The Research Of Feature Selection And Segmentation Of Medical Image Based On Fusion

Medical image segmentation has become a hot issue of medical image processing and analysis research filed. There are many different types of brain medical mage and every kind of image has its respective merits and defects, so it has important significance of how to effective using various information from different images.In this thesis we researches fixed features segmentation, automatic features selection algorithm based on fusion and fusion rules on medical image of CT and MRI, as a result, a kind of automatic features segmentation scheme is formed.Trought a great deal of experiments, the thesis acquired a series of valuable results which can be summarized in the following aspects.Firstly, traditional cluster is initialization-sensitive, difficult to find out the number of prototypes and noise-sensitive, in order to overcome these shortcomings; adaptive FCM and ECM segmentation algorithm based on fixed features is proposed. In this method the cluster centers is obtain first using SSCL , and then treat the cluster centers as initial value of ECM and FCM, so adaptive image classification can achieve. At last post processing is implemented using space information. Experiment results show that proposed algorithm can adaptively ascertain number and center of classification and is a better solution of sensitive of noise in ECM and FCM, and prominently improve the accuracy of classification.Secondly, in order to utilize information of features, this thesis proposes automatic features segmentation algorithm. The automatic features segmentation algorithm fuse mutual information feature selection and cross-validation feature selection by fusion rules of DS to select features, then segmentation feature image make use of the segmentation algorithm presented in part one. The benifit of this method is that it filters noisy features and avoids influence of them, so the speed and accuracy of segmentation is improved.Thirdly, on the basis of the research DS evidential theory, we find out that it is deficient when evidences are conflicting. In this thesis, a kind of weighted fusing algorithm is presented based on classical sun quan'fusing method, it first computes the weight of every evidence according collision degree among evidences, then fuse all evidence by sun quan'fusing method. In this chapter, the new fusion method is applied to SSCLECM and MICVFS. At last, segmentation experiments on brain medical image have good effect.