Spectrum separation of Magnetic Resonance Spectroscopy based on sparse representation

In this paper, a novel spectrum separation technique based on sparse representation is proposed to deal with Magnetic Resonance Spectroscopy (MRS) quantification which is used to measure the levels of different metabolites in brain tissues. Since a measured MR spectrum contains the spectra of numbers of metabolites and a baseline, the separation and quantification of them becomes difficult. A nonnegative pursuit algorithm based on regularized FOCUSS algorithm is proposed here to decompose a measured spectrum with respect to an overcomplete dictionary. Benefitting from the a priori knowledge, the dictionary is built by Lorentzian and Gaussian basis functions representing different metabolites and baseline. Using this algorithm, not only the baseline is separated from the spectra of interest, but also the spectra of different metabolites are separated. The accuracy of quantification and the robustness are improved, from simulation data, compared with a commonly used estimation method. The quantification on tumor metabolism with in vivo brain MR spectra is also demonstrated.