Angiographic vessel segmentation for CT registration Antonio Hern´ andez * , Carlo Gatta *† , Petia Radeva *† , Laura Igual *† , Rub´ en Leta ‡ and Sergio Escalera *† * Computer Vision Center Edifici O, Campus UAB 08193 Bellaterra , Barcelona Email: {ahernandez, cgatta, petia, ligual, sescalera}@cvc.uab.cat † Departament de Matem` atica Aplicada i An` alisi, Facultat de Matem` atiques Universitat de Barcelona Gran Via de les Corts Catalanes, 585 08007 Barcelona ‡ Secci´ on de Imagen Card´ ıaca y Unidad de Hemodin´ amica. Hospital de la Santa Creu i Sant Pau, Barcelona. Email: Cardiologia@santpau.es Abstract—Angiographic vessel image registration is a really challenging problem due to difficulties like the absence of some fragments of vessel and the non-rigid deformations they suffer caused by the breathing and heart beating of the subject. In order to make this registration easier, we propose a prior automatic segmentation of the vessels in the images using graph-cuts. I. I NTRODUCTION Chronic total occlusions (CTO) are obstructions of native coronary arteries with the presence of Thrombolysis In My- ocardial Infarction (TIMI) flow grade 0 within the occluded segment with an estimated occlusion duration of more than three months. Recanalization of a CTO still remains a chal- lenge for invasive cardiologists. New imaging technologies may help selecting those candidates with more chances to have successful recanalization and a higher likelihood to improve regional function after percutaneous coronary inter- vention (PCI). Recent studies try to implement new imaging techniques to improve the success rate of CTO recanalization. Multislice Computed Tomography (MSCT) has emerged recently as a valuable technique for the non-invasive visu- alization of both the lumen and the features of the arterial wall of coronary vessels [3]. The importance of registration of CT to X-Ray images has been reported as a valuable tool to provide complete and high quality 3D information in addition to the poor data provided by Xray images [4]. Moreover, prior segmentation of the vessels is a typical step to apply before registration, since a lot of background noise is removed and the registration method can be simplified to work only with binary images -the segmentation masks obtained-. In this paper, we present a segmentation method for angio- graphic vessels, based on the graph-cuts energy minimization framework. We define new unary graph-cuts potentials which fit the arterial structures present in X-ray images. The structure of the paper is as follows. The segmentation method is explained in section II. Images from the dataset and some qualitative results are shown in section III. Finally, section IV ends the paper with the conclusion. II. METHOD The vessel segmentation problem is posed as an energy minimization problem, solved by means of graph-cuts [1]. In this framework, an energy function E(A) is designed such that the minimum value of this function corresponds to the optimal segmentation of the image. In order to minimize this energy function, a graph is constructed from the image in the following way: each pixel in the image is mapped to one node in the graph, and these nodes are interconnected following a neighbouring criterion. Furthermore, two additional nodes T and S called Terminal nodes are added to the graph, and connected to all the rest of nodes. Considering the image as a vector of pixels I = {I 1 ,I 2 , ...I n }, the segmentation mask is defined as A = {A i ,A 2 , ..., A n } where each A i is either 0 or 1 classifying the corresponding pixel as background or foreground. The energy function is divided in two weighted terms: E(A)= λR p (A)+ B(A) (1) The first term R p (A) is called the unary potential and encodes information at the pixel level. This potential give values to the edges connecting each pixel node in the graph to the terminal nodes. The second term B(A), or pairwise potential, encodes