Split-Bregman-based group-sparse reconstruction of multidimensional spectroscopic imaging data

Abstract

4D Magnetic Resonance Spectroscopic Imaging data provides valuable biochemical information in vivo, however, its acquisition time is too long to be used clinically. In this paper, 4D phantom MRSI data are retrospectively under-sampled 4X, 6X, and 8X then reconstructed with Compressed Sensing and Group Sparsity. A derivation for the Group Sparse problem solution within the Split-Bregman framework is provided which allows for arbitrary, over-lapping groups of transform coefficients. Results show that Group Sparse reconstruction with over-lapping groups is more accurate at each under-sampling rate than Compressed Sensing reconstruction with superior peak line-shape and amplitude reproduction. The acceleration factors used in these experiments could potentially reduce scan times from 40 minutes to 5 minutes.