A BONE MECHANICAL STUDY INTEGRATED BY DEEP-LEARNING METHOD AND MECHANICAL MODELING

Bone defects are common and complex conditions in orthopedic clinics. Designing personalized bone implants with biomechanical properties that match the mechanical properties of the bone tissue in the patient's defect area holds great promise for desired bone defect reconstruction. However, the current design of personalized bone implants faces numerous challenges in the microstructural analysis of bone tissues in vivo, the characterization and modeling of heterogeneous anisotropic mechanical behavior, making it difficult to achieve mechanical property matching, which results in suboptimal bone reconstruction outcomes. To address these issues, this paper establishes an integrated approach combining data-driven and mechanical modeling for the mechanical theory, computation, and experimental methods of bone defect reconstruction, enabling accurate characterization of the mechanical properties of bone tissue under clinical conditions. Firstly, the distal femoral bone tissues of sheep are adopted as the experimental subject, a data-driven model for accurately predicting the morphological parameters of cancellous bone tissue under clinical CT imaging was proposed. A multi-neural network model combining high-resolution micro-CT and clinical CT was established. By correlating the macroscopic bone density distribution from low-resolution clinical CT and the microstructural morphological characteristics of cancellous bone from high-resolution micro-CT, a mapping relationship between bone density distribution and microstructural characteristics was established, which realized the accurate prediction of morphological parameters such as heterogeneous bone density distribution and fabric tensor of in vivo bone tissue using clinical CT. Furthermore, an anisotropic constitutive model and a Bayesian calibrated experimental method for cancellous bone based on heterogeneous bone density and fabric tensor were developed, which revealed the relationship between the mechanical behavior of cancellous bone at different locations and the growth direction of its microstructure. Combining a Bayesian-based method for identifying constitutive model parameters, the systematic errors introduced by the deviation between the principal material direction and the loading direction in cancellous bone experiments were corrected. The accuracy of the established constitutive model and parameter identification method was validated through experiments, addressing the challenges of microstructural analysis of cancellous bone under clinical medical imaging, and lays the foundation for the design of personalized bone implants.