A Fast Patch-Based Hankel Low-Rank Method for Magnetic Resonance Spectroscopy Reconstruction.

Abstract

Sparse sampling is an effective strategy for accelerating the acquisition of multi-dimensional magnetic resonance spectroscopy (MRS), crucial in disciplines such as chemistry and structural biology. The state-of-the-art low-rank reconstruction methods enable the high-fidelity recovery of sparsely-sampled MRS but are limited by lengthy reconstruction times, posing a significant challenge. In this work, we introduce a novel approach that significantly reduces the dimensionality of the constructed low-rank Hankel-like matrix. This reduction leads to lower computational complexity and, as a result, a substantial acceleration in reconstruction times compared to conventional low-rank methods. Experimental evaluations on both simulated and real MRS demonstrate that our method achieves a reduction in reconstruction times by over fourfold without sacrificing the quality of spectrum reconstructions.