Research On Medical Image Registration Method Based On SIFT And Mutual Information Algorithm

With the rapid development of medical imaging technology, the modality ofmedical images is more and more diversified. In clinical application, integrating these modality medical images is helpful to observe comprehensive information visually,improve the accuracy of clinical diagnoses and develop an appropriate treatmentprogram. However, the primary task is image registration due to the difference ofmedical images’ imaging principle, parameters and resolution.In order to achieve higher registration accuracy, this paper studies the ScaleInvariant Feature Transform (SIFT) algorithm and the mutual information algorithm inmedical image registration. The main work of this paper is as follows:(1) To overcome the shortcoming of the SIFT algorithm in medical imageregistration, this paper proposes an improved method. The method integrates Harrisalgorithm into SIFT so that it can extract more structural and significant points inoriginal image. Experiment results show that the proposed algorithm can increase the quantity of structural points and improve the accuracy of medical image registration.(2) To overcome the shortcoming of the Particle Swarm Optimization (PSO)algorithm in the registration based on mutual information, this paper proposes animproved method. This method adds the ideas of genetic algorithm (GA)’s copy,hybridization and variation into PSO so that it can solve the problem of local optimaand slow convergence efficiently. Experiment results show that the improved PSOmethod can obtain a faster and more accurate registration result.(3) To meet the higher registration accuracy, this paper combines with theimproved SIFT method and the registration based on mutual information. Though theimproved SIFT method can improve the accuracy at a certain extent, higher accuracy isrequired in medical application. Thus, it’s helpful to add the further registration withmutual information method.The accuracy and robustness of the proposed algorithm has been analyzed with alarge number of experiments in registration of ultrasound images, CT images and MRIimages. The results show that the two-step registration method can achieve betteraccuracy and robustness.