Research And Application Of Nonrigid Medical Image Registration Based On B-spline And Mutual Information

Image registration techniques from birth to now have 30 years' history,because the development of the computer technology and medical imaging technology by leaps and bounds,all sorts of different modal medical images constantly emerge,containing different medical diagnostic information,providing abundant data for clinical application.Due to various imaging principle of the equipment,the providing medical information is greatly different.However,single-mode state image for the doctor's clinical diagnosis are single one-sided information.To get a more complete comprehensive complementary image information,We must fuse multiple modal images of different types of information together so that doctors make more accurate and reliable diagnosis and work out a more reasonable treatment plan.The current domestic and foreign technology in the field of rigid registration is ripe,but in fact human organs and tissues often exist all kinds of complicated nonlinear deformation.Only the nonrigid registration can meet the requirements,so medical image nonrigid registration is especially critical in the clinical treatment and has very important practical application significance.Nonrigid medical image registration will certainly be a hot spot in future medical digital technology.In this paper,we primarily study the nonrigid image registration key technology of the two big modules: one is geometric transformation model of reflecting the spatial relations between two images,and the other oneis similarity measure measuring whether two images have complete registration.Concrete research contents have the following several aspects:(1)We studied the framework,process and specific steps of the registration,concluded and summarized in detail the four key technologies of nonrigid image registration about geometry changes,gray level interpolation,similarity measure and optimization algorithm,expounded the spline function model,LBFGS optimization algorithm,and nonrigid image registration based on space transform and method based on the physical model.(2)We analyzed and deduced the relative mathematical expressions and matrix representation of b-spline curve and curved surface,then deeply analyzed the nonrigid registration model based on b-spline transform,elaborated the initialization of the grid method and the principle of free deformation model.Aiming at the condition of the nonlinear deformation image,it used nonrigid registration method based on b-spline,the experimental results showed that b-spline can fit nicely irregular deformation between two images;For the randomness of selection of b-spline control grid spacing,the density of control grid directly affect the effect of registration,the grid is too thin registration will faster but the precision is not enough and the grid is too dense with high precision but time is consuming,we can't well weigh the relationship between registration accuracy and efficiency,so put forward the multilevel b-spline nonrigid registration method;For uniform deformation field of multilevel b-spline can't very well simulate the image of local large deformation problems,put forward the local area multilevel b-spline nonrigid registration method.(3)The measurement of mutual information needs without any pretreatment operation such as feature extraction,detection,segmentation before the registration,only calculation for two image each information entropy and joint entropy.It is really simple and easy,but for this reason it ignores the spatial information of the existence of the image itself,affects the registration accuracy and robustness of the results.So this article introduced local mutual informationconcept which puts information in the image itself on a high status,analyzed the principle,calculation method and steps of local mutual information in detail;And put forward image registration method based on the local mutual information,the results of experiment indicated that the proposed scheme effectively improved the accuracy and robustness of registration,but there also be a big defect of time cost.(4)We conducted nonrigid registration application research of a prostate image based on b-spline and local mutual information,and expounded registration algorithm and steps of registration process on the whole;Then,introduced the key main function framework and the graphical user interface of registration process;Finally,compared the experimental results,the proposed solutions to the medical image of nonrigid deformation had a good effect of registration.