Multi-modality Medical Image Registration

With the rapid development of medical technology,more and more medical imaging methods have been invented.A variety of imaging equipment provides a wealth of medical images for clinical diagnosis and treatment,such as CT,PET,MRI,etc.The combination of these different types of images can provide a more comprehensive diagnosis and treatment information for doctors.Multimodal medical image registration is the key and foundation of medical image fusion.In this paper,the registration method of multimodal medical images was studied,focusing on the registration method based on the mutual information and gradient information.An in-depth study was made on the extraction of gradient information,the combination method of mutual information and gradient information as well as improvements of the registration speed,which were then fully verified.The main work of the paper is as follows:(1)A multi-resolution image registration method based on the combination according to weight ratio of normalized mutual information to gradient information was proposed.First,low resolution images were obtained by wavelet decomposition of the original images,and then the coarse registration was performed on the low resolution images,so as to determine the approximate scope of search and the combination weight ratio of mutual information and gradient information.Finally,precise registration was made on the original images.This method can effectively speed up the registration and improve the registration accuracy.(2)The extraction method of gradient information was improved.The gradient information mainly reflects the contour information of bone edge on the images.In this paper,based on the analysis and experiment results,the preliminarily extracted gradient amplitude images were multiplied with the original grey-scale images after Gauss fuzzy and operation open,which can better highlight the major information such as bone edge and effectively control the interference information such as noise and soft tissue.(3)In order to improve the registration speed and accuracy,the step-by-step optimization algorithm combining the particle swarm optimization(PSO)algorithm and Powell algorithm was used in the paper.First,the PSO algorithm was used for global optimization of the low resolution images obtained by wavelet decomposition,and then the PSO optimization algorithm was used as the guidance of the Powell algorithm for the precise local optimization of the original images.This optimization algorithm can greatly speed up the registration speed and effectively avoid the phenomenon of local optimization.