Clustering based feature selection methods from fMRI data for classification of cognitive states of the human brain
Awijit Gupta, Arjun Gupta, Kapil Sharma · International Conference on Computing for Sustainable Global Development · 2016
Functional Magnetic Resonance Imaging (fMRI) is a neuroimaging technique that has proven to be useful for decoding and analyzing cognitive states of the human brain. It generates 3D images with high spatial resolution resulting in high dimensional images. In order to avoid curse of dimensionality, we need good feature selection techniques. Through this paper, we propose a two phase algorithm for finding a small subset of relevant and non-redundant voxels for classification of cognitive states in a human being. In the first phase, the entire feature set is clustered using normalized cut with maximal information compression index as the similarity. The most informative voxel is selected from each cluster using t-statistics and a pool of k voxels is created. In the next phase, a wrapper based sequential forward feature selection technique is used to obtain the optimal voxels for a given classifier.