Heat Diffusion Based Dissimilarity Analysis for Schizophrenia Classification

We apply shape analysis by means of heat diffusion and we show that dissimilarity space constructed using the features extracted from heat diffusion present a promising way of discriminating between schizophrenic patients and healthy controls. We use 30 patients and 30 healthy subjects and we show the effect of several dissimilarity measures on the classification accuracy of schizophrenia using features extracted by heat diffusion. As a novel approach, we propose an adaptation of random subspace method to select random subsets of bins from the original histograms; and by combining the dissimilarity matrices computed by this operation, we enrich the dissimilarity space and show that we can achieve higher accuracies.