A Hybrid Generative/Discriminative Method for Classification of Regions of Interest in Schizophrenia Brain MRI

Schizophrenia research based on magnetic resonance imaging(MRI) traditionally relies on the volumetric analysis of brain matter,either characterizing the whole intracranial volume or studying the attributesof small regions of interest (ROI), corresponding to well-knownfunctional parts in the brain. In this work, we addressed the secondscenario, proposing a novel approach able to automatically distinguishschizophrenic patients from normal controls using multiple ROIs. Weexplore a hybrid generative/discriminative approach, exploiting state ofthe art generative models via Fisher kernel and support vector machines(SVM). Experimental results, on a dataset of 124 subjects and 7 ROIs,are really encouraging, also in comparison with pure discriminative methods.Moreover, our results find some agreements with previous medicalstudies in schizophrenia research.