[nwrnationalJournal of CardiacImaging 10: 67-70, 1994. 67 @ 1994 KluwerAcademic Publishers. Printed in the Netherlands. Three-dimensional coronary arteriography Anne Roug6e, Catherine Picard, Didier Saint-Fdlix, Yves Trousset, Thierry Moll I & Michel Amiel 1 GE Medical Systems, 283 Rue de la Minikre, 78 530 Buc, France; 1 Hospices Civils de Lyon, Lyon, France Accepted27 September 1993 Key words: biplane acquisition, coronary arteriography, three-dimensional reconstruction Abstract In this paper we present a new imaging technique for three-dimensional (3-D) X-ray coronary arteriography. The goal is to provide in near to real-time a 3-D representation of the coronary arterial tree, helpful to better understand its topology and locate the possible lesions. The 3-D reconstruction of the coronary arteries is obtained from a set of X-ray conic projections acquired during a rotation of the imaging chain around the patient. Images are taken before and after injection of contrast agent. A subset of mask and opacified images is selected, corresponding to the same phase in the cardiac cycle. These images are subtracted and corrected for geometric distortion. The reconstruction is performed by using a two-step non-parametric detection/estimation method. Due to heart motion and propagation of the contrast agent, the number of available projections is very small. Typically 4 or 6 projections are available if the opacification is stable during 2 or 3 cardiac cycles and when using a biplane acquisition system. High resolution 5123 reconstructions of the coronary arteries from a cadaver heart are presented, with a voxel size of 0.4 ram. The 3-D reconstruction provides a good 3-D representation of the global structure, even with a number of projections as small as 4. Introduction In recent years, several methods based on image pro- cessing and computer vision concepts have been pro- posed for the 3-D reconstruction of blood vessels from a very limited number of 2-D X-ray digital angiograph- ic projections - typically one or two pairs of biplane images [ 1-4]. Most of these methods rely on an elliptic model of vessel cross-section, thus providing a good solution only for normal and simple coronary arterial trees. Furthermore, for the sake of robustness, these methods have to allow some degree of user interaction and in any case can provide a good result only under good acquisition conditions, i.e. when the arteries are all well seen on each projection image with no super- imposition. This usually requires several injections in order to find the best incidences in which one chooses the images appropriate for the reconstruction, since the global 3-D geometry of the arterial tree is unknown. Thus, these methods cannot be used automatically and efficiently under clinical conditions. We introduce a new concept of 3-D coronary arte- riography, in which the blood vessels are reconstructed by using a methodological framework closer to com- puterized tomography than to computer vision. This approach has the potential of offering in near to real- time a robust, fully automatic method for a global 3-D representation of the coronary arterial tree. Although it is not claimed that finest details could be reconstruct- ed with high precision due to the very low number of projections available, the 3-D informations provided by this technique are very helpful to better understand the topology of the coronary arterial tree and locate the possible lesions. Furthermore this technique may be useful to find a good positioning of the imaging chain for further quantitative analysis using the above- mentioned techniques for example.