Discriminating between major depressive disorder and bipolar depression: Aberrant EEG microstate dynamics and machine learning classification.

BACKGROUND: Major depressive disorder (MDD) and bipolar depression (BD) are common mood disorders with overlapping clinical features, posing significant challenges for accurate diagnosis and effective treatment. Electroencephalography (EEG) microstates reflect transient, quasi-stable patterns of brain activity that index fast, large-scale neural network dynamics and may provide novel insights into the neural abnormalities associated with mood disorders. METHODS: In this study, 210 participants (78 MDD, 45 BD, and 87 healthy controls) completed demographic, clinical, and microstate assessments. Resting-state EEG microstate features were analyzed and used in machine learning models to classify MDD versus BD, MDD versus HCs, and BD versus HCs. RESULTS: MDD patients showed higher microstate C metrics, lower microstate D metrics, increased transition probabilities from B to C, and reduced transition probabilities between B and D, suggesting enhanced sequential activation from the occipital visual cortex to the default mode network but disrupted sequential activation from the visual cortex to the executive control network. BD patients primarily showed significantly longer microstate B duration, indicating excessive visual network activity. Microstate-based machine learning models showed moderate to good discriminative performance, with Area Under the Curve (AUC) values of 83.4% (MDD versus BD), 86.0% (MDD versus HCs), and 93.3% (BD versus HCs). LIMITATIONS: The modest sample size may restrict generalizability, and refined methodological approaches could further enhance classification performance. CONCLUSIONS: These findings provide preliminary insight into neural alterations in MDD and BD, suggest potential diagnostic relevance of EEG microstates, and may inform future hypothesis-driven research on disorder-specific interventions.