EDT: An EEG-based attention model for feature learning and depression recognition

Numerous existing studies on machine learning-based depression recognition have focused on the frequency domain features of EEG data. Furthermore, their experiments have demonstrated the importance of frequency domain features of EEG data for depression detection. However, in the field of deep learning-based depression recognition, the frequency domain information of EEG data has received relatively limited attention. In this study, we propose a deep learning model named EEG-based Depression Transformer (EDT), which can extract features from the frequency, spatial, and temporal domains of EEG data and distinguish individuals with depression from healthy controls. We develop a specialized module for the extraction of frequency domain features. Our model combines the advantages of attention mechanism and convolutional neural network. We evaluate the performance of EDT, baseline models, and various EDT variants through ten-fold cross-validation. The experimental results indicate that EDT achieves a superior accuracy (92.25 ± 4.83 %) compared to other models.