Depression Diagnosis Modeling With Advanced Computational Methods: Frequency-Domain eMVAR and Deep Learning.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) signals are complex, non-linear, and non-stationary, requiring advanced analytical methods beyond traditional linear approaches.
• Automated EEG-based depression diagnosis holds promise for early and accurate detection of mood disorders.
• Non-linear methods are effective in analyzing biological signals like EEG, capturing intricate brain activity patterns.

Purpose of the Study:

• To develop and evaluate a novel methodology for EEG-based depression diagnosis.
• To compare a hybrid deep learning (DL) approach with a conventional machine learning (ML) framework using extended multivariate autoregressive (eMVAR) features.
• To assess the efficacy of advanced computational techniques in identifying depression-specific information from EEG signals.

Main Methods:

• A hybrid DL model combining a pretrained ResNet-50 and long-short term memory (LSTM) was developed.
• An eMVAR framework was employed to extract 8 causality measures (DC, DTF, PDC, gPDC, eDC, dDC, ePDC, dPDC) from EEG data.
• The performance of the DL model was compared against the eMVAR-based ML framework.

Main Results:

• The eMVAR framework achieved classification accuracies ranging from 84% to 95.9%, with partial directed coherence (PDC) and delayed PDC (dPDC) showing the highest performance.
• The DL framework (ResNet-50 + LSTM) achieved a classification accuracy of 90.22%.
• The DL methodology demonstrated competitive performance compared to the feature extraction-based ML methods.

Conclusions:

• The proposed DL methodology is a viable and competitive alternative for EEG-based depression classification.
• Advanced computational techniques, including DL and eMVAR, are effective in analyzing complex EEG signals for mood disorder detection.
• Further research into DL models can enhance the accuracy and efficiency of automated depression diagnosis systems.