Robust Principal Component Analysis for Improving Cognitive Brain States Discrimination from fMRI

Abstract

In this paper we propose a new method to discriminate cognitive brain states directly from functional Magnetic Resonance Images (fMRI). First, we apply Robust Principal Component Analysis (RPCA) to construct low dimensional linear-subspace representations from the noisy fMRI images for each subject and then perform a Gaussian Naive Bayes (GNB) classification. In previous studies the discrimination of cognitive brain states from fMRI is done by transforming the fMRI into a time sequence of voxels from which the brain states are inferred. RPCA improved the classification rate of a real benchmark fMRI data.