Déformation 3D automatique du fémur à partir d’un nombre arbitraire de radiographies EOS calibrées

Abstract

3D reconstruction of the femur is useful for orthopedic surgeons in order to plan their surgery as well as the type of implants to be used. In partnership with the company EOS Imaging, we will use their calibrated biplane X-ray images in order to propose a precise 3D reconstruction method of the femur allowing for an arbitrary number of calibrated EOS images to be used. This research project’s objective is to replace the semi-automatic 3D reconstruction method currently used in EOS Imaging’s commercial platform, SterEOS. Our review of the literature has shown that the 3D reconstruction algorithms currently used in bone reconstruction from x-ray images are mainly based on non-rigid 3D/2D registration algorithms. These methods, although robust and accurate, have several limitations and are complex to use. It is in this perspective that we propose a 3D deformation method using Pixel2Mesh++, a deep learning algorithm performing a dense deformation on a generic model. Our experiments focus on the use of a pair of DRR where the femur is seen from the face and the side view as well as on the impact of using additional radiographs in the reconstruction process and show that our method makes it possible to obtain a very precise 3D reconstruction, even when a single acquisition is used.

Date	17 Dec 2023
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Carlos Vázquez (Supervisor) & Jacques A. de Guise (Co-supervisor)