Machine learning-assisted resting-state electroencephalography improves diagnostic accuracy in psychiatric disorders: A narrative review

Psychiatric diagnoses remain largely symptom-based and lack validated biological markers. Resting-state electroencephalography offers a non-invasive, low-cost, and portable tool for assessing large-scale brain dynamics, while machine learning enables the extraction of complex, multivariate patterns from high-dimensional neural data. This narrative review synthesizes recent advances from 34 original studies published between 2015 and June 2025 that applied machine learning to resting-state electroencephalography for the diagnosis, stratification, and prognosis of major psychiatric disorders. Across studies, machine learning models trained on electroencephalography-derived features—including spectral power, entropy, microstates, and connectivity measures—showed diagnostic accuracies ranging from 74% to 99%. In major depressive disorder, deep-learning and connectivity-based models achieved > 95% accuracy in small samples, which was correlated with symptom severity. Bipolar disorder studies highlighted entropy and multiscale complexity as candidate biomarkers, supporting electroencephalography–machine learning pipelines for differential diagnosis. In schizophrenia, microstate and network-based models revealed reproducible alterations linked to cognitive and symptom dimensions. Evidence from generalized anxiety and obsessive–compulsive disorder suggested electroencephalography markers of severity and chronicity, while transdiagnostic analyses indicated shared electrophysiological signatures of cognitive dysfunction and vulnerability. Resting-state electroencephalography combined with machine learning represents a promising avenue for developing objective, scalable biomarkers across psychiatric conditions. However, most studies rely on small, single-site datasets with heterogeneous preprocessing. Future progress requires standardized analytic pipelines, multimodal integration, and multicenter replication for robust clinical translation.