A Novel Depression Recognition Network Based on Brain Functional Connectivity from Prefrontal Emotional State EEG

Area of Science:

• Computational neuroscience
• Neuroimaging analysis
• Machine learning for healthcare

Background:

• Electroencephalogram (EEG) signals offer a noninvasive approach for detecting neural dysfunctions linked to depression.
• Existing methods face limitations due to small sample sizes and insufficient modeling of spatiotemporal brain activity patterns.

Purpose of the Study:

• To develop an advanced EEG-based framework for recognizing depression by capturing dynamic neural activity patterns.
• To overcome limitations in data size and spatiotemporal feature representation in current depression detection models.

Main Methods:

• Proposed DR-Net framework integrates generative adversarial networks (GANs) for data augmentation, transition propagation graph convolution networks (TPGCN) for dynamic spatiotemporal modeling, and transformers for temporal dependencies.
• Brain functional connectivity networks were constructed using the phase lag index.
• Emphasis on the frontal lobe region to enhance feature discriminability and reduce dimensionality.

Main Results:

• Achieved 98.7% accuracy in depression recognition, outperforming baseline and state-of-the-art methods.
• Clinical dataset validation demonstrated strong robustness and generalizability with 96.13% accuracy.
• The method showed no significant class bias.

Conclusions:

• DR-Net effectively models EEG spatiotemporal dynamics, offering a reliable and generalizable approach for objective depression diagnosis.
• The framework advances computational neurodiagnostics by integrating dynamic graph propagation, self-attention, and data augmentation.
• DR-Net shows potential for supporting future EEG-based clinical assessments of depression.