Based On Improved Mutual Information Method Of Medical Image Registration

With the development of medical and computer technology, medical imaging technology has been applied in the whole process of clinical activities.As different devices with different imaging principle, medical imaging has a variety of images with different mode to provide different information.In order to acquire comprehensive information of pathology in clinical,doctors often image the same patients with different modes or many times, and finally integrate information of multiple images for comprehensive analysis.The new image which is integrated from different modes images called the medical image fusion and the first step of fusion is image registration. Medical image registration as a prerequisite for image fusion, its research is a hot in the area of medical image processing. In this paper, we have done the deep research for the method of medical image registration.From the framework of image registration, we have done the main research of the method of image registration based on maximization of mutual information and analyzed its advantages and disadvantages.So we have proposed measures both from aspect of the accuracy and speed for improving the mutual information shortcoming.To speed up the rate of image registration and overcome the large calculation and the presence of local extremum of mutual information, we have focused on sampling and optimization strategy. This paper proposed sampling methods based on regional of image information entropy and designed mixture optimization algorithm of Powell combined with simulated annealing for optimizing the mutual information function. In the end, we done the emulation experiment for analysis that the sampling method based on information entropy combined with mixture optimization algorithm can not only keep the image without losing important information, but also significantly speed up the rate of convergence of measure function in the process of registration.In order to avoid the situation of mutual information registration error, for instance, two images mixed with a lot of noise, smaller overlapping area, lower contrast and the image data lost, it is difficult to obtain accurate registration results only using the method of mutual information. This paper have proposed two improved method of mutual information, one method is introducing the spatial information of image to the mutual information function, that is, the registration method of mutual information combined with the edge correlation deviation. The other method is introducing a region of interest (ROI) of image into the mutual information function, namely, the registration method of the mutual information combined with the region of interest's mutual information. There are two combination forms with weighted mutual information and weighted probability of mutual information respectively has been used in constructing the measure function.In the method of mutual information combined with the edge correlation deviation, Susan operator can be used in detecting the edge of medical image according to its advantages of anti-noise and flexibility in controlling parameters.Finally, experiment for these improved methods shows that the accuracy and stability of improved methods are more than the mutual information alone.