英文摘要 | From 1970's, CT, MR, PET and other medical imaging instruments had been successfully applied to clinic medicine. With the development of computer graphics, pattern recognition, artificial intelligence, virtual reality and computer networking, a new branch of research medical imaging process and analysis is coming into being and in the ascendant. The emphasis of this dissertation is the research on medical image segmentation, which is one of the bottlenecks of medical image processing and is a fundamental building block for higher-level image analysis. Image segmentation remains a difficult task, however, due to both the tremendous variability of object shapes and the variation in image quality. In particular, medical images are often corrupted by noise and sampling artifacts, which can cause considerable difficulties when applying classical segmentation techniques. As a result, none of the method that had good result for general images had been proposed up to now. In practice, in order to obtain the boundary of a 3-D object, there's usually tens, even hundreds of slices to be segmented. It is very tedious that an expert anatomist manually delineates the boundaries of different structures. Then it is necessary to introduce the interaction of experts to the segmentation algorithm. Under the guidance of this idea and aiming at the character of medical imaging, several interactive segmentation methods are proposed for specific medical applications in this dissertation. The main work of this dissertation is as follows: 1. For semi-automatically segmenting medical image series, a new algorithm combining the modified live wire algorithm and the T-snake model is proposed. First, the robust anisotropic diffusion filtering is used to smooth the images while keeping the edges. Then, the traditional live wire algorithm is modified by combining it with the watershed method, and one or more slices in a medical slice series are segmented accurately by the live wire algorithm. Next, the computer will segment the nearby slice using the modified T-snake model. To make full use of the correlative information between contiguous slices, a gray-scale model is applied to the model to record the local region characters of the desired object, and a new functional definition of the external energy is proposed. Furthermore, when the initial contour crosses the desired object, the traditional T-Snake model may fail to recover the boundary. By finding the internal point via region growing this problem is solved. The experiment results show that this algorithm can recover the boundary of the desired object from a series of medical images quickly and reliably with only little user intervention. 2. The relative definitions of fuzzy connectedness are extended for the purpose of segmenting medical texture images. Fuzzy spel adjacent relation is modified to a sub-region based relation. The texture feature is introduced t |
修改评论