Segmentation of coronary arterial tree using localized deformable model embedded with automated seeds

Abstract

This paper presents a fully automatic approach for isolating the left and right coronary arteries from CTA images by embedding our improved fast seed detection method into localized active contour model. Usually active contour based methods require starting point known as seed for their evolution. Accurate provision of this seed point leads to the accurate segmentation. Manual feeding of seed point requires expertise as well as may lead to wrong segmentation. Therefore, in this paper we have combined the quantile and median based thresholded Hessian-based vesselness with that of local geometric features of the vessel to detect the coronary seed points accurately in an automatic fashion. Further, the detected seed points are fed to the active contour model which evolves in a localized way to track the entire coronary arteries to their distal ends. The obtained seed points as well as the obtained segmented left and right coronary arteries are verified by the radiologist at each step. The method is evaluated and validated on nine real clinical CTA datasets and also compared with the previous methods proposed by Lankton et. al and Khedmati et. al. Experimental results reveal that the proposed method outperforms the previous methods qualitatively as well as quantitatively. © 2005 – ongoing JATIT & LLS.