From Exercise to Strain: Rapid and Accurate Prediction of Femoral Neck Loading

Abstract

Femoral neck fractures pose significant morbidity and mortality risks, particularly among osteoporotic patients. This study aims to identify effective exercises for enhancing bone health and develop a neural network model to predict femoral neck strains during exercise using inertial measurement unit (IMU) data. We employed musculoskeletal modeling (MSK) and finite element (FE) analysis to assess femoral neck strains during various ballistic exercises (walking, running, countermovement jumps, squat jumps, unilateral hopping, and bilateral hopping) across three intensity levels: high, moderate, and low. Results showed that running at all intensities produced significantly higher strains compared to walking (1985 ± 802 μϵ tensile, 5053 ± 181 μϵ compressive, p < 0.001), with peak tensile strains reaching 3731 μϵ and compressive strains up to 9541 μϵ. Low-intensity unilateral hopping also yielded significantly higher strains (3003 μϵ, p < 0.001) than walking, suggesting its osteogenic potential. In contrast, squat jumps, countermovement jumps, and bilateral hopping generated lower peak strains. The neural network model demonstrated high prediction accuracy, achieving correlations up to 0.97 and root mean square errors as low as 145.20 μϵ. These findings support the use of neural networks and IMU sensors for practical, cost-effective interventions to improve bone health and reduce fracture risk.