Reconstruction 3D du genou ostéophytique à partir de radiographies bi-planes

Abstract

Stereo radiography is used to acquire two calibrated x-ray images of the patient in order to use them to perform 3D model reconstruction of bone. The 3D models obtained can be used in clinical settings for diagnostic purposes, surgical planning or for post-surgical follow-up. The problem of 3D reconstruction from two radiographs is formulated as a 2D/3D registration problem where the objective is to perform bone structure generic model deformation in accordance with the radiographs. 2D/3D registration-based 3D reconstruction presents several challenges in the case of osteophytic knees which are: The difficulty to precisely identify bone growths (osteophytes) on x-rays images, the limited amount of identifiable relevant information’s in only two radiographs and the superimposition of the knee bone structures on radiographs. The main objective of this thesis is to develop a 2D/3D registration method for 3D reconstruction of the osteophytic knee. We developed a pseudo-volume model of the knee which includes a combined statistical pose and shape of the femur, tibia and patella, coupled with the density information of each bone structure. This model, which allows for a compact representation of all the knee bone structures, is used as a generic model within the framework of the 2D/3D registration. We also proposed a hybrid optimization method that combines the genetic algorithm with a regression based local optimization method to perform iterative optimization within the framework of 2D/3D registration. As a result, the proposed method which is based on an iconic similarity measurement allows for the simultaneous 3D reconstruction of the femur, tibia and patella from only two radiographs. The database used consists of 83 cases for modeling and 10 cases for tests. Test x-rays are simulated images from CT scan volumes. By comparing the reconstructed 3D models of each bone structure with the equivalent reference model, we obtained an RMS (Root Mean Square) error of approximately 2.73 ± 0.54 mm for the distal femur, 2.59 ± 0.57 mm for the proximal tibia and 3.60 ± 1.44 mm for the patella. These results are better than those obtained for the 3D reconstruction of the same subjects’ model by considering each bone structure one at time and following a 2D/3D registration approach based on geometric similarity. As a result, this work opens perspectives for the application of 3D reconstruction techniques for the design of personalized surgical instrument for patients with osteoarthritis.

Date	17 Aug 2021
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Carlos Vázquez (Supervisor) & Jacques A. de Guise (Co-supervisor)