LSDD-EEGNet: An efficient end-to-end framework for EEG-based depression detection.

[Display omitted] • First, we collected EEG signals from 40 depressed patients and 40 healthy controls and eliminated the noise through wavelet transformation. • Second, the EEG signals are divided into 5 frequencies and fed into mutiple models to evaluate the performance. • Third, we propose a model named LSDD-EEGNet for EEG-based depression detection, which connects CNN and LSTM layer as the extractor for EEG signals and introduces the domain discriminator to deal with the feature shift problem between training dataset and test dataset. • Finally, we evaluate the performance of LSDD-EEGNet and compare the model with other representative models. Depression is a mood disorder that causes negative effects on people's life and has become a leading health burden worldwide. But the effective and low-cost detection for depression is still a great challenge. Electroencephalogram (EEG) measures the brain activities and can be used for depression-related research. In this paper, we propose an effective end-to-end framework named LSDD-EEGNet for EEG-based depression detection. Specially, LSDD-EEGNet has two distinguishing characteristics for depression recognition: (1) Considering the superiority of convolution neural network (CNN) on feature extraction and the efficiency of long-short term memory (LSTM) for time-series signals, we combine both as the extractor for LSDD-EEGNet. (2) We apply the domain discriminator to modify the data representation space and eliminate the discrepancy between training and test dataset. In addition, we collected EEG signals from 40 depressed patients (DPs) and 40 healthy controls (HCs) to evaluate the performance of the proposed deep framework for depression detection. Compared to other typical machine learning (ML) methods and deep learning (DL) models, LSDD-EEGNet achieves superior performance on subject-independent evaluation that shows the LSDD-EEGNet can be a promising detection method for depression. [ABSTRACT FROM AUTHOR]