Create a Finite Element model of experiment involving cadaveric scapulae

Hypothesis: Nowadays, the number of people suffering from shoulder osteoarthritis increases as the population is ageing. An end-stage treatment is the total shoulder arthroplasty (TSA), but it still suffers from a high failure rate in comparison to other joints arthroplasty. To better understand the causes and the mechanisms of this high failure rate, researchers tend to build patient-specific model. To build these models, a workflow composed of different steps has to be carried out. The first step is the segmentation process, which allows to extract the geometry of the patient scapula. Different methods of segmentation are used and two of them were investigated. Indeed, it has been hypothesise that the uncertainty in the segmentation can translate into a larger one in the modelling outputs. The quantification of such errors has never been done. The goal was to estimate the errors between the two segmentation methods, which are, the "manual" and the "semi-automated" ones. Methods: The two segmentation methods are applied on one cadaveric scapula. The manual segmentation is realised thanks to thresholds values and manual adjustments, while for the semiautomated segmentation, the cortical bone is extracted by the use of thresholds values and manual adjustments, but the trabecular bone is obtained by a shrunk of 3mm of the cortical contour segmented manually. Then, each bone geometry obtained were implanted, and a FE model was build for each of them. The exactly same steps were applied to each bone geometry in the steps following the segmentation process, to influence as less as possible the error estimation. The error was estimated by comparing the modelling outputs of both models. Results: The semi-automated segmented bone geometry went through all the steps and the FE outputs were as expected. The manual segmentation suffered from invalid geometries and no proper mesh could have been generated due to the extreme thin cortical thickness in the glenoid cavity. No errors estimation was then performed. It was remarked that the difference of segmented volume between the two methods was important. Conclusion: The semi-automated segmentation process is an easy and fast method to implement. The manual segmentation is extremely time consuming and the build up of the FE model is more challenging, but more accurate. The huge different in segmented volume makes believe that the segmentation process influences the modelling outputs. The comparison of the two methods should be made on more scapulae to drawn global conclusions.