Clustering-Fusion Feature Selection Method in Identifying Major Depressive Disorder Based on Resting State EEG...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Depression is a complex syndrome with significant individual variability.
• Effective feature selection methods are crucial for accurately recognizing depression by capturing both commonalities and differences within patient groups.
• Existing methods may not fully address the heterogeneity inherent in depression.

Purpose of the Study:

• To propose and evaluate a novel clustering-fusion feature selection method (HCSNF) for depression recognition.
• To improve the classification accuracy of depression using electroencephalography (EEG) data.
• To identify discriminative features and brain network characteristics associated with depression.

Main Methods:

• Utilized Hierarchical Clustering (HC) to analyze subject heterogeneity.
• Employed Average and Similarity Network Fusion (SNF) algorithms to characterize brain network atlases.
• Applied differences analysis to identify discriminant features in EEG data (sensor and source layers).

Main Results:

• The proposed HCSNF method outperformed traditional feature selection techniques in depression recognition.
• Achieved optimal classification results for electroencephalography (EEG) data across sensor and source layers.
• Demonstrated a significant improvement (over 6%) in classification performance in the beta band at the sensor layer.
• Identified long-distance connections between the parietal-occipital lobe and other regions as highly discriminative and correlated with depressive symptoms.

Conclusions:

• The HCSNF method offers a robust approach for depression recognition, particularly with EEG data.
• This study provides methodological guidance for discovering reproducible electrophysiological biomarkers for depression.
• Findings offer new insights into the neuropathological mechanisms underlying heterogeneous depression.