A Coronary Artery Segmentation Method Based on Graph Cuts and MultiScale Analysis Chaima Oueslati 1,2(B ) , Sabra Mabrouk 1,2 , Faouzi Ghorbel 1,2 , and Mohamed Hedi Bedoui 1,2 1 CRISTAL Laboratory, GRIFT Research Group, National School of Computer Science, 2010 Manouba, Tunisia {chaima.oueslati,sabra.mabrouk}@ensi-uma.tn, faouzi.ghorbel@ensi.rnu.tn, medhedi.bedoui@fmm.rnu.tn 2 Laboratory of Biophysics, Faculty of Medicine of Monastir, TIM Team, University of Monastir, 5019 Monastir, Tunisia Abstract. In this paper we propose a new multi-scale fully automatic algorithm based on Graph cuts for vessel extraction. In fact, we combine vesselness, geodesic paths, a multi-scale edgeness map and the directional information for vessel tracking in order to personalize the Graph cuts approach to the segmentation of tubular structures. Keywords: X-Ray · Angiography · Segmentation · Graph cuts 1 Introduction Cardiovascular diseases are the leading cause of death in the developed countries essentially due to coronary atherosclerosis [1]. The medical imaging modality and the minor procedures most currently used to diagnosis the coronary dis- eases are the X-ray angiography. As angiographic images are subject to noise and radiological contrast agent that is widely heterogeneous, it is difficult to identify the arteries compared to the background. Segmentation is therefore nec- essary to extract the coronary arteries by eliminating artifacts contained in the background. Several methods were proposed for the coronary artery segmenta- tion. In fact, Vessel segmentation algorithms are the fundamental component of automated radiological diagnostic systems such as diagnosis of the vessels (e.g. stenosis or malformations) and registration of patient images obtained at differ- ent times. Two categories of vessel segmentation algorithms are distinguished. The first one is skeleton based technique which aim first to extract median blood vessel and then connect these centerlines and estimate the vessel width to develop the vessel tree. The main idea of non skeleton methods is to directly extract blood vessels based mostly to the pixels intensity. In this context, sev- eral works were proposed to solve the vessel segmentation issue including basic ones such as thresholding and morphological operator. In [2], the authors pro- pose a fuzzy clustering where each data point can belong to more than one cluster and clusters are determined via similarity measures, such as distance, c  Springer International Publishing AG 2017 B. Ben Amor et al. (Eds.): RFMI 2016, CCIS 684, pp. 141–151, 2017. DOI: 10.1007/978-3-319-60654-5 12