Intrinsic Prior Knowledge Driven CICA fMRI Data Analysis for Emotion Recognition Classification

The purpose of this paper is to solve an urgent problem about the selection of differential features under different emotional states. The fMRI data of positive and negative emotional (PNE) states were obtained through strict experimental design, and an improved GICA-IR method was proposed to extract the intrinsic prior information implied in fMRI data and decompose the corresponding functional independent component (IC) in the two states by integrating intrinsic prior information into constraint independent component analysis (CICA) method. The IC-fingerprints corresponding to the functional independent components of the above two states were also calculated, and then they were regarded as the classification features to accurately identify and classify PNE states by the ways of random sub-sampling and support vector machine. The experimental results showed that the presented method had a commendable effect on the classification of positive and negative emotions with an average classification rate of 99%. The proposed improved GICA-IR method was superior to the classical ICA method in terms of the source signal recovery from two aspects of spatial spectrum and power spectrum density, which provided a reliable basis for the good performance of the proposed method based on prior information IC-fingerprint in the emotional classification.