Artificial intelligence and wearable technology in athlete load management and injury prediction: a systematic literature review

Abstract

The emergence of sports science is marked by a change in the integration of the artificial intelligence segment to examine the high-rate burst of development of the wearable devices that is needed to decipher functional, physiological, and biomechanical areas of interest. The given review of literature performs the analysis of the status of artificial intelligence to summarize this information to facilitate proactive injury prevention and performance optimization. In this study, an attempt will be made to list the systematic, categorized approaches of existing predictive modelling solutions using AI, to trace the key methodological gaps and technical limitations to embark on the path of clinical adoption of AI-grounded predictive models. This review is based on strict adherence to the PRISMA principles and includes a thorough search of databases such as PubMed, Web of Science, and IEEE Xplore with the help of stringent requirements of including high-quality research. Deep learning (CNN, ResNet) models are superior to traditional models in the range of 87–91% and ROC-AUC up to 0.94 (analysis of 87 studies). Better predictive power was achieved with multimodal sensor configurations than with single sensor configurations. But the high data heterogeneity, class imbalance, and poor interpretability are major hurdles to real-time clinical implementation. The critical issues have been identified in this review, including inter-subject variability and class imbalance, and the need to standardize data acquisition protocols as a way of reducing the effect of feature noise on the generalizability of the model. Finally, this summary shows that longitudinal validation studies are desperately needed that do not focus on individual data sets to confirm whether AI-informed interventions are effective in high-performance sports settings.