Prediction and prevention of overuse injuries in long-distance runners using recurrent neural networks and wearable sensor data

A deep learning-based engineering framework is proposed for the early prediction and prevention of overuse injuries in long-distance runners, leveraging synchronized multi-modal wearable sensor data and advanced recurrent neural network architectures. Additionally, the system integrates advanced optical technologies, including motion capture systems, near-infrared spectroscopy (NIRS), and LiDAR, to provide richer spatial and physiological information. The system utilizes an inertial measurement unit, force sensors, and surface electromyography technology to collect detailed biomechanical signals. Optical data extends the dimensionality of biomechanical monitoring, capturing high-resolution motion trajectories, muscle oxygenation, and environmental features. Key technical advancements include: a bidirectional long short-term memory network constructed via dynamic sliding window optimization and hierarchical attention mechanisms; the ability to extract instantaneous and long-term risk patterns from continuous running data. Clinical validation through a 16-week prospective cohort study of competitive athletes confirmed the model's ability to accurately determine injury event timing. Model interpretability based on hierarchical correlation propagation enables detailed attribution of biomechanical and physiological risk factors, supporting personalized interventions. Results demonstrate that combining wearable and optical data improves early detection of overuse injury risk while maintaining robustness against data variability and sensor failures.