Hybrid CNN-LSTM-SVM Model to Detect MDD from EEG Signals

Abstract

According to World health organization 5.7 % of adult suffer from depression around the world and these people die 10 to 20 year earlier then general population. Stressful environment or personal factors may be major cause of depression. The diagnosis of depression exclusively depends on interaction between doctor and patient as there is no direct lab test available to detect depression. Electroencephalograms (EEG) are non-invasive and has potential to capture brain behaviour in form of signals. In our study an attempt was made to distinguish between Major Depressive Disorder (MDD) patients and healthy control. The EEG data used in this work is publicly available dataset, consisting of 19 channel recordings from both 34 MDD patients and 30 healthy subjects under controlled conditions. EEG recording with closed eyes was used in this experiment. EEG signals have a specific range for healthy person but it altered in MDD patient. On these EEG recording, time domain and frequency domain features were extracted. Random Forest, XGBoost, Stacking Ensemble, EEGNet, CNN-LSTM and proposed CNN-LSTM-SVM were applied. Highest subject level accuracy was attained by proposed method 96.43 % followed by CNN-LSTM (90%). Other performance parameters like F1-score (0.96), Sensitivity (0.96), Specificity (0.97), AUC-ROC (0.98), kappa value (0.93) which showed proposed model CNN-LSTM-SVM overperformed other models. Our system is designed as a decision-support tool for MDD assessment.