Osteoporosis is defined as low bone density, and results in a markedly increased risk of skeletal fractures. It has been estimated that about 40% of all women above 50 years old will suffer from an osteoporotic fracture leading to hospitalization. Current osteoporosis diagnostics is largely based on statistical tools, using epidemiological parameters and bone mineral density (BMD) measured with dual energy X-ray absorptiometry (DXA). However, DXA-based BMD proved to be only a moderate predictor of bone strength. Therefore, novel methods that take into account all mechanical characteristics of the bone and their influence on bone resistance to fracture are advocated. Finite element (FE) models may improve the bone strength prediction accuracy, since they can account for the structural determinants of bone strength, and the variety of external loads acting on the bones during daily life.

Several studies have proved that FE models can perform better than BMD as a bone strength predictor. However, these FE models are built from Computed Tomography (CT) datasets, as the 3D bone geometry is required, and take several hours of work by an experienced engineer. Moreover, the radiation dose for the patient is higher for CT than for DXA scan. All these factors contributed to the low impact that FE-based methods have had on the current clinical practice so far.

This thesis work aimed at developing accurate and thoroughly validated FE models to enable a more accurate prediction of femoral strength. An accurate estimation of femoral strength could be used as one of the main determinant of a patient’s fracture risk during population screening.

In the first part of the thesis, the ex vivo mechanical tests performed on cadaver human femurs are presented. Digital image correlation (DIC), an optical method that allows for a full-field measurement of the displacements over the femur surface, was used to retrieve strains during the test. Then, a subject-specific FE modelling technique able to predict the deformation state and the overall strength of human femurs is presented. The FE models were based on clinical images from 3D CT datasets, and were validated against the measurements collected during the ex vivo mechanical tests. Both the experimental setup with DIC and the FE modelling procedure have been initially tested using composite bones (only the FE part of the composite bone study is presented in this thesis). After that, the method was extended to human cadaver bones. Once validated against experimental strain measurements, the FE modelling procedure could be used to predict bone strength.

In the last part of the thesis, the predictive ability of FE models based on the shape and BMD distribution reconstructed from a single DXA image using a statistical shape and appearance model (SSAM, developed outside this thesis) was assessed. The predictions were compared to the experimental measurements, and the obtained accuracy compared to that of CT-based FE models. The results obtained were encouraging. The CT-based FE models were able to predict the deformation state with very good accuracy when compared to thousands of full-field measurements from DIC (normalized root mean square error, NRMSE, below 11%), and, most importantly, could predict the femoral strength with an error below 2%. The performances of SSAM-based FE models were also promising, showing only a slight reduction of the performances when compared to the CT-based approach (NRMSE below 20% for the strain prediction, average strength prediction error of 12%), but with the significant advantage of the models being built from one single conventional DXA image.

In conclusion, the concept of a new, accurate and semi-automatic FE modelling procedure aimed at predicting fracture risk on individuals was developed. The performances of CT-based and SSAM-based models were thoroughly compared, and the results support the future translation of SSAM-based FE model built from a single DXA image into the clinics. The developed tool could therefore allow to include a mechanistic information into the fracture risk screening, which may ultimately lead to an increased accuracy in the identification of the subjects at risk.