Multi-View Graph Contrastive Learning via Adaptive Channel Optimization for Depression Detection in EEG Signals.

Area of Science:

• Bioinformatics
• Neuroscience
• Machine Learning

Background:

• Automated depression detection using Electroencephalogram (EEG) signals is a growing field.
• Existing methods face challenges with data redundancy, underutilization of spatial information, and lack of interpretability.

Purpose of the Study:

• To propose a novel model, Multi-view Graph Contrastive Learning via Adaptive Channel Optimization (MGCL-ACO), for enhanced depression detection from EEG signals.
• To address limitations in current models by incorporating data redundancy reduction, spatial proximity, and functional connectivity.
• To improve the interpretability of depression detection models.

Main Methods:

• Adaptive channel optimization to eliminate redundant noise by maximizing mutual information.
• Multi-view graph construction integrating functional connectivity and spatial proximity.
• Graph convolutions and contrastive learning for fine-grained feature extraction.
• Visualization of channel significance for model interpretability.

Main Results:

• The proposed MGCL-ACO model demonstrated superior performance compared to state-of-the-art methods on public datasets.
• Effective reduction of data redundancy and improved feature extraction by integrating multi-view graph learning.
• Successful visualization of channel significance, enhancing model interpretability.

Conclusions:

• MGCL-ACO offers a promising approach for automated depression detection using optimized EEG signals.
• The model has the potential to enhance the accuracy and interpretability of depression diagnosis in clinical settings.
• This work advances the application of bioinformatics in mental health diagnostics.