Reconstruction 3D+T des artères coronaires à partir d'une séquence rotationnelle angiographique

Abstract

Coronary heart diseases are an important cause of death in industrialized countries. To treat those pathologies, a percutaneous intervention that consists in inserting a catheter in the femoral artery is performed. The instrument is directed to the affected arteries, and coronary angiography is used to lead the surgeon in an interventional context. However, 2D angiography does not consider depth, and its use is frequent during an intervention, leading in high doses of contrast agent and an extended exposure to X-rays. To mitigate the impact of those problems, medical imaging techniques like 3D+t visualization of coronary arteries are used to assist the surgeon during the intervention. Many imaging modalities are used to acquire the sequences, but the rotational angiography is favored due to its lower contrast agent use and its ease of use in an interventional context. This imaging allows the surgeon to guide the catheter in 3D in a clear manner, and limit the use of X-rays and contrast agent by reducing the duration of the intervention. First, the current study proposes a segmentation method based on a multi scale classification approach using connected components. The algorithm extracts coronary arteries while removing the noise caused by background artifacts like the ribs and the lungs. To remove the shift caused by breathing motion, the movement is modeled using a cyclic mathematical model that synchronizes the angiographic images from the rotational sequence. 3D reconstruction is then executed by pairing bifurcations in the images to model arteries segments. The remaining pixels are paired using epipolar geometry and an energy minimization graph traversal algorithm. To achieve temporal smoothness in the reconstructed volume, an affine deformation using non- inear optimization is applied to the reference volume. The proposed method is fully automatic, and the results show that when cardiac and respiratory motion is present, the reconstruction is robust and is computable in an interventional context.

Date	28 Mar 2013
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Luc Duong (Supervisor)