A priori-driven multivariate statistical approach to reduce dimensionality of MEG signals

A magnetoencephalography (MEG) multivariate data exploratory analysis is described and implemented that combines the variance criterion used in principal component analysis with some prior knowledge about the sensory experimental task. By using the idea of rearranging the data matrix in classification pairs that correspond to the time-varying representation of either stable or stimulus phases of the specific task, the feature extraction method is constrained reducing significantly the number of principal components necessary to represent most of the total variance explained by the MEG signals.