The role of artificial intelligence for the diagnosis and classification of ankle instability: a systematic review

Background Ankle instability, particularly chronic ankle instability (CAI), is a prevalent musculoskeletal condition associated with recurrent sprains and long-term functional limitations. Current diagnostic approaches rely heavily on clinical examination and subjective interpretation, resulting in inconsistent identification and classification of instability. Artificial intelligence (AI) has emerged as a promising approach for improving diagnostic accuracy through automated analysis of imaging, biomechanical, and clinical data. This study aimed to review the current evidence regarding the application of AI for the diagnosis and classification of ankle instability. Methods A systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and registered in PROSPERO (CRD420261463600). PubMed, Scopus, Web of Science, and Google Scholar were searched from inception to July 27, 2026. Studies investigating AI-based models for the diagnosis, classification, or subtype identification of ankle instability were included. Data regarding study characteristics, AI approaches, input modalities, and model performance were extracted. The methodological quality and risk of bias were evaluated using the Prediction Model Risk of Bias Assessment Tool for Artificial Intelligence (PROBAST + AI). Results Thirteen studies published between 2022 and 2026 were included. Reported diagnostic performance was high, with area under the receiver operating characteristic curve (AUC) values ranging from 0.892 to 0.999. Classification studies reported heterogeneous performance measures, including accuracy, AUC, and F1-score, with individual studies reporting accuracy values of up to 100%, AUC values of up to 1.00, and F1-scores of up to 0.95. However, these metrics were not directly comparable because of substantial differences in study designs, input modalities, and target conditions. According to the PROBAST + AI assessment, seven studies demonstrated low overall risk of bias, whereas five were judged as having high risk and one as unclear, with concerns primarily related to model development, validation, sample size, and generalizability. Conclusion Current evidence suggests that AI-based approaches may have substantial potential for supporting the diagnosis and classification of ankle instability, with high performance reported across imaging, biomechanical, wearable-sensor, and clinical modalities. However, the certainty of these findings is limited by methodological heterogeneity, small or selected samples, limited external validation, and potential overfitting in several studies. Therefore, the reported performance should be interpreted as preliminary rather than as evidence of established clinical effectiveness. Larger, multicenter studies with standardized reference standards, transparent reporting, external validation, and prospective clinical evaluation are needed before routine clinical implementation.