Failure of total hip replacement (THR) can be a highly variable phenomenon. Many factors of a different nature may affect the outcome of this operation, e.g. loads occuring in the joint, patient geometry, implant design, genetic factors, etc. Building a predictive model for failure of a THR is a technically challenging task requiring the incorporation of many uncertainties and modelling failure long-term processes.

In this work a stochastic numerical framework was developed to predict aseptic loosening of the femoral component, which remains one of the main reasons for the failure of THR. To enable the development of the stochastic framework two technologies were created. First, a fully automated method to reconstruct proximal femur geometry from planar X-rays, based on contour extraction of the femur and warping a generic surface model, was developed. It was capable of reconstructing proximal femur geometries with an average surface distance error not exceeding 3.8 mm, which stands in line with other existing methods to reconstruct femur from medical images. Additionally, an FE analysis of strain dependency on the geometric error was performed. It was shown that the error level of the reconstructed geometries does not dramatically change the strains. Secondly, a fully automated technique to mesh a surgical plan, i.e. an implant placed in the patient femur, was developed. This technique was based on morphing a pre-generated generic mesh employing parameterization of the meshed surfaces. An FE solver was employed to warp the internal structure of the generic mesh. The method produced good quality meshes in a rapid and fully automated manner. These techniques enabled the creation of the stochastic framework for modelling failure of total hip replacement. An FE-based approach for modelling aseptic loosening due to the damage accumulation scenario was employed as a basis for the stochastic framework. A Response Surface Methodology was utilized as an approximation to perform large trial Monte Carlo simulations. This surface was generated based on predicted values for a deterministic training set. Implant migrations was used as an indicator of revision risk.

Two potential applications were chosen to demonstrate the capability of the framework. These applications used patient morphology, implant design and surgical position as explanatory variables. First, a patient-specific study was performed. Three implants were placed into a publicly available CAD model (reference positions), then 6 deviations (rotations) were applied. This deterministic training set of aseptic loosening simulations allowed construction of response surfaces, which were father used to generate the stochastic set. Secondly, a population-based study was undertaken. For this study, the training set was extended to another 2 patient morphologies with similar implant positions. The difference between the two studies was that different random variables were chosen and, thus, allowed the answering of different research questions.

Analysis of the predicted values indicated that, the varus/valgus angle of the implant was the most strongly correlated factor with implant migration, hence revision risk. Reference positions generated according to templating procedure did not necessarily produce the best result. However, using a training set of templated position plus deviations, it was also impossible to discriminate the implant performance with statistical significance. The stochastic approach was necessary to enable the statistical discrimination of different factors. Furthermore, an implant performance analysis carried out on a single femur would give a false prediction compared to a sample of femora from a population. Finally, based on a single femur predicted rankings for implant performance were different from the rankings received using population-based scenario. Thus, a stochastic approach to investigation is very important and strongly recommended for the failure modelling of joint replacements.

This framework is a powerful tool for hypothesis testing in the mechanics ofjoint replacements and has a large range of potential applications, e.g. patient-specific pre-operative planning and population-based device testing.