PATTERN-RECOGNITION ANALYSIS OF H-1-NMR SPECTRA FROM PERCHLORIC-ACID EXTRACTS OF HUMAN BRAIN-TUMOR BIOPSIES

Pattern recognition techniques (factor analysis and neural networks) w ere used to investigate and classify human brain tumors based on the H -1 NMR spectra of chemically extracted biopsies (n = 118), After remov ing information from lactate (because of variable ischemia times), uns upervised learning suggested that the spectra separated naturally into two groups: meningiomas and other tumors, Principal component analysi s reduced the dimensionality of the data. A back-propagation neural ne twork using the first 30 principal components gave 85% correct classif ication of meningiomas and nonmeningiomas. Simplification by vector ro tation gave vectors that could be assigned to various metabolites, mak ing it possible to use or to reject their information for neural netwo rk classification, Using scores calculated from the four rotated vecto rs due to creatine and glutamine gave the best classification into men ingiomas and nonmeningiomas (89% correct). Classification of gliomas ( n = 47) gave 62% correct within one grade. Only inositol showed a sign ificant correlation with glioma grade.