Efficient Reconstruction of Coronary Vessels from 2D Angiography

The recovery of three-dimensional (3D) object information from biplane CT angiographic images is still a challenge in application of modelling coronary vessel trees. The proposed method of 3D reconstruction is obtained from the two-dimensional (2D) corresponding feature points by using triangulation projection based on the rotated mechanical angles of the CT angiographic machine. Those 2D feature points are on the centrelines of vessels. This research also includes how to extract the interested coronary vessels from CT angiographic images and how to represent those vessels as a group of connected 2D points that are on centrelines of them. One of the contributions in the paper is that we developed automatic searching algorithm to obtain the 2D corresponding points from a pair of biplane skeleton images under the two constraints of epipolar lines and corresponding bifurcations. We apply the Lucas-Kanade optical flow algorithm to register the corresponding bifurcations from an image sequence for the application of the proposed searching algorithm. The 3D reconstruction applies the readings of rotated mechanical angles and isocentre of the CT angiographic machine (Siemens Axiom) to establish the biplane image relationships of rotation and translation, and then projects the 2D centrelines of coronary vessels onto 3D world coordinates. Another contribution of the paper is that we apply the back projection algorithm to project the 3D points onto 2D image planes for evaluating the accuracy of the 3D reconstruction. The proposed method has been tested on patient data.