Research On Medical Image Registration Technology Based On Mutual Information

Medical images are achieved in different time and (or) in different space and (or) by different imaging machines. Diversified images reflect different kinds of attribute information of the same patient, some information is about anatomical structure, and some is about physiological characters. Essentially, the pieces of information are uniform, but now they are spreaded around different images. In order to provide more exact and comprehensive information to iatrical diagnose and therapeutic plan, different images are needed to be registered, that is to say, the space and anatomical location of the corresponding pixels should be on all fours and form a new 2D image, each pixel of the 2D image contains many kinds of information.In this paper, the method that combining Canny edge detect operator with mathematical morphology was firstly used to pick-up the figures of brain entity, then Moment and principal axis method based on the figures was used to do rough registration. The characters of normalized mutual information measure based on Shannon's entropy and Renyi's entropy were researched secondly. And the result is that normalized mutual information measure based on Renyi's entropy under some parameters can remove some unwanted local extrema, but normalized mutual information measure based on Shannon's entropy has the"depth"of the basin of attraction, so in this paper the normalized mutual information measure based on Renyi's entropy and Shannon's entropy were mixed to form the similarity metric and different entropy was used in different searching phases. At first, Adaptive genetic algorithm was used to searching the global initial extremum of normalized mutual information measure based on Renyi's entropy. Then, Simplex method was used to locate the exact global optimal solution by the current global initial extremum and the normalized mutual information measure based on Shannon's entropy was taken as the similarity metric. Finally, the following three methods were compared: 1)normalized mutual information measure based on Shannon's entropy as similarity metric, Genetic algorithm as optimize algorithm; 2)normalized mutual information measure based on Shannon's entropy as similarity metric, Moment and principal axis method completed rough registration, and then took Genetic algorithm as optimize algorithm to complete exact registration; 3)normalized mutual information measure based on mixed entropy as similarity metric, Moment and principal axis method completed rough registration, and then took Adaptive genetic algorithm and Simplex method as optimize algorithm to complete exact registration. The results of the experiments proved that the third algorithm is effective. It takes less time and the result is more exact than the other two.