Segmentation Of Abdominal CT Image Based On Super-pixel

Medical image segmentation is the first phase of medical imaging data analysis and visualization, which are also the precondition and crucial to computer-aided diagnosis, image-guided surgery, virtual endoscopy and many medical image applications. Compared with osseous organs, human abdominal soft-tissues are more complex and deformable, including the liver, the kidney, the gallbladder, the spleen as well as vascular such as the veins and arteries. However, due to the limitation of imaging device and the peristalsis of tissues, they often exhibit intensity inhomogeneity and overlapping in abdominal CT series. Besides, the blurred organs and the ambiguous of the edge of lesions also bring some considerable difficulties for segmentation. And for abdominal multiple organs segmentation, there will be much more influential factors, including high similarity of adjacent organs, partial volume effects and the relatively high variations of organ position and shape. Therefore, multi-object segmentation in abdominal CT image is still a challenge task.In this paper, the abdominal structure, the properties of organs and the characteristic of CT imaging have been studied. We propose a series of segmentation algorithms for abdominal CT image based on super-pixel. The main contents of the research are as follows:(1) There is a super-pixel pre-segmentation technique proposed for medical image, which generates the visual patch by clustering pixels based on both intensity similarity and spatial proximity.(2) An unsupervised and fully automatic method for segmentation has been proposed. Inspired by biology, there is a segmentation algorithm based on super-pixel clustering, which simulates the human visual selection mechanism to carry out the saliency map by calculating the center-surround contrast of region’s texture and intensity feathers. Finally, the target regions are segmented.(3) Combined with multi-classification theory of machine learning, a multi-object and multi-level supervised segmentation algorithm is put forward for abdominal CT image, which established the directed adjacency graph of super-pixel as spatial relationships of abdominal organs to restrict the classification. Experimental results prove the higher accuracy of the segmentation.Compared with the advanced medical image segmentation algorithms in The Cancer Imaging Archive database, the segmentation algorithms of this paper, which reduce the time cost and increase the accuracy of segmentation, have a higher adaptability and robustness.