AUTOMATED FEATURE-EXTRACTION FOR THE CLASSIFICATION OF HUMAN IN-VIVO C-13 NMR-SPECTRA USING STATISTICAL PATTERN-RECOGNITION AND WAVELETS

If magnetic resonance spectroscopy (MRS) is to become a useful tool in clinical medicine, it will be necessary to find reliable methods for analyzing and classifying MRS data, Automated methods are desirable be cause they can remove user bias and can deal with large amounts of dat a, allowing the use of all the available information. In this study, t echniques for automatically extracting features for the classification of MRS in vivo data are investigated. Among the techniques used were wavelets, principal component analysis, and linear discriminant functi on analysis. These techniques were tested on a set of 75 in vivo C-13 spectra of human adipose tissue from subjects from three different die tary groups (vegan, vegetarian, and omnivore), It was found that it wa s possible to assign automatically 94% of the vegans and omnivores to their correct dietary groups, without the need for explicit identifica tion or measurement of peaks.