Data-driven MRSI spectral localization using non-cartesian sampling trajectories

The ability to non-invasively visualize spatially-localized maps of metabolite concentrations in vivo as afforded by Magnetic Resonance Spectroscopic Imaging (MRSI) is an attractive prospect in clinically-focused biomedical imaging. However, the current gold standard implementation, known as Chemical Shift Imaging (CSI), is plagued by various artifacts, due primarily to the limitations dictated through use of the Fourier transform. To counter these impediments, numerous “constrained” reconstruction methods have been suggested, which typically inject some type of a priori information, usually with the aid of structural MR images, into the signal model. While this may be desirable for some applications, it introduces an assumption which posits a general equivalency between the spatial and spectral distributions, which may not always be appropriate. This work examines an alternative formulation in which, with the aid of statistical techniques and spatial regularization, constituent high-resolution spatial and spectral components are estimated from the raw MRSI data. We demonstrate the efficacy of this technique, and the robustness of the estimated components to alternative sampling strategies, thereby broadening the applicability of the method and offering the prospect of reduced acquisition times in more pressed clinical settings.