Application of Deep Learning Models in the Detection Depression Using Time-Frequency Transformation of Electroencephalogram Signals

Major Depressive Disorder (MDD) is a prevalent mental disorder worldwide, and timely diagnosis is necessary for efficient treatment. In the present study, an electroencephalogram (EEG) signal was utilized to automatically and precisely detect MDD using deep learning models.

Thirty MDD and twenty-eight healthy subjects participated, and their psychological evaluation was conducted by a specialist psychiatrist using the standard Beck questionnaire. 19-channel EEG signals were acquired from all participants in a resting state with eyes closed. Short-Time Fourier Transform (STFT) was applied to the sequential segments of the EEG signals and resulted two-dimensional matrix fed to the deep learning models. DeepEEGNet model was developed based on the EEGNet model utilized for the MDD classification and Healthy subjects. A holdout data was used to test the final model.

The DeepEEGNet model proposed in this study classified MDD and healthy participants with 84.1% accuracy, 86% sensitivity, and 82.7% specificity.

The deep learning model proposed in this study could accurately classify Healthy subjects and MDD patients using EEG signals and can be utilized as a helpful tool by psychiatrists.