摘要
针对低剂量CT投影数据存在伪影和噪声的现象,提出了一种基于字典学习与三维块匹配滤波(Block-Matching and 3D Filtering,BM3D)的最大后验(Maximum A Posteriori,MAP)投影域降噪算法.该算法首先使用区别性字典对低剂量CT投影数据进行预处理,达到抑制部分噪声的目的;再运用MAP算法框架,构造一种由中值能量函数与BM3D算子组成的联合先验模型,对整个投影数据进行平滑处理.分别采用两种模型图像对该算法进行验证,实验结果表明,与滤波反投影(Filter Back Projection,FBP)算法、惩罚重加权最小二乘(Penalized Reweighted Least-Squares,PRWLS)算法和各向异性加权先验正弦图平滑算法相比,所提算法重建出的图像伪影较少,较好地保持了图像的边缘信息,具有较高的结构相似性和峰值信噪比.
Aiming at the phenomenon of artifacts and noise in low-dose CT projection data,this paper propose a maximum posterior projection domain denoising algorithm based on dictionary learning and block-matching 3D filtering.The algorithm firstly uses a discriminative dictionary to preprocess low-dose CT projection data to achieve the purpose of suppressing part of the noise.Then,the MAP algorithmic framework is used to construct a joint prior model consisting of the median energy function and the BM3D operator to smooth the whole projection data.The algorithm uses two types of model images for verification.The experimental results show that the artifacts in the images reconstructed by the algorithm are less so the edge of the images are well preserved,and have higher structural similarity index measurement and peak signal-to-noise ratio,compared with other three algorithms:Filter Back Projection,Penalized Reweighted Least-Squares,and the anisotropic weighted prior sinogram smoothing algorithm.
作者
杨一鸣
刘祎
方帆
YANG Yiming;LIU Yi;FANG Fan(Shanxi Provincial Key Laboratory for Biomedical Imaging and Big Data,North University of China,Taiyuan 030051,China;PLA Unit 63961,Beijing 100012,China)
出处
《测试技术学报》
2020年第1期16-21,27,共7页
Journal of Test and Measurement Technology
基金
国家自然科学基金资助项目（61671413）
国家重大科学仪器设备开发专项资助项目（2014YQ24044508）
关键词
低剂量CT
字典学习
三维块匹配滤波
最大后验
low-dose computed tomography
dictionary learning
block-matching and 3D filtering
maximum a posteriori
