3rd TAGRA

Accurate prediction of bone fracture probability and identification of areas susceptible to damage is an important challenge in biomechanics and medical engineering. This study aimed to develop a framework for bone fracture risk assessment, combining Finite Element Analysis (FEA) and Explainable Machine Learning (XAI). By applying loading conditions related to standing and walking postures, stress distribution, stress concentration areas, and bone mechanical behavior are investigated using the finite element method. Fracture mechanics approach is also used to simulate surface cracks with different geometries and positions to determine the stress intensity factor at the crack tip and investigate crack growth. The modeling results show that bone geometry, bone elastic modulus, load application, and crack location and size are important factors affecting the mechanical response and vulnerability of the femur. A decrease in bone elastic modulus, which can indicate a decrease in bone quality, causes a significant increase in tensile stress in some conditions. Fracture mechanics analysis also shows that the location of cracks in different bone regions has a direct impact on the stress intensity factor of the first mode and, consequently, the probability of crack propagation and failure. In the second stage, numerical outputs from the finite element analysis, including tensile and compressive stresses, displacement, stress intensity factor, crack size and location, and mechanical parameters, are used as input variables to develop machine learning models. Using explainable machine learning methods, the relative contribution of each input parameter is determined. Thus, unlike purely predictive models, the proposed framework can explain the mechanical and geometric factors affecting the model's decision. The proposed framework can be used for personalized assessment of bone fracture risk and identification of critical areas. Combining FEA with explainable machine learning enables the connection between the mechanical behavior of bone and data-driven prediction.