Carotid artery segmentation in 3D ultrasound images using a hybrid framework

In this paper, we propose a semiautomatic hybrid framework including region growing algorithm, marching cubes method and deformable model for the segmentation of carotid artery in 3D ultrasound (US) images. In this framework, double-threshold method is used to preprocess the initial images before using region growing method to achieve rough segmentation. Because of the effect of inherent speckle noise in the US images, there are some small holes in the images segmented and the contours are discontinuous. So it is necessary to apply morphologic closing computation to image slices, in order to provide a better estimation of the arterial wall. Then we apply marching cubes method to construct a deformable mesh surface based on the rough segmentation result as the initial active contour of deformable models. At last we deform the 3D mesh model using external image forces so that the model converges to the authentic object surface. We take full advantage of region-based methods and boundary-based methods to achieve semiautomatic method, and the quality of the segmentation of 3D US images is improved. All of the modules in the hybrid segmentation framework are developed with the context of the Insight ToolKit (ITK). We show the experimental segmentation results of carotid artery in 3D ultrasound (US) images with high quality segmentation results and computational efficiency. Meanwhile, we also detect this methodology through liver data in 3D US images which roughly have the same organization structure features as carotid artery, consequently, we get the expected experimental segmentation results.