Application of Artificial Intelligence for Predicting Sports Injuries and Customizing Personalized Prevention Strategies: A Scoping Review

Résumé fourni par la source

Background: Sports injuries impose a substantial burden on athletes. Machine learning (ML) and deep learning (DL) methods, collectively referred to as artificial intelligence (AI), are increasingly applied to develop predictive models and targeted prevention strategies. Objective: This scoping review aimed to map contemporary trends in AI applications for sports injury prediction and personalised prevention strategies, critically appraising the existing methodological approaches and identifying future research directions. Methods: Following PRISMA-ScR guidelines, we systematically searched five electronic databases, i.e., PubMed, Web of Science, Institute of Electrical and Electronics Engineers Xplore, Scopus, and Google Scholar, for peer-reviewed studies published up to February 2026 that applied AI methods for injury prediction and/or prevention in athletic populations. Results: Thirty-nine studies were included. Tree-based ML algorithms were the most common (59% of studies) methods used, with reported area under the curve values ranging from 0.82 to 0.95. DL was used in 18% of studies, with one hybrid model reporting 92% accuracy. Integrating multi-modal data was associated with improved model performance in 37% of studies. Among included studies, AI-informed prevention strategies were associated with injury reductions ranging from 23% to 42%, derived from synthesis-level and single-centre intervention evidence, respectively. The key challenges identified were heterogeneous injury definitions, small sample sizes, and data privacy concerns. Conclusions: AI models can inform personalised injury prevention, but their clinical use is limited by methodological issues. Key limitations include heterogeneous injury definitions, small sample sizes, and a lack of external validation. Standardised protocols are needed to improve the reliability and application of these models in practice.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.