Research On Medical Image Demons Registration Algo-Rithm Based On Anisotropic Regularization

Medical image registration is the process of matching the moving image with the corresponding point of the reference image in the spatial position through spatial transformation.It is the basis of medical image 3D reconstruction and medical image fusion.At the same time,it has important application value in organ motion correction,image-guided surgery,radiotherapy planning and other clinical applications.In recent years,a large number of non-rigid image registration algorithms have been proposed by many scholars to effectively describe irregular complex deformation.Among the non-rigid registration algorithms,the Demons registration algorithm has been widely concerned by scholars because of its complete theoretical basis.However,the Demons algorithm smoothes the deformation through isotropic regularization,which easily leads to the decline of registration accuracy in the areas with rich details of medical images or in the movement process of tissues and organs,so it can not well fit the movement process of tissues and organs.In view of the above problems,this thesis studies the Demons registration algorithm for medical images based on anisotropic regularization,which is mainly used to solve the problem of low accuracy of organ motion registration caused by isotropic registration model,and further improve the implementation efficiency of registration.The main research of this thesis is as follows:(1)The isotropic regularization model is used for the registration processing of the classic Demons registration algorithm,which leads to large registration errors in areas with rich image details,as well as the problem that the smooth operation of per-pixel regularization is easy to lead to low efficiency in the registration processing of large size images.A deformation guided regularization based Demons registration algorithm,namely DGR Demons,is proposed in this thesis.The algorithm has three advantages:Firstly,the DGR Demons introduces the anisotropic filters as the regularization term of the registration process,which effectively preserves the edge details of images and reduces the registration error of complex organ motion.Secondly,DGR Demons employs the deformation field between reference and moving images to guide the regularization process,and makes full use of the spatial information of the deformation field to obtain more accurate registration results.Finally,by down-sampling the deformation field,the smoothed maps in the regularization are performed on low-resolution deformation,which effectively reduces the registration time.Experimental results show that DGR Demons can rapidly achieve more accurate registration results.Compared with the popular Demons algorithms,the registration accuracy is improved by about 40%and the registration efficiency is improved by about 8%.(2)The Diffeomorphic Demons algorithm uses Gaussian regularization to smooth deformation,and the isotropic property makes it unable to effectively handle the sliding motion of tissues and organs.At the same time,the sum of squared differences is used to measure the similarity of the registration images,which is not suitable for accurately measuring the images with large gray difference,resulting in low registration accuracy of complex medical images.In order to solve the above problems,this thesis proposes an adaptive anisotropy regularization based Demons registration algorithm for medical images,namely A2 Demons.The algorithm has three advantages:Firstly,the adaptive anisotropic model is used as the registration regularization term,and the regularization of strong anisotropy is obtained by optimizing the weight value of local neighborhood variance.Meanwhile,the weighted average is used to realize the diffusion maximization,so as to truly simulate the complex movement of organs.Secondly,according to the different spatial information contained in the deformed image,the smoothness of regularization is adjusted by adaptive selection of kernel size in the regularization process,so as to obtain a good filtering effect and better retain the image details,further reducing the influence of parameter selection on the registration accuracy.Finally,the mutual information-sum of squared differences similarity measure is designed,which can effectively solve the problem of gray level intensity difference between tissues and organs to be registered,and improve the accuracy and robustness of registration results.Experimental results show that A2 Demons can handle complex organ movements well,and accurate registration results can be obtained at the same time.