Quantification in magnetic resonance spectroscopy based on semi-parametric approaches

Abstract

Magnetic resonance spectroscopy (MRS) is a value-added modality to magnetic resonance imaging (MRI) that is often used in diagnosis, treatment and progression monitoring, as well as in non-destructive, non-invasive studies of disease states in humans and model systems in animals. The availability of high magnetic field strengths and use of hyperpolarized nuclei, combined with the possibility of acquiring spectra at very short echo-time, have dramatically increased the potential of MRS. For the last two decades, a challenge has been to quantify short echo-time proton spectra that exhibit many metabolites, and to estimate their concentrations. Quantification of such spectra is challenging. Because the model function describing the acquired MRS signal is incomplete, semi-parametric techniques are required for estimation of the wanted metabolite concentrations. The semi-parametric approaches, QUEST, AQSES, TARQUIN, LCModel and SiToolsFITT, are reviewed and discussed according to handling of macromolecule signal and unknown decay of the metabolite signal (lineshape). Estimation of noise-related errors on model parameters and compromise used in real-world applications are detailed, with emphasis on the bias-variance trade-off. Applications of the semi-parametric methods QUEST and AQSES to quantification of MRS, HRMAS and MRSI data are also provided.