Dimensionality Reduction Based Classification of Proton Magnetic Resonance in vivo Spectra from the Normal Human Brain

摘要

The purpose of this work was to exploit multivariate projection techniques of data analysis in evaluation of regional biochemical heterogeneity of the normal human brain. Principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) applied both to the absolute metabolite levels and to the unresolved spectra were chosen to study the metabolic differences between thalamus, pons, cerebellum, hippocampus, frontal and occipital areas. Dimensionality reduction offered by these techniques permitted graphical illustration of the separation between the spectra originating from different anatomical regions and biochemical interpretation of the observed spectra clustering.