Research Of Medical Image Registration Based On Mutual Information And Genetic Algorithm

Image registration is a process of mapping and overlying at least two imagesacquired under different transducers, different times and different conditions, whichhas been extensively applied on medical image processing, mainly including remotesensing images,medical imaging and image processing etc.Medical image registrationwhich used in clinical research in order to find an optimal transform which makes onemedical image consistent with another medical image on the corresponding point inspace.This consistency is the same anatomical point of the body in two images has thesame spatial location,for example: the same size, consistent point of view, alignedposition.The result of registration makes all of the anatomical points or mostdiagnostic significance of the points have reached match.The project of this paper originates from two research projects of Medical ImageRegistration Model Research Based on Knowledge Base and Evaluation ModelResearch on CT and MRI Image Registration and Fusion in the science andtechnology development plan of Jilin Province. The medical image registration of CTand MRI is mainly studied, which provides basis for image fusion. All the medicalimages are provided by The Second Clinical Hospital of Jilin University.Normalized mutual information (NMI) method is adopted in this paper. Hybridoptimization algorithm is taken as the optimization algorithm in medical imageregistration. A kind of registration algorithm based on NMI and genetic algorithm isdesigned and the registration between two medical images is realized.Firstly, this paper introduces the research background, research significance,foreign and domestic research status of image registration and basic concepts andtheories of medical image registration, including the method, classification andprocedures of image registration. Then, this paper focuses on introducing the similarity measurement andoptimization algorithm in medical image registration. In the image registration, thecommonly used the similarity measurement methods include AM measurement, grayvariance and mutual information, etc. Optimization algorithm methods include Powell,the simplex method, particle swarm algorithm and genetic algorithm etc. Through theintroduction and comparison of various algorithms, the similarity measurementfunction and optimization algorithm used in this paper are selected.Thirdly, the research based on similarity measurement and optimizationalgorithm confirms the normalized mutual information and genetic algorithm used inthis paper. The algorithm for image registration is experimented to prove its feasibility.A hybrid optimization algorithm is then presented to combine the global optimizationalgorithm and local optimization algorithm, namely the combination of geneticalgorithm and simplex method. The immigration operation is added to the geneticalgorithm, making the image in the registration process overcome the local maximumproblem and improve the precision of the algorithm.Finally, the result of medical image registration based on improved geneticalgorithm is evaluated. The evaluation methods in image registration are divided intotwo classes: Subjective evaluation method and objective evaluation method. Thispaper evaluates the experimental result from last chapter with many evaluationmethods. The improved genetic algorithm has higher accuracy and better robustness.Through the research of medical image registration based on the normalizedmutual information and genetic algorithm research, the implementation process andthe theoretical basis of the technology is clearly understood. A medical imageregistration system is designed, which can be used for rigid registration of2dmultimode medical images.