The human brain consists of a myriad of chemical compounds critical to its functioning. Audgroup of these compounds, collectively known as metabolites, have been a research interest forudyears because the pathogenesis of neurodegenerative diseases, a tumours classification, the effectivenessudof a drug, etc., can be investigated via variations in brain metabolite concentrationudlevels. Nuclear Magnetic Resonance Spectroscopy (NMRS) enables investigators to conductudnon-invasive in vivo studies of metabolites in the human brain and the rest of the body. Howeveruda number of problems have hindered the usage of NMRS as a clinical diagnostic tool. Oneudis the non-uniqueness of the most widely used analysis methods, i.e. as the parameters and/orudprior knowledge data of an analysis method are changed, the results also change. A secondudproblem is the lack of a method that can automatically classify the signal components estimatedudvia signal decomposition based signal analysis methods. Additionally, some of the mostudwidely used analysis methods, by virtue of their algorithms, intrinsically assume the nature ofudNMRS signals, e.g. stationary, linear, Lorentzian, etc. Hence, this thesis explores a new analysisudapproach, based on a theoretical and practical understanding of NMRS, that (a) avoids makingudassumptions about the nature of experimentally acquired NMRS signals, (b) relies on a uniqueuddecomposition analysis method, and (c) automatically classifies the estimated peaks of an analysis.udUnique decomposition analysis was conducted via the rarely used unique and non-linearudsignal decomposition method − the Fast Pad´e Transform (FPT). The FPT is compared withudthe main decomposition based NMRS analysis methods via a detailed mathematical analysis,udand a comparative analysis. Automatic classification was conducted via a novel classificationudmethod, which is introduced herein, and which is based on quantum mechanical predictions ofudmetabolite NMRS behaviour.
展开▼
