A Novel Brain Connectivity-Powered Graph Signal Processing Approach for Automated Detection of Schizophrenia from Electroencephalogram Signals

Schizophrenia is a severe neural disorder that affects around 24 million individuals globally. In this context, Electroencephalogram (EEG) signal-based analysis and automated screening for Schizophrenia (SZ) have gained importance. EEG-based Schizophrenia (SZ-EEG) analysis is traditionally done by extracting features from individual EEG electrodes' signals and utilizing these features for Machine Learning (ML)-based classification models. However, these methods do not exploit the Schizophrenia-induced alteration of functional brain connectivity between neuronal masses. The present study proposes a novel graph-signal (GS) representation of multi-channel SZ-EEG data that fully encompasses local brain activation and global interactions between brain regions. The proposed GS representation comprises the underlying connectivity network and the signal values on the network's vertices. Here, the EEG signal's entropy at each electrode is used as GS values, and a phase lag index (PLI)-based functional connectivity measure is utilized as the underlying connectivity network. Further, these connectivity-informed GSs are transformed to the spectral domain by the Graph Fourier Transform (GFT), and relevant discriminatory features are extracted from them using the Graph Signal Processing (GSP) technique. Those features are fed to basic ML-based classification models. The efficacy of the proposed PLI-GSP framework is validated using a publicly available SZ-EEG dataset, and a 99.77% classification accuracy is achieved that outperforms most of the state-of-the-art models.