Comparing the predictions of CT-based subject-specific finite element models of human metastatic vertebrae with digital volume correlation measurements

Several conditions can increase the incidence of vertebral fragility fractures, including metastatic bone disease. Computational tools could help clinicians estimate the risk of vertebral fracture in these patients; however, comparison with in vitro data is mandatory before using them in clinical practice. Nine spine segments were tested under compression and imaged with micro-computed tomography (µCT). The displacement field was calculated for each vertebra using a global digital volume correlation (DVC) approach. Subject-specific homogenised finite element models of each vertebra were built from µCT images, applying experimentally matched boundary conditions at the endplates. Numerical and experimental displacements, reaction forces, and locations showing higher strain concentrations were eventually compared. Additionally, given that µCT cannot be performed in clinical settings, the outcomes of a µCT-based model were also compared to those of a model built from clinical CT scans of the same specimen. Good agreement between DVC and µCT-based FE displacements was found, both for healthy (<jats:italic>R</jats:italic> <jats:sup>2</jats:sup> = 0.69 ÷ 0.83, RMSE = 3 ÷ 22%, max error &lt; 45 μm) and metastatic (<jats:italic>R</jats:italic> <jats:sup>2</jats:sup> = 0.64 ÷ 0.93, RMSE = 5 ÷ 18%, max error &lt; 54 μm) vertebrae. Strong correlations were found between µCT-based and clinical CT-based FE model outcomes (<jats:italic>R</jats:italic> <jats:sup>2</jats:sup> = 0.99, RMSE &lt; 1.3%, max difference = 6 μm). Furthermore, the models qualitatively identified the most deformed regions identified with the experiments. In conclusion, the combination of experimental full-field technique and in-silico modelling enabled the development of a promising pipeline to validate bone strength predictors in the elastic range. Further improvements are needed to analyse vertebral post-yield behaviour better.