A Multi-Branch Network for Depression Recognition Based on Temporal-Spatial Features of EEG Signals

Depression is a prevalent mental disorder worldwide, significantly impacting individuals' emotional and psychological states. Early detection and diagnosis of abnormalities remain a critical challenge. Electroencephalography (EEG), which records brain activity, has been extensively utilized in studies as a diagnostic tool for depression. However, many existing models lack consideration of the pathogenesis of depression, resulting in limited interpretability of the networks. Given that the majority of methods do not effectively utilize the temporal and spatial characteristics of EEG signals at the same time, this study proposes a multi-branch network designed to extract multi-scale temporal and spatial features concurrently. This model aggregates spatial input and dynamically models spatial interactions between electrodes using a dynamic graph convolutional network. In the meantime, the signals' temporal and spatial information are extracted using parallel convolutional layers. On publicly available datasets, the proposed method demonstrates superior classification performance compared to existing typical models. The suggested strategy offers a successful way to diagnose depression.