Global optimization for 3-D reconstruction of coronary artery trees from angiographic image sequence

In this paper, a novel method is developed for the 3-D reconstruction of coronary artery trees from two angiographic images. Due to the rotational distortion and unpredictable motion of the imaging chain, the commonly used pinhole camera model cannot obtain desirable reconstruction of the vascular trees. Curvature Scale Space (CSS) approach is used to find correspondence by choosing two similar views between the various views obtained by the angiographic technique in one cardiac cycle. Therefore, in this study the curve matching and epipolar constraint are integrated to calculate the correspondences in two different angiographic images. Also, a non-linear optimization method is developed for the refinement of the vascular structures and the transformations of the angiographic system using the fundamental matrix and Levenberg-Marquardt (LM) algorithm. The variables available for optimization are verified extensively which ensures achievement of the global minimums and at the same time make corrections in the low order geometric distortions. Experimental result shows that the proposed framework can guarantee high accuracy of reconstruction result. General terms: angiographic images, Curve matching.