Detection of malignant tumours by multivariate analysis of proton magnetic resonance spectra of serum

Abstract

Proton magnetic resonance spectra of blood serum have been subjected to multivariate data analysis to discriminate between samples from cancer patients and from controls. The main feature was the use of digitally defined resonance profiles. The methyl and methylene lipoprotein signals centred at 1.3 and 0.9 parts per million are non-lorentzian composite peaks that cannot be described properly by the line width at half-height. Instead 71 and 76 data points were used to describe the methylene and methyl peak profiles, respectively. These data points were used as input to a principal component analysis to distinguish between malignant (n = 29) and control samples (n = 55). At a probability level of 0.01 (F-test) modelling classified all patients except 2 correctly, while 1 control was slightly above the predictive level for malignancy.