EEG-based Depression Detection Using Convolutional Neural Network with Demographic Attention Mechanism

Electroencephalography (EEG)-based depression detection has become a hot topic in the development of biomedical engineering. However, the complexity and non-stationarity of EEG signals are two biggest obstacles to this application. In addition, the generalization of detection algorithms may be degraded owing to the influences brought by individual differences. In view of the correlation between EEG signals and individual demographics, such as gender, age, etc., and influences of these demographic factors on the incidence of depression, it would be better to incorporate demographic factors during EEG modeling and depression detection. In this work, we constructed an one-dimensional Convolutional Neural Network (1-D CNN) to obtain more effective features of EEG signals, then integrated gender and age factors into the 1-D CNN via an attention mechanism, which could prompt our 1-D CNN to explore complex correlations between EEG signals and demographic factors, and generate more effective high-level representations ultimately for the detection of depression. Experimental results on 170 (81 depressed patients and 89 normal controls) subjects showed that the proposed method is superior to the unitary 1-D CNN without gender and age factors and two other ways of incorporating demographics. This work also indicates that organic mixture of EEG signals and demographic factors is promising for the detection of depression.