Étude morphologique des structures osseuses de l’épaule par la caractérisation d’un modèle statistique

Abstract

Shoulder pathologies, like rotator cuff tears, are numerous and require precise imaging techniques in order to have personalized surgery plans. There is currently a method proposed in the imaging and orthopedics research laboratory (LIO, ÉTS) which permits the reconstruction of 3D shoulder models from bi-planar x-ray images, which reduces the radiation to which a patient is exposed through similar reconstruction approaches that use CT scans. However, using such an imaging device involves certain limitations which can explain the inaccuracy and lack of robustness of the current reconstruction method. The main objective of this master's thesis is to develop and validate statistical models that can be integrated in the 3D reconstruction method mentioned above. These statistical models will bring a priori information on the morphology and variation of the bony structures of the shoulder and will thus improve the reconstruction method. Firstly, we present a review of shoulder morphology and the reconstruction method previously proposed at the LIO, while identifying its limitations. Then, we use three databases of 3D shape models of the shoulder (asymptomatic, pathological, and hospitalized) to generate statistical models using a principal component analysis (PCA). Finally, the statistical models are characterized and validated to determine if they can be integrated in 3D reconstruction methods. We deduce that morphological variations of the statistical models correspond to those described in the literature. According to the results of the PCA, there is more variability in the hospitalized database, but we cannot determine whether this variability is due to a more severe pathology, or because the models of this database were derived from CT scans instead of xray images like the other databases. We conclude that the statistical models developed in this study can be included in a reconstruction method, excluding those of the hospitalized database. The next step will be to evaluate the method once the statistical models are integrated.

Date	15 Nov 2022
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Nicola Hagemeister (Supervisor) & Jacques A. de Guise (Co-supervisor)