Defended by Stéphanie UK on 29-01-2021

Thesis director: Christine CHAPPARD

Abstract

Currently, medical imaging modalities involving X-rays play an essential role in the diagnostic thanks to the good visualization of bone structure. The fluoroscopy is used in the operating room to assist and guide the surgical gesture, now available with planar sensors on mobile systems which can adapt to restricted areas of operating rooms.

These systems can generate two-dimensional (2D) images such as radiographs or three-dimensional (3D) images such as scanner based on the principle of tomography. The generation of these images in 2D/3D dimensions is possible thanks to the mobile C-arm or O-arm already used for surgery. However, these devices have one major disadvantage: the amount of radiation dose they create. When exposed to high radiation doses, they become dangerous threat to both patients and operators.

In this thesis, for we will try to respond to two problematics: For which tomographic configurations, for a strategy dose reduction for applications in orthopaedical surgery? what reconstruction methods are best suited to the use of morphometric criteria? How to improve the diagnosis of osteo-articular pathologies based on the analysis of 3D scanner images?

I will address these objectives, on the one hand, in the field of orthopaedical surgery. To respond to the first two objectives, our approach has been experimental. The knee specimen was firstly imaged by a prototype C-arm coupling a cone-beam X-ray source to a plane sensor. This prototype was developed by our partners in the FUI-3D4Carm project with the aim of being integrated on a C-arm to get an intraoperative 3D image.
We tested several dose reduction scenarios with 2 types of reconstruction algorithms, 1) classical Felkamp (FDK) and 2) Iterative (SART-TV). We tested the reduction of the intensity (mAs), the number of projections and the angular range on the quality of the images. Various evaluation approaches were performed by expert qualitative assessment, digital image quality indicators based on grey levels, structure and finally segmentation methods.

On the other hand, in the context of rheumatology, the objective is to quantify the variations in joint width as a biomarker using 3D images for the diagnosis of osteoarthritis in order to answer our third objective. We studied variations in joint spacing at the initial time and at 36 months in a longitudinal study of patients suffering from osteoarthritis. To carry out this quantification, a hybrid segmentation method was developed. It is based on two segmentation methods: a classification segmentation to select the bone part, followed by an active contour method, called “Snake”, to extract the joint space.

We have been able to demonstrate which criteria for which images are potentially usable for orthopaedic surgery: i) an intensity greater than 5mAs, ii) a minimum threshold angular range of 170°, iii) a minimum number of projections of 200.
Then we were also able to demonstrate that the choice of the reconstruction algorithm depends on its use. Indeed, the FDK algorithm gives a better visualization of the bone structure and observers tend to favor it. Whereas images reconstructed with the SART-TV algorithm offer a better similarity on geometric criteria and allow a more stable segmentation.

Finally, we were able to obtain preliminary results on the quantification of joint spacing using a semi-automated segmentation method on patients suffering from osteoarthritis. We have identified good segmentation results in 60% of the cases, which suggests that this method is potentially interesting but still needs to be improved.

Visit theses.fr website