Systematic review of machine learning and deep learning models for EEG-based detection of depression.

Clinical Snapshot

Key Findings

• P Value: Non-significant in exploratory non-parametric comparison between ML and DL accuracy distributions (exact statistic and p-value not specified in abstract)

• Effect Size: DL mean reported accuracy: 93.92%; ML mean reported accuracy: 90.78%; overall accuracy range across all studies: approximately 76%–100%

• Primary Outcome: Classification accuracy for EEG-based depression detection using ML or DL algorithms across 42 included studies (23 ML-based, 19 DL-based)

• Nnt Or Sensitivity: Sensitivity and specificity not pooled; individual study values reported descriptively. No NNT calculable. SROC curve analysis not performed. QUADAS-2 identified high risk of bias in patient selection and index test domains across multiple studies.

• Confidence Interval: Not reported — no pooled estimates or confidence intervals calculated; descriptive averages only

Clinical Application

EEG acquisition is feasible in specialist psychiatric, neurological, and sleep medicine settings, but is not routinely available in primary care. Clinical-grade QEEG systems with integrated ML/DL pipelines are not yet commercially validated or regulatory-approved for depression diagnosis. Implementation would require standardised EEG protocols, trained operators, and validated software — none of which are currently standardised across the included studies. In Australia, EEG is Medicare-rebatable (MBS item numbers 11000–11027) for neurological indications but is not currently indicated or rebated for psychiatric diagnosis of depression. The TGA has not approved any ML/DL-based EEG diagnostic device for depression. RACGP guidelines for depression management (including the RACGP Red Book and beyondblue clinical practice guidelines) rely on validated clinical instruments (PHQ-9, DASS-21) and structured clinical assessment — not biomarker-based tools. PBS-listed antidepressant prescribing does not require EEG confirmation. This technology remains investigational in the Australian context, and any future clinical translation would require TGA regulatory approval, MBS pathway development, and prospective validation in Australian populations. The review's findings do not alter current Australian clinical practice. Adults with suspected or confirmed depressive disorder in whom objective, scalable biomarker-based diagnostic adjuncts are being considered. The evidence base does not yet support application in any specific clinical subgroup (e.g., treatment-resistant depression, adolescents, elderly) given the heterogeneity and methodological limitations of included studies.

Abstract

OBJECTIVE: Depression is a leading cause of global disability, motivating the development of objective and scalable diagnostic approaches. Quantitative electroencephalography (QEEG) combined with machine learning (ML) and deep learning (DL) techniques has gained increasing attention for depression detection. This systematic review aimed to critically examine and descriptively compare the methodologies, performance metrics, and limitations of ML- and DL-based models applied to EEG data for depression detection. METHODS: A systematic review was conducted in accordance with PRISMA 2020 guidelines. Seven electronic databases (PubMed, Scopus, IEEE Xplore, ScienceDirect, Web of Science, SAGE Journals, and MDPI) were searched for peer-reviewed studies published between 2020 and 2024. Eligible studies included human participants, used EEG signals for depression detection, and applied ML or DL algorithms. Extracted information comprised algorithm type, sample size, EEG acquisition parameters, validation strategies, and reported performance metrics, which were synthesized descriptively across studies. Risk of bias was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. RESULTS: A total of 42 studies met the inclusion criteria, including 23 ML-based and 19 DL-based investigations. Reported classification accuracy ranged from approximately 76% to 100%. DL studies showed a higher mean reported accuracy than ML studies (93.92% vs. 90.78%); however, this difference was not statistically significant in the exploratory non-parametric comparison. Near-perfect performance values were frequently observed in studies with small sample sizes and subject-dependent or exclusively internal validation strategies, raising concerns regarding overfitting and limited generalizability. Studies relying on publicly available datasets tended to report more stable performance. QUADAS-2 assessment revealed recurrent risk-of-bias concerns, particularly in the domains of patient selection and index test conduct. CONCLUSIONS: Both ML and DL approaches demonstrate potential for EEG-based depression detection, but reported performance differences between them should be interpreted cautiously. Although DL studies tended to report higher accuracy values, this pattern was not statistically significant in exploratory analyses and was strongly influenced by sample size, validation strategy, and methodological design. Future research should prioritize larger and more diverse samples, subject-independent or external validation strategies, and standardized reporting frameworks to enhance methodological rigor and clinical applicability.

References

1. 1.Luna-Guevara, G. R., Vargas-Huicochea, I., & Rodríguez-Machain, A. C. (2026). Systematic review of machine learning and deep learning models for EEG-based detection of depression. Journal of Psychiatric Research. https://doi.org/10.1016/j.jpsychires.2026.04.030