans. Approach: We propose a novel learning-based motion correction method that incorporates SENSE-based regularization, internal priors (spatial sensitivity and smoothness) and external priors (multi-reference motion-free images). Results: The proposed method has been validated on both simulation data and in-vivo data from AD and stroke patients with involuntary head motions, producing high-quality water and metabolite images. Impact: A novel motion correction method has been developed for high-resolution, non-water-suppressed MRSI of the brain. This method has the potential to significantly improve the robustness and clinical applicability of non-water-suppressed MRSI.