Identification of Schizophrenia and Depression Based on Deep Learning and EEG

Schizophrenia (SCH) and depression (DEP) may be misdiagnosed, mainly because their symptoms have certain similarities and are susceptible to emotional changes. In this study, a deep learning model (CEEGA-Net) based on convolutional neural network (CNN), gated circulation unit (GRU) and attention mechanism (AM) is proposed to classify EEG signals from SCH and DEP to improve the diagnostic accuracy of patients with these two types of diseases. In this paper, a deep learning model that can automatically extract and classify important features was adopted. Data from Peking University Sixth Hospital, a total of 154 participants. The experimental paradigm was auditory oddball task. The local features are extracted by convolutional neural networks, then the features are sequentially modeled and weighted by the GRU layer and the attention mechanism, and finally the fully connected layer is used for classification. Our CEEGA-Net model has an average accuracy of 98.3% for trial-based classification and 85.1% for subject-based classification under the ten-fold cross-validation method. The results of the study show that CEEGA-Net can distinguish SCH and DEP well under this experimental paradigm, and the high accuracy and reliability of the model may help doctors to distinguish these two diseases more accurately in clinical practice.