Traditional methods for sports injury prevention and rehabilitation rely predominantly on subjective clinician-guided assessments and standardized intervention protocols. These approaches often result in limited accuracy, delayed responsiveness, and insufficient personalization. Recent advances in artificial intelligence (AI), wearable sensor technologies, and multimodal analytics provide novel opportunities for objective, real-time, and personalized injury management strategies. Despite these advances, there remains a critical need for systematic synthesis and evaluation of integrated multimodal approaches. This systematic review critically evaluates contemporary developments in multimodal AI technologies applied specifically to sports injury prediction and rehabilitation. We systematically describe the biomechanical and physiological foundations of common acute and chronic sports injuries and present them within an integrated, five-stage injury recovery pipeline. Our analysis emphasizes AI methods including sensor fusion frameworks, time-series classification algorithms, and predictive analytics that enhance early injury detection, accurate risk modelling, and timely interventions. For the rehabilitation phase, we critically assess AI-supported motion quality assessment methods, adaptive feedback mechanisms, and individualized recovery protocols facilitated by wearable and vision-based technologies. Furthermore, we evaluate the real-world deployment and athlete-specific modelling strategies of AI systems, addressing challenges of environmental robustness, computational efficiency, and personalized adaptation. Multimodal AI technologies offer substantial potential for revolutionizing sports injury prediction and rehabilitation by enabling highly individualized, data-driven, and context-aware solutions. Nevertheless, significant challenges persist in the areas of model generalization, interpretability, privacy concerns, and clinical validation. Promising future research directions include the advancement of explainable AI frameworks, digital twin technologies, and multi-agent modelling approaches, aimed at overcoming these limitations and advancing personalized, intelligent sports medicine.
Wang et al. (Thu,) studied this question.