Study On The Application Of D-S Theory To Medical Image Fusion Segmentation Of The Human Brain

Medical image segmentation, as an auxiliary means of medical clinical diagnosis, could be particularly significant in medical applications. It is the basis of structure analysis of medical image, dynamic image analysis,3D image visualization, image guides surgery and tumor radio-therapy research, etc. The properties of medical image makes segmentation even more difficult, because of the complexity and diversity of the images, the noise during the process of imaging technology, and illegibility of the edge of the images. In order to improve the accuracy of segmentation, satisfactory results could not be obtained by a single segmentation algorithm, some special algorithms or effective combination of a variety of algorithms for specific image was used. According to the imaging characteristics of an object of segmentation, the more information the images possess, the more accurate the segmentation result is.In order to improve the automation and accuracy of segmentation, according to the feature of D-S evidence theory that it can integrate multivariate information, this paper reintegrated segmentation results of MRF method and Two-dimensional histogram of fuzzy clustering method, to segment the human brain MR images. Satisfactory segmentation result was obtained. The main research results can be concluded as follows:In this paper, features, importance, existing problems and present situation of the medical image segmentation were analyzed. And the methods of medical image segmentation were given and discussed.The MRF theory and the fuzzy set theory are studied. Probability distribution model of image information, the spatial features of image information, potential function model and its parameter estimation were studied.D-S evidence theory and its application in image fusion segmentation were researched. The basic probability assignment function was mainly researched. And this paper presents a new method for obtaining basic probability assignment based on histogram of Gaussian distribution.Experimental verification:MRF segmentation method can fully reflect spatial features of image information, so it can deal effectively with noise and degradation; Two-dimensional histogram of fuzzy clustering method can deal effectively with fuzziness of images. These two methods can get good segmentation results, but there will be different divisions of the results when dealing with controversial image pixels. So the theory of D-S fusion segmentation was applied to segmentation results from the above two methods, and quantitative analysis of segmentation result was obtained.Experimental result shows:D-S evidence theory resolves classification of controversial pixels through fusing and segmenting the corresponding pixels from two images, to make up for deficiencies cause of inaccuracy and uncertainty in cross-region of segmentation result. Finally more reasonable and accurate segmentation results were obtained.