Item

Abstract

Several multi-dimensional statistical evaluation methods were applied to single-voxel proton MR spectra of glial brain tumors and of healthy volunteers. Metabolic coordinates with histological relevance for future diagnosis were found by which spectra from controls, low-grade tumors, and highgrade tumors were completely separated. Significant differences between low-grade and high-grade glioma patients and controls were found for several metabolic ratios by variance analysis. Cluster analysis both with and without principal component analysis was applied. The outcome of these two approaches depended mainly on the lipid-to-creatine ratio. Two other approaches, discriminant factor analysis and the orthonormal discriminant vector method were then used to find discriminatory metabolic coordinates. It turned out that a linear combination of all evaluable metabolic ratios made it possible to separate the three groups completely. On the basis of these results, a classification method that uses the entire proton MRS spectrum is proposed.