BoneNet: A Framework for Rapid and Accurate Prediction of Femoral Neck strains from Exercise Data

Femoral neck fractures pose significant risks, particularly among osteoporotic patients. This study identifies effective exercises for bone health and introduces BoneNet, a neural network-based framework for predicting femoral neck strains using inertial measurement unit data. Using musculoskeletal modeling and finite element analysis, we evaluated femoral neck strains during various dynamic exercises, including walking, running, countermovement jumps, squat jumps, unilateral hopping, and bilateral hopping, across three intensities: high, moderate, and low. Running at all intensities and low-intensity unilateral hopping generated significantly (p < 0.001) higher strains than walking (up to 47% and 35%, respectively, for tensile and compressive strains), suggesting osteogenic potential. Other exercises produced lower peak strains than walking. BoneNet achieved high prediction accuracy, with correlations up to 0.97 and root mean square errors as low as 145.20 \u00b5\u03b5. These findings support neural networks and IMU sensors as practical, cost-effective tools to enhance bone health and reduce fracture risk.