Three-dimensional reconstruction of coronary arteries from angiographic sequences for interventional assistance.

摘要

The main objective of this work is to implement novel techniques necessary to extract and use the information contained in single plane angiographic sequences in order to develop a potential tool to assist the cardiologist concurrently during the procedure. Two specific hypotheses will need to be verified to justify our main research objectives. The first hypothesis states that the integration of a geometrical curvature constraint improves the precision of the point correspondence and improves reconstruction accuracy whereas the second hypothesis states that it is possible to recuperate the 3D geometry across a single plain angiographic sequence, for clinical assistance, by using an a priori 3D model of the structure of interest.;Recalling that the cardiologist usually makes use of a single plane angiographic view by looking at the X-ray fluoroscopy monitor during interventions, it would be advantageous to reconstruct the arteries in 3D using one view. Thus, we suppose that an a priori 3D model of the arteries is obtained and available at a first time instant. The model can be obtained from a biplane reconstruction before the intervention, or from preoperative data, such as an MRI or MDCT dataset, which are the registered thereafter in the coordinate frame of the fluoroscope during a procedure. The monoplane equations developed optimize for the unknown 3D displacements in subsequent time instants of the angiographic dataset. A clinical experiment was performed using arrhythmia data. The aim is to estimate the depth of the tip electrode of the ablation catheter in 20 distinct datasets. For a posterior/anterior view, the 3D mean and maximum depth errors were 2.68mm and 7.05mm, whereas for a left-right anterior oblique setup the 3D mean and maximal errors were 2.34mm and 5.84mm respectively.;In conclusion, our research work helped verify the two hypotheses outlined above. In order to develop a potential tool to assist the cardiologist during angioplasty procedures primarily (i.e. catheter ablation procedures may be targeted as well), we developed solutions that consider and extract the spatio-temporal information from X-ray images acquired by the fluoroscope. We implemented a two-click process that extracts the coronary artery centerline in the diastolic image and that is tracked temporally using all images representing an entire cardiac cycle. The 3D reconstruction obtained using a biplane setup was optimized by considering a novel geometric curvature constraint when choosing point correspondences. Single plane reconstruction is made possible if an a priori 3D model of the targeted structure is available at a 1 st time instant in the angiographic images. Subsequent estimates of the 3D structure are obtained by considering the spatial coordinates and displacements of the arteries as well. Future work should focus on including non-rigid constraint equations to consider the inherent movement of the heart. Other geometrical constraints such as length preservation terms and 3D perpendicularity constraints between normals and tangents could be exploited to improve results using a single view. (Abstract shortened by UMI.).;We implemented a novel automatic 2D segmentation algorithm to enhance healthy or diseased (stenosis) coronary arteries in a first image representing the diastolic cardiac phase. A 4-step filter, that included a homomorphic, anisotropic, shock filter and morphological component systematically suppressed the background and enhanced only the arteries. Then, we provide the cardiologist with the ability to target a specific coronary artery and investigate its motion by extracting its 2D centerline and automatically temporal tracking it across all the angiographic images in a cardiac cycle. A two click Fast Marching Method for the centerline extraction phase, whereas a gradient vector flow (GVF) active contour model coupled with a pyramidal Optical Flow approach is used for the temporal tracking procedure. These tools lead towards the reconstruction process of a targeted coronary artery. Valid 3D reconstruction relies on the successful correspondence of appropriate landmarks obtained from the arteries through a set of angiographic images. By introducing a novel curvature constraint that relates 2D-3D coronary curvature, we aim at reconstructing structures that undergo mainly rigid motion, since if this is the case, the structure would not undergo contractility and hence would have similar curvature between its 3D geometry and its projected shape in the 2D images. The proposed algorithms were validated by using 5 consecutive biplane image pairs, we obtain the following research results: for a posterior/anterior view, the 3D RMS improves from 2.8mm to 1.1mm, whereas for a left-right anterior oblique setup the 3D RMS improves from 3.1mm to 1.9mm using our geometric curvature constraint.