EEG Rhythm-Based Functional Brain Connectivity for Automated Detection of Schizophrenia Employing Deep Learning

In the present contribution, a novel framework for automated detection of healthy and schizophrenic (SCZ) electroencephalogram (EEG) signals is proposed employing multiplex weighted visibility graph (MWVG)-aided functional brain connectivity analysis and deep residual network (ResNet). For this purpose, EEG signals recorded from different regions of the brain using multichannel EEG system, have been channel-wise decomposed into different frequency bands known as brain rhythms. Following this, for each rhythm, a novel approach for construction of functional brain connectivity for both healthy and SCZ patients is proposed using inter-layer similarity of nodal local efficiency (LE) measures. The red-green-blue (RGB) images of rhythm-wise brain connectivity patterns obtained for healthy and SCZ patients were finally fed to a 19-layer customized lightweight ResNet model for automated feature extraction and classification purpose. It was observed that the brain connectivity patterns for each brain rhythm showed significant alterations between healthy and SCZ patients. Further, it was also observed that for the alpha brain rhythm, distinct difference is perceived, which yielded highest detection accuracy of 98.72% and 99.93%, respectively for two publicly available benchmark datasets.