基于非局部低秩稀疏矩阵分解的低剂量脑灌注CT统计迭代重建-《中国医学物理学杂志》

文章信息/Info

Title:: Statistical iterative reconstruction for low-dose cerebral perfusion CT using nonlocal low-rank and sparse matrix decomposition

作者:: 牛善洲1; 2; 李硕1; 2; 梁礼境1; 2; 谢国强1; 2; 刘汉明1; 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

Author(s):: NIU Shanzhou1; 2; LI Shuo1; 2; LIANG Lijing1; 2; XIE Guoqiang1; 2; LIU Hanming1; 1. School of Mathematics and Computer Science, Gannan Normal University, Ganzhou 341000, China 2. Ganzhou Key Laboratory of Computational Imaging, Gannan Normal University, Ganzhou 341000, China

Keywords:: Keywords: cerebral perfusion CT nonlocal low-rank and sparse matrix decomposition penalized weighted least-squares image reconstruction

摘要:: 脑灌注CT成像需要对患者头部进行连续反复扫描，相应的X射线辐射剂量较常规CT显著增加。为减少X射线辐射剂量，提出一种基于非局部低秩稀疏矩阵分解的低剂量脑灌注CT统计迭代重建方法。首先对脑灌注CT序列图像进行分块，然后构建非局部低秩稀疏矩阵分解模型，最后求解相应的目标函数重建出脑灌注CT序列图像。与滤波反投影算法和基于低秩稀疏矩阵分解的惩罚加权最小二乘方法相比，本文方法得到的脑血容参数图像的结构相似性指标分别提高38.07%和5.61%、特征相似性指标分别提高13.17%和2.47%；平均通过时间参数图像的结构相似性指标分别提高59.73%和0.28%、特征相似性指标分别提高20.26%和0.70%。本文方法能在去除低剂量脑灌注CT图像噪声和伪影的同时保持图像的边缘结构信息，并且获得更准确的脑血流动力学参数图像。

Abstract:: Abstract: The cerebral perfusion CT (CPCT) imaging requires continuous repetition scans, and the corresponding radiation dose is significantly increased compared with that of conventional CT. A low-dose CPCT statistical iterative reconstruction method using penalized weighted least-squares based on nonlocal low-rank and sparse matrix decomposition (PWLS-NLSMD) is present to reduce the radiation dose of CPCT imaging. After partitioning the sequence images of CPCT, a NLSMD model is developed, and the corresponding objective function is solved for reconstructing CPCT images. PWLS-NLSMD increases the structural similarity and feature similarity of the cerebral blood volume map and those of the mean transit time map by 38.07%, 13.17%, 59.73% and 20.26% as compared with filtered back-projection algorithm, and 5.61%, 2.47%, 0.28% and 0.70% as compared with penalized weighted least-squares based on low-rank and sparse matrix decomposition. PWLS-NLSMD can preserve the edge and structure information while effectively suppressing the noise, and obtain more accurate cerebral hemodynamic parameters.