Reduction of distortions in MRSI using a new signal model

We propose a new reconstruction scheme for magnetic resonance spectroscopic imaging (MRSI) signal based on minimizing the spatial total variation (TV) integrated with the lscr₁-norm of the spectral data. Furthermore, we propose to represent the MRSI signal as a linear combination of polynomials and spikes to capture the baseline and metabolite peaks. The proposed signal model provides a sparser representation of the spectral data, which enables us to suppress the field inhomogeneity induced line shape distortions and losses as well as reduce noise. We also take advantage of high resolution field map and anatomical prior derived from additional 3-D MRI scan to model system imperfections.