Topological feature extraction from brain networks of schizophrenic patients based on EFC-GCN model

Abstract

To accurately extract the topological features of brain functional networks from electroencephalogram signals of patients with schizophrenia, this study proposes a functional brain network model that uses a graph convolutional neural network to integrate frequency bands, segment lengths, and functional connectivity indexes. It combines time-frequency domain features and brain network topological features to divide electroencephalogram signals into different frequency bands and length epochs. This allows for the analysis of characteristic values of electroencephalogram signals of schizophrenic patients. Through comparative trials, the study evaluates the suggested model’s performance. The results revealed that the proposed model obtained an average accuracy of 91.12% using the phase-locked value functional connectivity metric at Theta frequency band and 6-s segment length, which was 2.85%, 8.74%, and 3.95% higher compared to the support vector machine, convolutional neural network, and temporal convolutional network models, respectively, and demonstrated a superior recognition performance. In addition, network topology analysis revealed that the parietal region was an important region in the brain network of schizophrenic patients. The proposed model is reliable and feasible for detecting schizophrenic patients. It can extract topological features from electroencephalogram signals to characterize the functional brain networks of patients with schizophrenia. These features can then be correlated with clinical symptom salience, which is expected to aid in diagnosis.