ObjectiveTo evaluate the diagnostic performance of artificial intelligence (AI) in strabismus detection through a systematic review and meta-analysis. MethodsThe PubMed, Embase, Cochrane Library, Web of Science, CBM, WanFang Data, and CNKI databases were systematically searched to identify clinical studies on AI-assisted strabismus diagnosis. Study quality was assessed using the QUADAS-AI tool. Diagnostic performance metrics were pooled using a random-effects model. Meta-regression and subgroup analyses were performed to explore sources of heterogeneity. ResultsNine studies involving 19872 participants were included. Meta-analysis revealed a pooled sensitivity of 0.94 (95%CI 0.88 to 0.97) and a pooled specificity of 0.95 (95%CI 0.90 to 0.98) in internal validation test sets. For independent external validation cohorts, the pooled sensitivity reached 0.97 (95%CI 0.93 to 1.00) and specificity was 0.99 (95%CI 0.98 to 1.00). Four studies (44%) had a high risk of bias. Meta-regression and subgroup analyses indicated that the performance of AI in diagnosing strabismus was associated with sample size, validation method, use of transfer learning, and feature extraction type. ConclusionAI demonstrates excellent sensitivity and specificity for strabismus screening and early diagnosis, showing potential as a clinical decision support tool. Future studies require standardized methodologies and rigorous external validation to ensure clinical applicability.