Magnetic Resonance Spectroscopy Data Quantification

(Nuclear) Magnetic Resonance ((N)MR) is a non-invasive technique that has been used to acquire spatially resolved images of living organisms. Another application of clinical MR is MR Spectroscopy (MRS) in which chemical information can be extracted from a well-defined region (e.g. a voxel) in for example the human brain [de Graaf, 1998]. Brain tumors [Nelson, 1999], multiple sclerosis, epilepsy and Alzheimer constitute some very important research areas where MRS can identify the pathology, provide insight in the underlying biochemical processes or monitor the effect of medication. Depending on the measurement protocol used, so-called single-voxel or multi-voxel (multiple signals acquired simultaneously over a larger region, one per voxel) longor short-echo-time proton signals (see Figure 1) are obtained. Applying spectral analysis methods can provide quantitative information about the metabolites. However the quantification is hampered by a low signal-to-noise ratio, deviations from the theoretical model function and the presence of disturbing components (like residual water and lipids). Moreover, the exact number of spectral components is unknown. To complicate matters further, short-echo-time signals are characterized by the presence of an unknown broad baseline underlying the resonances of the metabolite signals of interest and by heavily overlapping metabolite signals. To exploit the full potential of MRS the reliable determination of the metabolites present and their concentrations is essential. We discuss the state of the art of data processing methods for longand short-echo-time proton measurements including techniques to remove unwanted features that hamper the quantification.