Research progress in the clinical application of AI in strabismus

PURPOSE: Strabismus diagnosis and management remain subjective due to variable measurement techniques and lack of standardized surgical criteria. Artificial intelligence (AI) has shown potential in ophthalmology, but its clinical translation in strabismus is underexplored. This review systematically summarizes current AI applications and challenges in strabismus diagnosis, surgical planning, and postoperative training. METHODS: We searched PubMed, Web of Science, and CNKI for studies up to 2025 using keywords "artificial intelligence," "machine learning," "deep learning," "strabismus," and "eye tracking." Studies on AI-based detection, angle measurement, surgical dosage prediction, and outcome evaluation were included. RESULTS: Fifty-two studies were analyzed. AI applications focus on three areas: (1) Deep learning on corneal reflection and facial photographs achieved diagnostic sensitivity of 94-99.1% and specificity up to 99.3%, outperforming clinicians in some datasets; (2) Eye-tracking technology showed ~90% correlation with prism cover test and higher repeatability (ICC 0.99 vs. 0.91, p<0.02) in children; (3) Surgical planning models (support vector regression, decision trees) predicted muscle recession/resection with mean errors of 0.5-0.8 mm. However, most models lack multi-center prospective validation. CONCLUSION: Despite promising accuracy, clinical adoption faces challenges: single-center data bias, poor interpretability, infrastructure gaps in remote areas, and no regulatory approvals. Future work requires multi-center trials, explainable AI, and user-friendly deployment to ensure equitable global use.