ARTIFICIAL NEURAL-NETWORK ANALYSIS OF H-1 NUCLEAR-MAGNETIC-RESONANCE SPECTROSCOPIC DATA FROM HUMAN PLASMA

Nuclear magnetic resonance (NMR) spectroscopy is finding increasing us e in studies of plasma and lipoproteins in health and disease, includi ng cancer. Analysis of the NMR data is not straightforward due to comp lex systems and also partly unknown underlying biochemistry. Here we d emonstrate how artificial neural networks can be utilised in biomedica l NMR. Their quantification power is illustrated by establishing lipop rotein lipid quantification directly from plasma H-1 NMR data. The bio chemical rationale for this example is elucidated on the basis of the relative weights of the spectral inputs in a trained network. A novel application of a Kohonen-type network architecture to classify plasma H-1 NMR spectra is also presented.