The use of machine learning in predicting anterior cruciate ligament injury: a systematic review and meta-analysis

Machine learning (ML) models are used to analyze the relationship between risk factors and anterior cruciate ligament (ACL) injury outcomes. This review assessed the practicality of ACL injury prediction models, analyzed the existing issues, and provided valuable information for future research.

A comprehensive search was conducted across PubMed, Medline, Embase, Scopus, Web of Science, and Cochrane databases for relevant studies until 30th April 2024.

The search yielded 633 studies, of which eight articles (32 predictive models) were included in the final evaluation. Based on the results, the meta-analysis quantification yielded an overall area under the curve (AUC) of 0.79 (95 % CI: 0.75-0.82), a combined sensitivity of 0.57 (95 % CI: 0.45-0.68), and a combined specificity of 0.87 (95 % CI: 0.78-0.92). The included models comprised 10 ensemble algorithms and 22 non-ensemble algorithms. Ensemble methods demonstrated higher specificity (0.96 vs. 0.79) and AUC (0.79 vs. 0.68), whereas non-ensemble models showed better sensitivity (0.65 vs. 0.40).

ML models were effective in correctly identifying non-injury cases, however their ability to detect actual injury occurrences required significant improvement. Algorithm selection significantly influenced performance trade-offs: ensemble models favored specificity, whereas non-ensemble models provided superior sensitivity. These findings may guide algorithm selection to improve the accuracy and efficiency of injury prediction tools in sports medicine. Despite the challenges posed by the diversity of injury mechanisms, the study emphasizes the importance of high-quality biomechanical data, prospective study designs, and standardized methodologies in enhancing model reliability and clinical applicability, providing a basis for optimizing ACL injury prediction models.