U-Net Based Estimation of Functional Connectivity from Time Series Multi-Channel EEG from Schizophrenia Patients

Human brain, as a complex network, is affected by many mental disorders and neurodegenerative diseases. Brain functional and structural alteration could be captured by imaging modalities such as CT and MR imaging. However, these modalities have low sensitivity to properly capture the brain connectivity map as a biomarker for the early diagnosis of neurodegenerative diseases. In this light, complimentary examination tools (such as EEG) are employed to estimate the brain functional connectivity (FC) map. To decode the brain FC map from EEG signals, conventional approaches rely on hand-craft feature extraction, leading to suboptimal performance/effectiveness. In this light, this work set out to implement a novel deep neural network based on U-Net to extract the brain FC maps and identify (based on the obtained FC map) the type of neurodegenerative disease from the patient's EEG signals. Due to the absence of the ground truth brain FC maps, the proposed approach extracts the patient-specific brain FC maps in an unsupervised approach. To evaluate the performance of the proposed deep learning model, a publicly available dataset of EEG signals from healthy control and schizophrenia patients was employed. The proposed model exhibited an accuracy of 94.11% to classify schizophrenia patients. Moreover, the estimated brain FC maps for both healthy control and schizophrenia patients exhibited highly discriminative patterns to differentiate abnormalities from the healthy controls. The proposed unsupervised model, which is applicable to EEG and functional MR data, exhibited promising performance to extract brain FC maps and classify neurodegenerative diseases.