DXA-derived three-dimensional finite element models of the femur: validation against CT-based models

Three-dimensional finite element (FE) models derived from Computed Tomography (CT) images predict hip fractures better than areal bone mineral density measurements from Dual-energy X-ray Absorptiometry (DXA). Yet, these results have not justified the adoption of CT in clinical practice, and only 2D DXA images are clinically available. Statistical shape and appearance models can be used to reconstruct three-dimensional FE models from 2D DXA images. While ex vivo validations have been performed on 3D reconstructed DXA-based FE models, it is not clear how well 3D reconstructed DXA-based FE models can predict fractures compared to CT-based models. The aim of this study was thus to evaluate the ability of one such methodology, namely DXA2FEM, to predict fractures in a clinical cohort of pair-matched fractured and control subjects, for whom both DXA and CT images were available. 3D FE models of the femur were built from both DXA and CT, and FE simulations were run reproducing a sideways fall in 28 different femoral configurations. An absolute risk of fracture (ARF0) was then computed based on the FE-predicted femoral strength values. DXA- and CT-derived models were compared with respect to geometry, density distribution, and FE-predicted proximal femoral strength. DXA-derived 3D FE models had an average point-to-surface distance of -2 mm from CT-based models, whereas the Young's moduli were 29% higher. ARF0 by CT reported statistically significantly better diagnostic accuracy (0.83, 95% CI 0.75 to 0.91) than standard hip DXA (0.69, 95% CI 0.6 to 0.8) or FRAX (0.69, 95% CI 0.57 to 0.81). The diagnostic accuracy of ARF0 by DXA was between ARF0 by CT and standard hip DXA/FRAX (0.74, 95% CI 0.62 to 0.86), albeit neither difference was statistically significant in the analysed cohort.