Adolescent depression recognition and symptom prediction based on multi-view EEG network: an exploratory study

Area of Science:

• Neuroscience
• Machine Learning
• Psychiatry

Background:

• Adolescent depression diagnosis lacks objective neurobiological biomarkers, relying on subjective self-report.
• Current diagnostic methods are limited by their reliance on questionnaires.

Purpose of the Study:

• Compare the diagnostic performance of traditional psychological scales with a novel Multi-view Adaptive Graph Convolutional Network (MVA-GCN) using resting-state EEG.
• Explore associations between MVA-GCN-derived brain network features and psychological resilience in adolescents.

Main Methods:

• Collected resting-state EEG data from adolescents with major depressive disorder (MDD) and healthy controls (HCs).
• Developed an MVA-GCN model integrating phase-locking value (PLV), Pearson correlation coefficient (PCC), and phase lag index (PLI) connectivity views.
• Compared MVA-GCN performance against questionnaire-based machine learning models and examined correlations with clinical measures.

Main Results:

• MVA-GCN achieved 99.84% accuracy, significantly outperforming questionnaire-based models (86.43%).
• Occipital and fronto-central regions were key predictors; increased gamma-band connectivity and reduced alpha-band power were noted.
• Exploratory analyses suggested potential group-specific brain-resilience associations, with deviations correlating with symptom severity in the MDD group.

Conclusions:

• MVA-GCN shows promising discriminative capability for adolescent depression in an exploratory sample.
• Brain-resilience association findings require caution due to lack of statistical significance after multiple comparisons.
• MVA-GCN-derived features warrant further investigation in larger cohorts but are not yet clinically validated.