Principal component and linear discriminant analysis of T1 histograms of white and grey matter in multiple sclerosis

Twenty-three relapsing remitting multiple sclerosis (RRMS) patients and 14 controls were imaged to produce normal-appearing white and grey matter T1 histograms. These were used to assess whether histogram measures from principal component analysis (PCA) and linear discriminant analysis (LDA) out-perform traditional histogram metrics in classification of T1 histograms into control and RRMS subject groups and in correlation with the expanded disability status score (EDSS). The histograms were classified into one of two groups using a leave-one-out analysis. In addition, the patients were scanned serially, and the calculated parameters correlated with the EDSS. The classification results showed that the more complex techniques were at least as good at classifying the subjects as histogram mean, peak height and peak location, with PCA/LDA having success rates of 76% for white matter and 68%/65% for grey matter. No significant correlations were found with EDSS for any histogram parameter. These results indicate that there is much information contained within the grey matter as well as the white matter histograms. Although in these histograms PCA and LDA did not add greatly to the discriminatory power of traditional histogram parameters, they provide marginally better performance, while relying only on data-driven feature selection.