Abstract
with machine learning (ML) to forecast the fracture behavior of cortical bone while maintaining microstructural
fidelity. It has created a parametric dataset of about 450 three-dimensional XFEM models of single-edge notched bend
(SENB) specimens, including important µ-structure features (e.g., osteon orientation, cement line properties, interfacial
connectivity, etc.). Based on these simulations, 47 quantitative descriptors were obtained, and these were used to model
supervised ML models, namely, Random Forest and Artificial Neural Networks, to estimate fracture load. The Random
Forest model demonstrated exceptional predictive performance (R2 = 0.952, MAE = 6.1 N, MAPE = 1.9%, Pearson
r = 0.980), significantly outperforming the ANN model (R2 = 0.831, MAE = 11.8 N, MAPE = 3.7%), The Random
Forest model demonstrated strong predictive performance (R2 = 0.95, MAE = 6.1 N, MAPE = 1.9%), while reducing
computational time by nearly 300-fold and memory requirements by over 20-fold compared to full XFEM analyses. For
fracture resistance, feature importance analysis indicated the most salient features were osteon orientation, cement line
strength, and pore topology. The methodology was also robust, as further led to by sensitivity analyses and uncertainty
quantification. The hybrid method provides microstructure-based predictions of fracture that are very automated and
precise and result in a significant decrease in computational cost, hence allowing a scalable route to clinical translation
to characterize bone integrity
fidelity. It has created a parametric dataset of about 450 three-dimensional XFEM models of single-edge notched bend
(SENB) specimens, including important µ-structure features (e.g., osteon orientation, cement line properties, interfacial
connectivity, etc.). Based on these simulations, 47 quantitative descriptors were obtained, and these were used to model
supervised ML models, namely, Random Forest and Artificial Neural Networks, to estimate fracture load. The Random
Forest model demonstrated exceptional predictive performance (R2 = 0.952, MAE = 6.1 N, MAPE = 1.9%, Pearson
r = 0.980), significantly outperforming the ANN model (R2 = 0.831, MAE = 11.8 N, MAPE = 3.7%), The Random
Forest model demonstrated strong predictive performance (R2 = 0.95, MAE = 6.1 N, MAPE = 1.9%), while reducing
computational time by nearly 300-fold and memory requirements by over 20-fold compared to full XFEM analyses. For
fracture resistance, feature importance analysis indicated the most salient features were osteon orientation, cement line
strength, and pore topology. The methodology was also robust, as further led to by sensitivity analyses and uncertainty
quantification. The hybrid method provides microstructure-based predictions of fracture that are very automated and
precise and result in a significant decrease in computational cost, hence allowing a scalable route to clinical translation
to characterize bone integrity
Keywords
Biomedica
Computational Efficiency
Cortical bone
Extended Finite Element Method
Fracture Load prediction
machine learning