Amr R. Abdel-Dayem and Mahmoud R. El-Sakka, "Segmentation of Carotid Artery Ultrasound Images Using Graph Cuts", International Journal for Computational Vision and Biomechanics, Vol. 3, No. 1, pp. 61-71, 2010.

Abstract

This paper proposes a scheme for segmenting carotid artery ultrasound images using graph cuts segmentation approach. Region homogeneity constraints, edge information and domain specific information are incorporated during the segmentation process. A graph with two terminals (source and sink) is formed by considering every pixel as a graph node. Each pair of neighbouring nodes is connected by a weighted edge, where the weight is set to a value proportional to the intensity of the gradient along them. Moreover, each graph node is connected to the source and the sink terminals with weights that reflect the confidence that the corresponding pixel belongs to the object and the background, respectively. The segmentation problem is solved by finding the minimum cut through the constructed graph. Experiments using a dataset comprised of 40 B-mode carotid artery ultrasound images demonstrates good segmentation results with (on average) 0.677 overlap with the gold standard images, 0.690 precision, and 0.983 sensitivity.