Towards the complete automatization of BBCT-hip in silico methodology for femur fracture risk prediction.

Femoral fractures are common in the elderly, primarily due to age-related osteoporosis and the increased risk of falls. Recently, in silico methodologies have gained significant attention for predicting fracture risk, particularly in the context of osteoporosis management. To be effective in clinical settings, they require automatization, which reduces analysis time, enhances accessibility through hospital system integration, and standardizes results for consistency regardless of operator experience. In this context, this study presents a practical application of a novel in-silico methodology, the Bologna Biomechanical Computed Tomography at the hip (BBCT-hip), which integrates patient-specific Finite Element (FE) simulations and a stochastic mathematical model to predict the femoral absolute risk of fractures at time 0 (ARF0). The aim of this work was specifically to focus on the automatization of two central steps of the BBCT-hip process: the segmentation procedure, which derives the femur geometry from CT scan, and the material mapping phase, which is based on the CT Hounsfield Unit. Therefore, a fully automated segmentation method for the femur has been validated. In particular, the segmentations and the FE results coming from such method have been compared to those coming from semiautomated and manual segmentation methodologies. The first comparison was made between the proximal femurs geometries by calculating the Dice Coefficient (DSC), Jaccard Index (JAC), Hausdorff Distance (HD) and Average Hausdorff Distance (AHD). Subsequently, BBCT-hip simulations were conducted to compare the segmentation methodologies in terms of resulting failure loads, maximum strains, and ARF0. As for material mapping, the outcomes of an in-house automatic script developed in the Medical Technology Laboratory were compared to those from the commercial software Bonemat (V3.2, Istituto Ortopedico Rizzoli, Bologna, Italy). Moreover, a second sample was examined to study the segmentation variability. Metrics were computed within manual and semi-automatic segmentation and compared with those evaluated between automatic and the other segmentation procedures. Results from geometric comparison revealed that the automated segmentation differs more from manual (DSC = 0.96±0.01, JAC = 0.93±0.02, HD = 4.11±1.37 mm, AHD = 0.69±0.14 mm) than semiautomated segmentation (DSC = 0.98±0.01, JAC = 0.97±0.01, HD = 3.65±1.55 mm, AHD = 0.22 ± 0.07 mm). These discrepancies were also observed in the outcomes of the BBCT-hip simulations, highlighting differences between automatic and manual segmentation (R² > 0.58, RMSE < 35.15%), as well as between automatic and semi-automatic (R² > 0.55, RMSE < 47.05%). Regarding the comparison of material assignment methodologies, no significant differences were found (R² > 0.97, RMSE < 7.01%), making the in-house script preferable to the commercial software, due to its higher level of automatization. Lastly, the variability analysis revealed statistically significant differences between the metrics computed within manual and semi-automatic segmentations and those calculated between automatic and the other segmentation methods (p < 0.05). In conclusion, automating tools and processes in a clinical setting can reduce manual variability, enhance decision support, and improve time efficiency, ultimately enabling clinicians to deliver high-quality care that benefits patients.