A Computational Multiscale Framework for Bone Remodeling: Coupling Apparent Density Evolution and Microscale Shape Optimization

Abstract

Bone remodeling models are typically phenomenological or mechano-biological but often lack mechanisms to incorporate patient-specific data, limiting clinical use. We present a patient-specific multiscale framework that couples finite element (FE)-based shape optimization at the microscale with a mechano-biological model at the macroscale. The model predicts % bone mineral density (BMD) changes at the macroscale, which in turn drive microscale trabecular adaptation via % bone volume fraction (BV/TV) changes. Micro-QCT imaging data are used to train a DCGAN-based ReconGAN for virtual reconstruction of trabecular microstructures, from which FE models are generated. Apparent BMD changes predicted by the macroscale model guide the microscale shape optimization to simulate adaptation. The framework reproduces BMD losses of 9.8% (trabecular) and 4.9% (whole vertebra) over a 215-day spaceflight scenario, consistent with results from prolonged bed rest and controlled experimental datasets. In vertebral compression fracture simulations, it captures trabecular bone degeneration by reducing peak load from 3.532 to 3.280 kN and energy absorption from 0.243 to 0.218 J, and recovery restores close agreement to the original microstructure. These results demonstrate a path toward patient-specific simulation of bone remodeling and its mechanical consequences, with strong potential for treatment planning and assessment of skeletal interventions.