Interactive segmentation method for medical texture images

An interactive segmentation algorithm is proposed for medial texture images. It consisted of three steps: parameter training, fuzzy affinity-based region growing, and snake-based boundary refining. First, in parameter training, some seed points are specified by the user interactively in the desired object, which would improve the accuracy of segmentation. In region growing, the texture feature is introduced into the definition of fuzzy affinity relation proposed by J.K. Udupa in order to make fusion of the intensity and texture information, and a new method of selecting the optimal threshold is proposed to extract the object automatically. To refine the boundary of the object, an improved snake method is adopted, in which a new definition of the external energy is designed by incorporating the prior knowledge of the desired object. This method is applied to segment to CT and MR images with the regular texture information. The experiment results show that it can produce good results for this kind of image.