Resting-state EEG microstates as biomarkers for major depressive disorder.

OBJECTIVES: To investigate the temporal dynamics of resting-state electroencephalography (EEG) microstates in patients with Major Depressive Disorder (MDD) and explore their potential as objective biomarkers for MDD diagnosis using machine learning techniques. METHODS: Resting-state EEG data were obtained from the MODMA dataset. EEG signals were preprocessed for artifact removal. Microstate analysis was performed using the Atomize and Agglomerate Hierarchical Clustering (AAHC) method, identifying four canonical microstate classes (A, B, C, and D). Microstate parameters, including duration, occurrence, and contribution, were compared between the MDD and control groups using statistical analysis. Additionally, seven significant microstate parameters were selected and used to classify MDD patients with machine learning models. RESULTS: MDD patients exhibited significantly shorter microstate durations and increased occurrences of microstates A and B. Microstate A showed significantly higher contribution in MDD patients. Machine learning classification based on microstate parameters achieved a maximum accuracy of 87.5 %, with LDA performing the best. CONCLUSIONS: EEG microstate analysis revealed altered temporal dynamics in MDD, indicating increased instability in brain activity. These findings suggest that EEG microstate parameters could serve as biomarkers for MDD diagnosis. The high classification accuracy of machine learning models further supports their potential for early and objective MDD detection.