形变引导正则化的医学图像Demons快速配准算法  

Fast Demons Registration Using Deformation Guided Regularization for Medical Images

作　　者：杜晓刚 王玉琪 王福海 雷涛[1,2] 张学军[3] DU Xiao-gang;WANG Yu-qi;WANG Fu-hai;LEI Tao;ZHANG Xue-jun(Shaanxi Joint Laboratory of Artificial Intelligence,Shaanxi University of Science and Technology,Xi′an 710021,China;School of Electronic Information and Artificial Intelligence,Shaanxi University of Science and Technology,Xi′an 710021,China;School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]陕西科技大学陕西省人工智能联合实验室,西安710021 [2]陕西科技大学电子信息与人工智能学院,西安710021 [3]兰州交通大学电子与信息工程学院,兰州730070

出　　处：《小型微型计算机系统》2022年第12期2580-2590,共11页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61861024,61871259,61762058)资助;甘肃省自然科学基金项目(20JR5RA404,21JR7RA282)资助;陕西省创新能力支撑计划项目(2020SS-03)资助;陕西省杰出青年科学基金项目(2021JC-47)资助;陕西省重点研发计划项目(2021ZDLGY08-07)资助。

摘　　要：Demons配准算法通常采用各向同性的Gaussian滤波器作为正则化项,该操作忽略了图像灰度变化的空间各向异性和形变场信息,造成在目标边缘等细节信息丰富的区域配准误差较大.此外,由于正则化过程通常需要逐像素进行平滑操作,导致Demons算法针对大尺寸图像进行配准时效率有待提高.针对上述问题,提出了一种基于形变引导正则化的医学图像Demons快速配准算法(DGR Demons).该算法有3个优势:首先,DGR Demons通过引入各向异性的引导滤波器作为配准过程的正则化项,有效保留了图像边缘细节,避免了在图像边缘处发生梯度越变;其次,DGR Demons用待配准图像间的形变来引导正则化过程,由于充分利用了形变的空间信息,获得了更加精确的配准结果;最后,DGR Demons通过对形变进行下采样,使得正则化中的平滑映射关系计算在低分辨率形变上执行,从而有效减少了配准耗时.实验结果表明,提出的DGR Demons实现了更快、更精确的配准结果,与主流Demons算法相比,将配准精度提高了约40%,配准效率提高了约8%.Demons registration algorithm usually considers the isotropic Gaussian filter as the regularization term,which ignores the spatial anisotropy of the intensity change of images,resulting in clear registration error in areas with rich detail information such as object edges.In addition,since the regularization process usually requires pixelwise smoothing operation,the efficiency of Demons algorithm needs to be improved for large-scale images.To address these issues,a deformation guided regularization based Demons registration algorithm,namely DGR Demons,is proposed in this paper.First,DGR Demons introduces the anisotropic filters as the regularization term of the registration process,which effectively preserves the edge details of images and avoids the gradient change at the edge of images.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 the low-resolution deformation,which effectively reduces the registration time.Experimental results show that the proposed 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%.

关 键 词：医学图像配准 DEMONS 正则化 非刚性形变 各向异性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]