Artificial Intelligence-Based Exploration of the Relationship Between Running Motion Patterns and Achilles Tendon...

Inactive adults face a measurable risk of chronic Achilles tendon injury when starting to run, yet the mechanisms remain unclear. We leverage machine learning and Shapley Additive Explanations (SHAP)-core methodologies in artificial intelligence-to identify running motion patterns that elevate Achilles tendon stress in this population. Our findings aim to inform safer running practices, ultimately helping inactive adults integrate into the running community more healthily. A total of 189 inactive adults were recruited, and their running biomechanics were comprehensively assessed. Achilles tendon stress was estimated using OpenSim musculoskeletal modeling combined with ultrasound imaging. The relationship between running biomechanics and Achilles tendon stress was examined using three machine-learning models-Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Support Vector Regression (SVR)-integrated with the SHAP framework. Modeling results demonstrated that the XGBoost model outperformed both RF and SVR in prediction accuracy. SHAP analysis revealed that Achilles tendon stress increased significantly when the ankle dorsiflexion angle was less than 10.5°, the ankle plantarflexion moment exceeded 1.5 N·m/kg, the ankle eversion moment exceeded 0.1 N·m/kg, the hip internal rotation angle exceeded 8.2°, the ankle external rotation angle was less than 25.3°, or the knee flexion angle was less than 23.3°. Based on this analysis, for inactive adults, reducing ankle plantarflexor activation, moderately increasing ankle dorsiflexion and external rotation, and optimizing proximal joint movement patterns may be crucial for decreasing Achilles tendon stress. These findings may also inform future studies on Achilles tendon injury prevention.