When CT images come without calibration phantom: phantom-less calibration procedure for FE-based prediction of fracture risk

Osteoporosis affects a huge number of people and its prevalence is expected to increase. The gold standard for its diagnosis, the Bone Mineral Density (BMD)-based T-score, has not proven accurate enough for its prediction: the need to find new methods able to improve fracture risk estimation is therefore urgent. In this context, aiming to improve hip fracture risk detection, Computed Tomography (CT)-based Finite Element (FE) models have been shown to predict femoral fracture risk more accurately than T-score. In the development of CT-based FE analyses, the calibration of the CT images, fundamental to extract local BMD values related to the Hounsfield Units (HU) values, is commonly based on the availability of a calibration phantom: however, it is not always possible to have phantoms available in the clinical practice. When that happens, phantomless calibration represents the only viable option. The aim of this thesis was to implement an alternative phantomless calibration of CT images to extract local BMD values from HU in absence of a calibration phantom. CT images of the proximal femurs for a cohort of 28 post-menopausal women were examined, aiming to build CT-based 3D patient-specific models for hip fracture risk estimation. Since the CT images came without the calibration phantom, a phantomless calibration procedure selected from the literature was followed to calibrate them so that local material properties could be assigned to the FE models. Peaks of air, fat and muscle tissue were extracted from histograms of the HU in a region of interest for each patient. These peaks were linearly fitted to reference BMD values of the corresponding tissues in order to extract a patient-specific calibration of the images. Thus, HU-BMD calibration functions could be identified; subsequently, these calibration functions were employed to assign material properties to the FE models. Boundary conditions reproducing sideways fall conditions were eventually applied and static simulations performed. Tensile and compressive principal strains were extracted for the models and a Risk Factor (RF) calculated for each mesh element as the ratio between principal strains and corresponding thresholds. Furthermore, a Risk Factor Index (RFI), the highest superficial RF value, and the Femoral Strength (FS), the load at which fracture was estimated to occur, were extracted for each patient. The obtained outcomes were compared with those obtained from analogous models where the equivalent local densities were obtained with a literature-based non-patient-specific calibration. The corresponding element- specific principal strains and RF values were compared and relative errors computed. As far as principal strains and RF are concerned, the mean relative error considering the patient-specific average errors values were between 25 and 26%, while the maximum value among the average ones for each patient were between 31% and 34%. The fracture risk indicators (RFI and FS) turned out to be significantly correlated (p<0.05), but a greater number of patients resulted to be at high fracture risk according to the phantomless and patient-specifically calibrated models. Unfortunately, the lack of follow-up information did not allow the validation of the obtained results, but in the future further studies will allow the evaluation of the power of the proposed methodology.