Bi-LSTM-deep CNN for schizophrenia detection using MSST-spectral images of EEG signals

Schizophrenia (SZ) is a chronic mental disorder associated with functional impairment of human brain. Early-stage SZ detection can lead to better treatment, improving the quality of life of patients suffering from this disease. This work proposes an automated computer-aided diagnosis (CAD) system for SZ detection using multichannel EEG activity. The nonlinear and nonstationary nature of EEG signals require time-frequency domain analysis for accurate diagnosis of the diseased condition. In this research, an automated multisynchrosqueezing transform (MSST)-based Bi-CNN model is proposed for SZ disease detection. The proposed methodology consists of 2D time-frequency representation of EEG activity using MSST, and further, features extraction and classification of the decomposed EEG is performed by using the Bi-CNN model for SZ detection. The classification results obtained using the proposed MSST-Bi-CNN methodology indicate an accuracy of 84.42%. The contribution to the body of knowledge is a method for clinicians for early SZ disease detection and screening method based on inductive learning on nonstationary data processed with multisynchrosqueezing transformation.