Representation of strong baseline contributions in 1H MR spectra

A comparison is made between two optimization procedures and two data models for automated analysis of in vivo proton MR spectra of brain, typical of that obtained using MR spectroscopic imaging at 1.5 Tesla. First, a shift invariant wavelet filter is presented that provides improved performance over a conventional wavelet filter method for characterizing smoothly varying baseline signals. Next, two spectral fitting methods are described: an iterative spectral analysis method that alternates between optimizing a parametric description of metabolite signals and nonparametric characterization of baseline contributions, and a single-pass method that optimizes a complete spectral and baseline model. Both methods are evaluated using wavelet and spline models of the baseline function. Results are shown for Monte Carlo simulations of data representative of both long and short TE, in vivo 1H acquisitions.