Study On Non-rigid Registration Method Of Multimodal Medical Images Based On B-spline And Mutual Information

With the development of medical imaging technology and equipment innovation,medical image resolution and imaging accuracy continue to increase,making medical images widely used in clinical and medical research.High-precision medical image registration is an indispensable important technology in many medical image applications.It can match multiple types of information to the same space,which is convenient for disease analysis and diagnosis.However,the current non-rigid registration of different modal medical images lacks a universal method,and the rigid registration with wide adaptability cannot express the nonlinear deformation in tissues and organs,and has defects in the quality of registration.Therefore,exploring the non-rigid registration method of multi-modal images has important practical significance and is one of the hotspots of current research.In this paper,the key transformation models and similarity measures in the current multi-modal image non-rigid registration technology are studied in depth.The method is designed according to the characteristics of the two modal data of T1 and DTI in the magnetic resonance image,so that the solving method of deformation field can be improved to enhances the registration stability,and the GPU parallel processing method is used to increase the calculation speed of the similarity measure.And a visualization software is designed and implemented based on research.The main work and results of this paper are as follows:(1)For the image registration with non-linear deformation,b-spline function is selected as the deformation model,and normalized mutual information is used as similarity measure to evaluate the registration result,so that the registration has a good local deformation effect;A similarity measure method for unified image gray scale is proposed to solve the problem that the gray scale-based similarity measure is difficult to accurately express the relationship between images.(2)Considering the influence of grid size of traditional b-spline method on registration accuracy and efficiency,hierarchical b-spline registration is used to get close to the appropriate grid size and improve registration stability;Aiming at the problem that the deformation accuracy of the stratified b-spline of uniform grid is low in the area with too large difference,a local b-spline registration method based on block mutual information comparison is proposed,which improves the registration accuracy in the area with too large difference.(3)To solve the problem that traditional mutual information does not contain image spatial information and similarity measure is easily affected by local isolated points,the regional mutual information measure containing spatial information is used to replace the normalized mutual information measure and reduce the influence of isolated points on registration.A new method based on GPU parallel computing is proposed to solve the problem of high computational complexity and long time.(4)A medical image registration software was designed to be able to register and visualize T1 and DTI images.