Quantitative Analysis of Deformable Model-Based 3-D Reconstruction of Coronary Artery From Multiple Angiograms

Abstract

Three-dimensional (3-D) reconstruction of the coronary artery is important for the diagnosis and interventional treatment of cardiovascular diseases. In this paper, a novel mean composited external force back-projective composition model is proposed and integrated into the deformable model framework for the 3-D reconstruction of coronary arteries from multiple angiograms. The parametric snake evolves toward the real vascular centerline in 3-D space based on the integrated internal energy and composited external energy. In addition, a polynomial function is constructed to determine the diameter of the cross section of the vascular segments, which fully utilizes the back-projection information of multiple angiograms. The deformable and proposed methods are comparatively validated using phantom datasets and routine angiographic images with respect to space and reprojection Euclidean distance errors. The experimental results demonstrate the effectiveness and robustness of the proposed model, which can achieve a mean space error of 0.570 mm and a mean reprojection error of 0.351 mm. In addition, the influence of the angle difference to the reconstruction accuracy is discussed and validated on phantom datasets, which demonstrate that an angle difference of for any two angiograms is suitable for the 3-D reconstruction process.

Reconstruction of coronary arteries from X-ray angiography: A review

“…In order to calculate 2D external energy terms for each landmark point, common approaches are to compute the Gradient Vector Flow (GVF) (Xu and Prince, 1998b), Generalized Gradient Vector Flow (GGVF) (Xu and Prince, 1998a), and Potential Energy (PE) (Cohen and Cohen, 1993) from the 2D projection images and use the resulting 2D vector fields (Cañero et al, 2000(Cañero et al, , 2002Cong et al, 2013Cong et al, , 2015Yang et al, 2014). Centrelines segmented from 2D projection images are used as the feature map input to GVF, GGVF and PE computation.…”

“…Therefore, the new 3D position of the landmark is found as the 3D position, which minimizes the distance to all projection lines (Figure 6a). Another strategy is to compose a 3D external force term using the 2D external force terms (Cong et al, 2013(Cong et al, , 2015Yang et al, 2014). To this end, the 2D external forces are back-projected onto the world coordinate system by ignoring the out-of-plane component of the 3D external force.…”

“…The back-projected external energy forces are added together to obtain the 3D external force (Figure 6b). The main advantage of this strategy is to update the 3D landmark points without violating the epipolar constraints, which is proven to increase the accuracy and the convergence rate (Cong et al, 2015;Yang et al, 2014). Additionally, this strategy is easy to adapt to multi-view scenarios since the 3D external force is given by a simple vector addition operation (Cong et al, 2015).…”

“…Initialization of the active contour model in 3D is performed manually (Cañero et al, 2000(Cañero et al, , 2002Cong et al, 2013Cong et al, , 2015Yang et al, 2014;Hui and Friedman, 2002;Zheng et al, 2010). Some corresponding points (including the start and end point) for a branch are selected from different views and rough reconstructions are obtained for these points.…”

“…(Cong et al, 2015) combined back-projections of 2D external forces to compute a 3D external force for an active contour based reconstruction method. (Blondel et al, 2006) generated 3D reconstructions for every pair of multiple views and fused them together to find the final reconstruction.…”