Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 144 of the 152 references it cites were found in a registry or an index; 5 of the 13 sources its claims cite were checked for support.

AI-assisted multi-modal information for the screening of depression: a systematic review and meta-analysis

Open access

Abstract

Depression is a prevalent and costly mental disorder across all ages. Artificial intelligence (AI)-assisted physiological and behavioral information-such as electroencephalography (EEG), eye movement, video or audio monitoring, and gait analysis-offers a promising tool for depression screening. We systematically reviewed the classification performance of these AI-assisted measures in depression screening. A comprehensive literature search was conducted in Google Scholar, Web of Science, and IEEE Xplore, with the search date up to June 7, 2025. The reported AUC values are pooled estimates calculated from all results of eligible studies. AI-assisted multi-modal methods achieved a pooled AUC of 0.95 (95% CI: 0.92-0.96), outperforming uni-modal methods (pooled AUC: 0.84-0.92). Subgroup analysis indicated deep learning models showed higher performance, with an AUC of 0.95 (95% CI: 0.93-0.97). These findings highlight the potential of AI-based multi-modal information in depression screening and emphasize the need to establish standardized databases and improve research design.

Overview

CoChat Summary

## TL;DR

This systematic review and meta-analysis found that AI-assisted multi-modal approaches for depression screening performed better than uni-modal approaches, with pooled AUC up to 0.95 and 0.96 for the highest-performing multi-modal results. The authors conclude that multi-modal AI is promising, but larger standardized datasets and stronger study designs are needed.

## Key findings

  • The search identified 1,564 records; after removing 981 duplicates and excluding 108 by title/abstract and 397 at full text, 80 studies were included in the qualitative synthesis and meta-analysis.
  • For all multi-modal depression screening studies, the pooled performance was sensitivity 0.88, specificity 0.90, and AUC 0.95 (95% CI: 0.92–0.96); selecting the highest-performing contingency tables gave sensitivity 0.89, specificity 0.91, and AUC 0.96 (95% CI: 0.94–0.97).
  • Uni-modal pooled AUCs were lower: EEG 0.92, eye movement 0.88, video 0.86, audio 0.92, and gait 0.84; the corresponding highest-performing contingency tables ranged up to AUC 0.95 for audio and 0.94 for EEG.
  • Deep learning showed higher performance than traditional machine learning in subgroup analysis, with an AUC of 0.95 (95% CI: 0.93–0.97); the abstract states multi-modal methods outperformed uni-modal methods, which had pooled AUCs of 0.84–0.92.

## Methods

  • Systematic review and meta-analysis conducted according to PRISMA and registered in PROSPERO (CRD420251049107).
  • Searches were run in Google Scholar, Web of Science Core Collection, and IEEE Xplore up to June 7, 2025; only English-language technical studies on AI/ML/DL-based depression classification with extractable classification outcomes were included.
  • The authors extracted or derived 2 × 2 contingency tables from reported accuracy/sensitivity/specificity/recall/precision/confusion matrices, then used random-effects meta-analysis, SROC curves, forest plots, and I² heterogeneity statistics.

## Limitations

  • The paper states several limitations: only English-language articles were included; many pooled estimates were based on a limited number of studies and small samples; subgroup analyses still had heterogeneity; few studies examined cross-ethnic or cross-age generalizability; studies reporting only continuous metrics or only F1 scores without TP/TN/FP/FN were excluded.
  • Evident from the text: all included studies used retrospective data, and nearly half of the studies (n = 38) had high or unclear risk of bias in patient selection, which may limit confidence in the pooled estimates.
  • Evident from the text: most public and private datasets had no more than 280 subjects (range 24–275), which likely constrains model robustness and generalizability.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41746-025-01933-3

Cites

144 of 144 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Brain activation patterns in depression across key functional domains. Brain activation patterns in depression across key functional domains, including altered emotional processing, negative attentional bias, delayed auditory responses, and impaired sensorimotor-emotional integration. BG Basal Ganglia, ATL Anterior Temporal Lobe, FFA Fusiform Face Area, VP Ventral Pallidum, SC Superior Colliculus, OC Occipital Cortex.
Fig. 1 Brain activation patterns in depression across key functional domains. Brain activation patterns in depression across key functional domains, including altered emotional processing, negative attentional bias, delayed auditory responses, and impaired sensorimotor-emotional integration. BG Basal Ganglia, ATL Anterior Temporal Lobe, FFA Fusiform Face Area, VP Ventral Pallidum, SC Superior Colliculus, OC Occipital Cortex.CC BY-NC-ND · PubMed Central
Fig. 2 PRISMA flowchart of study selection. Illustration of the identification, screening, eligibility assessment, and inclusion of studies in the systematic review.
Fig. 2 PRISMA flowchart of study selection. Illustration of the identification, screening, eligibility assessment, and inclusion of studies in the systematic review.CC BY-NC-ND · PubMed Central
Fig. 3 Hierarchical SROC curves of studies in meta-analysis of depression and NC classification using different paradigms. a The SROC curves comparing uni-modal and multi-modal approaches. b The SROC curves stratified by deep learning and machine learning methods.
Fig. 3 Hierarchical SROC curves of studies in meta-analysis of depression and NC classification using different paradigms. a The SROC curves comparing uni-modal and multi-modal approaches. b The SROC curves stratified by deep learning and machine learning methods.CC BY-NC-ND · PubMed Central
Fig. 4 Outlook for multi-modal physiological and behavioral information integration. Illustration of Multi-modal Fusion, AI Model Updating, Dataset Standardization, and Cross-group Research in Diverse Applications.
Fig. 4 Outlook for multi-modal physiological and behavioral information integration. Illustration of Multi-modal Fusion, AI Model Updating, Dataset Standardization, and Cross-group Research in Diverse Applications.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

4 Valid · 1 Not valid · 5 Not checked · 39 Own results and interpretations

21 more claims

Create a free account to see all 49 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 13
  • It may state two separate claims as one. 14

Create a free account to see these claims.

Citations

Cites

Show all 152 references

Research works printed but not linked: 5

  1. Bai, R., Guo, Y., Tan, X., Feng, L. & Xie, H. An EEG-based depression detection method using machine learning model. Int. J. Pharma Med. Biol. Sci. 10, 17–22 (2021).
  2. Prokhorenkova, L., Gusev, G., Vorobev, A., Dorogush, A. V. & Gulin, A. CatBoost: unbiased boosting with categorical features. Adv. Neural Inf. Process. Syst. 31 (2018).
  3. Gönen, M. & Alpaydın, E. Multiple kernel learning algorithms. J. Mach. Learn. Res. 12, 2211–2268 (2011).
  4. Lundberg, S. M. & Lee, S.-I. A unified approach to interpreting model predictions. Adv. Neural Inf. Process. Syst. 30 (2017).
  5. Flores, R., Tlachac, M., Toto, E. & Rundensteiner, E. Audiface: multimodal deep learning for depression screening. In Proceedings Machine Learning for Healthcare Conference 609–630 (2022).

Cited by

See all 33

Metadata

Authors
Wang, L., Wang, C., Li, C., Murai, T., Bai, Y., Song, Z., Zhang, S., Zhang, Q., Huang, Y., Bi, X., & Jiang, J.
Affiliations
  • Shanghai University (CN)
  • Kyoto University (JP)
  • Second Military Medical University (CN)
Corresponding author
Jiang, Jiehui
Venue
npj Digital Medicine
Published
2025-08-16
Volume
8
Issue
1
Pages
523-523
ISSN
2398-6352 (ISSN-L)
OpenAlex
W4413270591
Cited by
33
References
152
Citations by year
2026: 28 · 2025: 5
FWCI
5.44
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Digital Medicine (publishedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Shanghai Industrial Collaborative Innovation Project (No. XTCX KJ-2023-37)
  • Science and Technology Innovation 2030 - Major Projects (2022ZD021600)
  • National Natural Science Foundation of China (No. 62376150)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
  • Life Sciences › Neuroscience › Cognitive Neuroscience › EEG and Brain-Computer Interfaces
Keywords
depression screening, artificial intelligence, multimodal data, EEG, eye movement analysis, gait analysis, deep learning, classification performance, AUC

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: AI-assisted multi-modal information for the screening of depression: a systematic review and meta-analysis. https://app.cochat.ai/papers/10.1038/s41746-025-01933-3
The paper
Wang, L., Wang, C., Li, C., Murai, T., Bai, Y., Song, Z., Zhang, S., Zhang, Q., Huang, Y., Bi, X., & Jiang, J. (2025). AI-assisted multi-modal information for the screening of depression: a systematic review and meta-analysis. npj Digital Medicine. https://doi.org/10.1038/s41746-025-01933-3

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.