Tensor gradient based discriminative region analysis for cognitive state classification

Abstract

Extraction of relevant features from high-dimensional multi-way functional MRI (fMRI) data is essential for the classification of a cognitive task. In general, fMRI records a combination of neural activation signals and several other noisy components. Alternatively, fMRI data is represented as a high dimensional array using a number of voxels, time instants, and snapshots. The organisation of fMRI data includes a number of Region Of Interests (ROI), snapshots, and thousand of voxels. The crucial step in cognitive task classification is a reduction of feature size through feature selection. Extraction of a specific pattern of interest within the noisy components is a challenging task. Tensor decomposition techniques have found several applications in the scientific fields. In this paper, a novel tensor gradient-based feature extraction technique for cognitive task classification is proposed. The technique has efficiently been applied on StarPlus fMRI data. Also, the technique has been used to discriminate the ROIs in fMRI data in terms of cognitive state classification. The method has been achieved a better average accuracy when compared to other existing feature extraction methods.