Answers are grounded in this paper and cite their sources.
Autism spectrum disorder (ASD) is a complex neurodevelopmental condition. Electroencephalography (EEG) combined with deep learning (DL) has emerged as a promising noninvasive approach for ASD detection. To systematically review the performance, methodologies, and limitations of DL models developed for ASD diagnosis using EEG data. A systematic search of PubMed, Scopus, and Web of Science was conducted through June 27, 2024, following PRISMA guidelines (PROSPERO: CRD42024564955). Eligible studies evaluated deep learning models for EEG-based classification of individuals identified as having ASD compared with non-ASD controls. A meta-analysis employing a random-effects model was conducted to estimate the pooled sensitivity and specificity. A sensitivity analysis and a meta-regression were also performed. Quality assessment was performed using QUADAS-2. Thirty-four studies published between 2018 and 2024 met inclusion criteria for the systematic review. The meta-analysis included 15 studies. The pooled sensitivity was 0.94 (95% CI: 0.92–0.96), and the pooled specificity was 0.87 (95% CI: 0.83–0.90). The subgroup analysis demonstrated that studies employing task-based EEG attained a pooled sensitivity of 0.95 (95% CI: 0.92–0.97) and a specificity of 0.94 (95% CI: 0.92–0.96). Studies utilizing resting-state EEG achieved a pooled sensitivity of 0.95 (95% CI: 0.92–0.97) and a specificity of 0.94 (95% CI: 0.92–0.96). EEG-based DL models demonstrated high reported diagnostic performance. However, methodological heterogeneity, small sample sizes, limited external validation, and potential overfitting preclude conclusions regarding their clinical readiness or superiority over established diagnostic tools. Larger prospective multicenter studies are required before clinical implementation.
No takes yet. Share an insight, caveat, or question.
Yousefi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: