| Image segmentation,an image processing method to divide the picture into meaningful units by features such as texture,spectrum,shape,and color,plays a vital role in medical imaging.The liver is the largest substantial organ in the abdominal cavity of the human body.Contouring liver tumor on Computed Tomography(CT)image,a necessary procedure for diagnosis and operation planning in the treatment of liver cancer,is a laborious job usually manually performed by experienced clinicians.Automated liver tumor target segmentation may substantially improve efficiency.The level-set method is widely used in image segmentation.The evolution curve of the level-set is closed,vwhich can converge at the edge of the target reasonably.However,the current level set method usually assumed that the image intensity in a small local area is approximately homogeneous,which restricts its precision of segmenting intensity-inhomogeneous images,especially the CT images.The present study propo sed a novel level-set algorithm incorporating multiscale information,which leaves the segmentation curve unrestricted by a single scale during the evolution process and improves the segmentation accuracy for intensity inhomogeneous images.To verify the superiority of the multiscale level method,a systematic empirical study was performed using three independent sets of CT images of liver cancer,namely the big and smooth tumor,the small tumor within the liver,and the small tumor at the edge of the liver.Compared with traditional level-set methods(CV and LSACM),the proposed method performed better in segmenting all those three kinds of liver tumors,which was statistically significant.Experiment testing sensitivity to initial contour also revealed that the new algorithm was reasonably robust.The present study demonstrates that the multiscale level-set method can substantially improve the segmentation of intensity-inhomogeneous CT images,which has potential clinical value. |
PDF Full Text Request