Detection of Schizophrenia Based on EEG Signal Using Vision Transformer Techniques

Schizophrenia is a serious and long-lasting condition characterized by disturbed beliefs, difficulties with thinking, and experiencing things that are not real, which impact emotions, behavior, and thoughts. Detecting and treating schizophrenia early is important to prevent long-term consequences. Electroencephalogram (EEG) data is a biological marker that can identify hidden changes in the brain during schizophrenia. However, EEG signals are unstable and have low intensity, making extracting meaningful information challenging. In this study, we propose using a vision transformer technique, which operates in the time-frequency domain, to detect schizophrenia automatically. The model was built and evaluated using three different validation approaches, including ten-fold cross-validation, with separate publicly available schizophrenia data sets. A new method for automated EEG-based schizophrenia detection has been developed using the DeiT model time-frequency (TF) input images. The method was evaluated on EEG recordings from 45 schizophrenia patients and 39 healthy controls, using morlet wavelet scattering. The overall accuracy of the method was 91% for subject-independent classification. The authors suggest that the ViT model could be used as a disease detection tool for not only schizophrenia but also other neurological symptoms. The suggested model's accuracy rate on dataset 2 was 84.64%, with 49 participants with schizophrenia and 32 healthy controls.