Development and pre-clinical validation of a computer-assisted predictor of the risk of vertebral fracture

Pathologies such as metastasis and osteoporosis affect the mechanical properties of the vertebrae, increasing fragility fractures incidence. Patient-specific models, generated from diagnostic images, could predict fracture risk, helping clinicians in the correct therapy choice. However, validation is required to use them in clinical practice. Many proposed finite element (FE) models consider single vertebrae only, neglecting the role of intervertebral discs in load transmission. Multi-vertebrae models would include more physiological boundary conditions, but computed tomography (CT) imaging does not provide information about discs' mechanical properties. Consequently, in the first part of the thesis CT-based multi-vertebrae FE model was generated assigning to the discs a linear isotropic material with Young’s modulus values in the literature range. Boundary conditions were assigned experimentally-matched and computational displacements and strains on vertebral surface compared to experimental ones coming from Digital Image Correlation measurements. Good agreement was obtained for displacements (R2>0.9 and RMSE%<8%), but strains local distribution differed substantially. Hence, in the second part the focus was shifted on individual vertebrae. Specifically, µCT scans were acquired before and after compressive tests and used as input to Digital Volume Correlation algorithm, allowing to extract deformation field within bone structure. FE models were generated from unloaded scans and subjected to experimentally-matched boundary conditions. Predicted displacements were compared pointwise against experimental data, showing good agreement, both for healthy (R2=0.69÷0.83, RMSE%=3÷22%) and metastatic (R2=0.64÷0.93, RMSE%=5÷18%) vertebrae. Additionally, qualitative comparison between computational and experimental strains was performed, correctly identifying regions with highest strains concentration. In conclusion, when the boundary conditions are accurately modelled, subject-specific CT-based FE can predict displacements and strains with accuracies comparable to that of experimental method used for validation. But in multi-segment models, predictions are highly sensitive to biomechanical properties of intervertebral discs; thus, accurate models can be built only if disc's properties are personalised.