Advanced signal processing for magnetic resonance spectroscopy

Assertive diagnosis of cancer, Alzheimer?s disease, epilepsy and other metabolic diseases is essential to provide patients with the adequate treatment. Recently, different invasive and non-invasive techniques have been developed for this purpose, nevertheless, due to their harmless properties the non-invasive techniques have gained more value. Magnetic Resonance is a well-known non-invasive technique that provides spectra (metabolite peaks) and images (anatomical structures) of the examined tissue. In Magnetic Resonance Spectroscopy (MRS molecules containing certain excitable nuclei, such as 1H, provide the metabolite information. As a consequence, the peaks in the MR spectra correspond to observable metabolites which are the biomarkers of diseases. Finally, metabolite concentrations are computed and compared against normal values in order to establish the diagnosis. The method to obtain such amplitudes is also called quantification and its accuracy is essential for diagnosis assessment. Quantification of MRS signals is affected by a relatively low signal-to-noise ratio (SNR), a residual water resonance, lineshape distortions, overlapping resonances and underlying macromolecules and lipids affecting their baseline. Our ultimate goal was, therefore, the development of signal processing tools and algorithms to improve the quantification of MRS signals. For this, we first proposed a heuristic method to improve the residual water filtering using Hankel Singular Value Decomposition (HSVD). Additionally, a method for lineshape estimation of distorted MR spectra was developed and evaluated in simulated, in vitro and in vivo signals. Furthermore, a baseline approach was implemented via a parametric modelling method based on prior knowledge acquired from a set of in vivo macromolecular signals (measured via inversion recovery), which aims at avoiding long measurements and allowing a flexible set of baseline components. Finally, for the analysis of quantification results we focused on an automatic evaluation of the residual, thereby benefiting MRS spectral analysis in the clinic. This work was developed within the context of the European project ?FAST? (MRTN-CT-2006-035801) – Advanced Signal Processing for Ultra Fast Magnetic Resonance -, which was a Research and Training Network granted by Marie Curie Actions in the 6th Framework Program (2007-2010), http://fast-mrs.eu.