SZ-RAN: A Residual Attention Network for Early Detection of Schizophrenia using EEG Signals

Schizophrenia (SZ) is a complex and debilitating mental disorder which affects 1% of the global population. Electroencephalogram (EEG) emerged as a promising, low-cost and non-invasive tool for recording, understanding and scrutinizing the neurological underlying factors of SZ to refine diagnostic procedures. Thus, a novel lightweight deep learning architecture named SZ-RAN is proposed for early SZ recognition using EEG signals. SZ-RAN incorporates the advantage of both residual connection and attention mechanism. Residual connections aid in removing vanishing gradient problems and attention mechanism focuses on significant features and reduces transmission of irrelevant features. This leads to fast convergence of the model. An IBIB-PAN dataset is utilized to test SZ-RAN and the model achieved a classification accuracy of 99.9%. Further, comparing the proposed work with five recent techniques prove the effectiveness of the proposed technique in medical applications.