基于低维流形先验的低剂量CT重建方法  

Low-Dose CT Reconstruction Using Low-Dimensional Manifold Prior

作　　者：牛善洲[1,2] 梁礼境 李硕 张梦真 邱洋 刘汉明[1] 李楠[3] NIU Shanzhou;LIANG Lijing;LI Shuo;ZHANG Mengzhen;QIU Yang;LIU Hanming;LI Nan(School of Mathematics and Computer Science,Gannan Normal University,Ganzhou,Jiangxi 341000,China;Ganzhou Key Laboratory of Computational Imaging,Gannan Normal University,Ganzhou,Jiangxi 341000,China;School of Economics and Management,Gannan Normal University,Ganzhou,Jiangxi 341000,China)

机构地区：[1]赣南师范大学数学与计算机科学学院,江西赣州341000 [2]赣南师范大学赣州市计算成像重点实验室,江西赣州341000 [3]赣南师范大学经济管理学院,江西赣州341000

出　　处：《计算机工程与应用》2023年第18期242-248,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(62261002,11701097);江西省科技创新杰出青年人才培养计划(20192BCB23019);江西省重点研发计划一般项目(20202BBE53024);江西省“双千计划”科技创新高端人才青年项目(jxsq2019201061);赣州市科技创新人才计划。

摘　　要：针对低剂量CT成像问题,提出了一个基于低维流形先验的低剂量CT重建方法。对投影数据进行统计建模,将低维流形正则化作为先验信息引入到投影数据恢复过程中,从而达到抑制噪声的目的,再使用传统的滤波反投影(filtered back-projection,FBP)算法进行CT图像重建。在Shepp-Logan体膜实验新方法重建结果与FBP、PWLS-QM(penalized weighted least-squares via quadratic membrane)以及PWLS-DL(penalized weighted least-squares via dictionary learning)方法重建结果相比,相对均方根误差分别降低了64.87%、54.81%和7.02%;结构相似性指标分别提高了16.78%、1.88%和1.91%。在XCAT体膜实验中,新方法重建结果与FBP、PWLS-QM以及PWLS-DL方法重建结果相比,相对均方根误差分别降低了37.46%、22.17%和11.48%;结构相似性指标分别提高了3.33%、0.73%和1.22%。在临床数据实验中,新方法重建结果与FBP、PWLS-QM以及PWLS-DL方法重建结果相比,相对均方根误差分别降低了45.96%、25.61%和15.87%;结构相似性指标分别提高了19.12%、7.46%和8.63%。仿真和临床数据实验结果表明,新方法在有效抑制低剂量CT图像中噪声和伪影同时,可以很好地保持图像的结构信息和空间分辨率。In this paper,a low-dose CT reconstruction method using the low-dimensional manifold prior is presented.Incor-porating with the low-dimensional manifold prior,a penalty weighted least-squares(PWLS-LDMM)approach is presented to reduce the noise in projection(sinogram)domain,and then the image is reconstructed by filtered back-projection(FBP)algorithm.The relative root mean square errors of Shepp-Logan image reconstructed by PWLS-LDMM method are reduced by 64.87%,54.81%and 7.02%,respectively,as compared with that of FBP,PWLS-QM and PWLS-DL meth-ods.The structural similarity values of Shepp-Logan image reconstructed by PWLS-LDMM method are increased by 16.78%,1.88%and 1.91%,respectively,as compared with that of FBP,PWLS-QM and PWLS-DL methods.The relative root mean square errors of XCAT image reconstructed by PWLS-LDMM method are reduced by 37.46%,22.17%and 11.48%,respectively,as compared with that of FBP,PWLS-QM and PWLS-DL methods.The structural similarity values of XCAT image reconstructed by PWLS-LDMM method are increased by 3.33%,0.73%and 1.22%,respectively,as com-pared with that of FBP,PWLS-QM and PWLS-DL methods.The relative root mean square errors of clinical image recon-structed by PWLS-LDMM method are reduced by 45.96%,25.61%and 15.87%,respectively,as compared with that of FBP,PWLS-QM and PWLS-DL methods.The structural similarity values of clinical image reconstructed by PWLS-LDMM method are increased by 19.12%,7.46%and 8.63%,respectively,as compared with that of FBP,PWLS-QM and PWLS-DL methods.The simulation and clinical data empirical results demonstrate that the new method can effectively suppress the noise and artifacts in low-dose CT images,and maintain the image structure information and spatial resolution.

关 键 词：低剂量CT 图像重建 投影数据恢复 惩罚加权最小二乘 低维流形先验 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]