Research On Non-Rigid Medical Image Registration Technology

Non-rigid medical image registration technology restores the corresponding relationship of spatial position and anatomical structure of two medical images with non-rigid deformation by searching for some non-linear spatial transformation.Non-rigid medical image registration technology can provide complementary information of medical images about the same or multiple modalities.Therefore,non-rigid medical image registration technology is one of the essential research contents in the computer vision and medical image processing fields.And it has been widely applied in many medical fields such as neurodegenerative disease analysis,tumor radiotherapy localization and image-guided surgery navigation.Since the 21^st century,with the development of various novel registration models and efficient optimization methods in non-rigid medical image registration technology,the accuracy and applicability of non-rigid medical image registration have significantly been improved.However,the existing non-rigid medical image registration methods do not have satisfactory performance in global complex deformation and multi-modal medical image registration.This thesis mainly focuses on the non-rigid mono-modal medical image registration method which combines Demons algorithm and SIFT algorithm,as well as the non-rigid mono-modal and multi-modal medical image registration method based on the non-linear elastic model.The main work of this thesis is summarized as follows:1.This thesis introduces the background and significance of non-rigid medical image registration technology,expounds the development and research status of non-rigid medical image registration technology,and discusses the classification and methods of non-rigid medical image registration technology in detail.2.This thesis describes a basic model of non-rigid medical image registration,introduces the process of non-rigid medical image registration,discusses the contents including image preprocessing,image feature searching and matching,construction and optimization of target energy function,interpolation transformation.3.To solve the problem that Demons algorithm cannot deal with large deformation medical image registration,a non-rigid mono-modal medical image registration method combining Demons algorithm and SIFT algorithm is proposed.In the framework of the standard registration model,Demons algorithm and SIFT algorithm are used to construct a new driving displacement field energy function,and the SIFT feature term is used to constrain the deformation displacement of the displacement field,which makes up for the inadequacy of Demons algorithm in processing large deformation medical image registration.Simultaneously,feature pairs optimization strategy,bidirectional registration strategy and multi-resolution strategy are added to the algorithm to improve the efficiency and accuracy of registration.Experiments show that the improved algorithm can solve the registration of small deformation region and large deformation region at the same time.4.To solve the problem of discontinuous displacement field caused by only using the gradient information of image to drive the registration method based on diffusion model,a non-rigid mono-modal medical image registration method based on the non-linear elastic model is proposed.Firstly,a non-linear elastic registration model based on the linear elastic registration model is constructed.The non-rigid deformation of the image is simulated as the elastic fluctuation,which ensures the continuity and smoothness of the displacement field.Then,the gradient of the Sum of Squared Differences is added to the displacement field equation of the nonlinear elastic model as the driving force.Finally,the transform displacement field is obtained iteratively by using multi-resolution strategy,and the two images are registered.Experiments indicate that the proposed method has a higher registration accuracy when dealing with the global complex deformation registration problem.5.To solve the problem which the non-rigid medical image registration method based on image intensities cannot deal with multi-modal medical image registration,a non-rigid multi-modal medical image registration method using local structure tensor conservation as similarity measure is proposed.Firstly,a similarity measure based on local structure tensor conservation is constructed.Then,the gradient of similarity measure based on local structure tensor conservation is used as the driving force of the nonlinear elastic model,and a non-rigid multi-modal medical image registration method based on the nonlinear elastic model could be obtained.Finally,the method is tested by selecting several sets of multi-modal medical image sequences.The experimental results show that the method performs well in multi-modal medical image registration.