Major Depressive Disorder Classification Based on Different Convolutional Neural Network Models: Deep Learning...

Area of Science:

• Computational Neuroscience
• Neuroimaging
• Artificial Intelligence

Background:

• Major Depressive Disorder (MDD) lacks specific diagnostic biomarkers.
• Evaluating brain's structural and functional connections is crucial for neurodegenerative diseases.
• Deep learning (DL) shows promise for identifying translational biomarkers in mood disorders.

Purpose of the Study:

• To develop an electroencephalography (EEG)-based diagnosis model for MDD using deep convolutional neural networks (CNNs).
• To explore the potential of DL in identifying spatial and temporal features for MDD diagnosis.
• To compare the performance of different DL architectures (ResNet-50, MobileNet, Inception-v3) for MDD classification.

Main Methods:

• EEG recordings from 46 MDD patients and 46 healthy controls were analyzed.
• Data from 19 electrodes across four frequency bands (Δ, θ, α, β) were utilized.
• Three deep CNN architectures (ResNet-50, MobileNet, Inception-v3) were employed for classification.

Main Results:

• MobileNet architecture achieved 89.33% and 92.66% classification accuracy using location data.
• The delta frequency band with ResNet-50 showed 90.22% predictive accuracy and an AUC of 0.9.
• Distinctive spatial and temporal features were delineated to differentiate MDD subjects from controls.

Conclusions:

• DL models, particularly CNNs, show significant potential for accurate and rapid diagnosis of MDD.
• EEG-based DL analysis can serve as a valuable translational biomarker for mood disorders.
• Computational methods offer advantages in speed and accuracy over traditional diagnostic approaches for psychiatric disorders.