New Feature for Schizophrenia Classification Based on Functionally Homogeneous Brain Regions

In this work, we investigate functionally homogeneous regions segmentation method (FHR) to obtain features for binary classification of patients with schizophrenia and healthy controls using support vector machine classifier (SVM) based on resting-state functional magnetic resonance imaging (rs-fMRI) data. For comparison, we used 4 feature-type approaches: functional connectivity maps (FCM), Amplitude of low frequency fluctuations (ALFF) and fractional amplitude of low frequency fluctuations (fALFF), Regional Homogeneity (ReHo). Four different feature selection algorithms were used (χ2, F_test, L1 and L2). SVM classifier was trained and tested on a rs-fMRI dataset of 36 patients with schizophrenia and 36 healthy controls, obtained using Siemens Magnetom Verio MRI 3TL scanner. The best results were achieved by features obtained by the ReHo approach (93% accuracy) and the FHR approach (91% accuracy). The ReHo approach showed best accuracy with χ2 and F test feature selection algorithms, and the FHR approach showed best accuracy with L1 and L2 feature selection algorithms.