Predicting the Cognitive States of the Subjects in Functional Magnetic Resonance Imaging Signals Using the Combination of Feature Selection Strategies

The functional magnetic resonance imaging (fMRI) provides very useful information about the activities from different brain areas during a task. This information can be used to train a classifier and predict the sensory and motor functions and also different mental states of the subject's brain in a particular task. Using a high resolution fMRI, normally the activities from many voxels are obtained with respect to time and not all of these voxels involve actively in a particular task. Here we propose a combination of feature selection strategies using an evolutionary computation algorithm and the support vector machines to find out those feature dimensions that are actively involved in representing the brain activities in a particular task. We show that using this lower dimensional space we can predict the cognitive state of the subjects in a particular task more accurately.