CT-based Finite Element Modeling of the human femur: comparison with in-vitro strain measurements under stance loading conditions

Proximal femoral fractures are among the most common osteoporosis-related injuries and are projected to affect roughly 21 million individuals worldwide by 2025. These fractures occur predominantly as a consequence of falls, but may also arise as spontaneous failures under paraphysiological loading, particularly in older adults with osteoporosis, who exhibit reduced bone mineral density (BMD). Although BMD is clinically employed to estimate fracture risk, this latter ultimately depends on bone strength—the capacity of bone to withstand physiological, paraphysiological, and pathological loads—which is not fully captured by BMD alone. Computed Tomography (CT)–based finite element (FE) modeling could represent a valid alternative to estimate in vivo the subject-specific femur strength. Yet, such models need to be first validated against experimental in vitro data. The aim of this thesis was thus to construct CT-based FE models of seven human cadaveric femurs, reproducing the in-vitro experimental tests previously conducted on the same specimens to compare the models’ predictions with the experimental outcomes. In the experimental set up all femurs were instrumented with up to 16 triaxial strain gauges and subjected to six loading configurations that span the entire cone of the hip joint resultant force while standing, including the single leg stance configuration and the most relevant loading configuration for spontaneous fractures (8° in the frontal plane). Corresponding boundary conditions were applied to the FE models built starting from the specimens CT images and simulated using Ansys Mechanical APDL. Principal strains were extracted at nodes in the same location as the strain gauges and averaged over a sphere of 3mm to ensure continuity of the strain and avoid local effects. The principal strains values predicted by the models could then be compared with the experimental measurements obtained for each strain gauge on each femur and for all loading configurations. Results across all femurs and all six loading configurations show higher tensile strains (ε₁) on the lateral aspect of the proximal femur, with peak values ranging from 749 µε to 1704 µε, and higher compressive strains (ε₃, in absolute values) on the medial aspect, with peak values ranging from −830 µε to −1675 µε, consistent with experimental observations. Nonetheless, the overall average percentage error is higher than expected, ranging between 32–285% for ε₁ and 18–250% for ε₃, with extreme outliers in cases where experimental strain values were very small. Conversely, the median error provides a more reliable measure of accuracy: it remains below 40% in well-predicted regions, ranges from 20% to 142% for ε₁ (with a few femur-specific and configuration-specific outliers), and from 15% to 127% for ε₃, with only one significant outlier affecting both ε₁ and ε₃ in a single femur and under one specific loading configuration. In conclusion, the model discreetly reproduces the experimental strain distribution, showing the expected pattern of higher tensile strains and higher compressive strains on the lateral and medial aspect of the femur, respectively. The major discrepancies emerge under specific loading conditions and in correspondence with very low experimental strain values. Predictive accuracy in future developments might be enhanced by refining the segmentation process and implementing a more robust algorithm for computationally locating the strain gauges, thus ensuring a closer match with their experimental positions.