Spatial spectral modeling for robust MRSI.

We propose a novel spatial spectral model for the reconstruction of magnetic resonance spectroscopic imaging (MRSI) signal. We penalize the compartmentalized spatial total variation norm of the signal to exploit the spatial properties of the metabolite peaks. The spectral signal is modeled as a sparse linear combination of spikes and polynomials to capture the peaks and baseline induced by unsuppressed water and lipids. We also use the high-resolution map of the magnetic field distribution within the slice to model the image acquisition, thus correcting for intra-voxel line shape distortions. The spectral model enables the stable recovery of the signal even in challenging spatial regions, while the spatial model suppresses the spectral leakage from extra-cranial fat and inter-voxel crosstalk. We acquire the MRSI signal using EPSI, while the high-resolution 3-D MRI information is derived using Dixon scans. The reconstruction of phantom and in vivo MRSI data demonstrate a significant improvement in spectral quality and accuracy over classical MRSI schemes.