基于多尺度低秩的心脏磁共振图像的高质量重构算法  

High quality reconstruction algorithm for cardiac magnetic resonance images based on multiscale low rank modeling

作　　者：衡阳 陈峰 徐剑锋[3] 汤敏[1,4,5] HENG Yang;CHEN Feng;XU Jianfeng;TANG Min(Department of Information Engineering, School of Information Science and Technology, Nantong University, Nantong, Jiangsu 226007,P.R. China;Department of Automation, School of Electrical Engineering, Nantong University, Nantong, Jiangsu 226007, P.R.China;Department of Medical Imaging, Affiliated Hospital of Nantong University, Nantong, Jiangsu 226007, P.R.China;Tongke School of Microelectronics, Nantong, Jiangsu 226007, P.R.China;Nantong University-Nantong Joint Research Center for Intelligent Information Technology, Nantong, Jiangsu 226007t P.R.China)

机构地区：[1]南通大学信息科学技术学院信息工程系,江苏南通226007 [2]南通大学电气工程学院自动化系,江苏南通226007 [3]南通大学附属医院医学影像科,江苏南通226007 [4]通科微电子学院,江苏南通226007 [5]南通大学-南通智能信息技术联合研究中心,江苏南通226007

出　　处：《生物医学工程学杂志》2019年第4期573-580,共8页Journal of Biomedical Engineering

基　　金：国家自然科学基金项目(81371663);江苏省自然科学基金项目(BK20151273);江苏高校品牌专业建设工程资助项目(PPZY2015B135)

摘　　要：借助信号内在的稀疏性或可压缩性,压缩感知利用随机投影实现以远低于奈奎斯特频率的采样频率下对压缩后数据的采集。结合压缩感知和低秩思想,可以加快心脏磁共振(CMR)图像的扫描速度,减轻患者不适,提高检查质量。本文提出CMR图像的多尺度低秩分解模型,并采用交替方向拉格朗日乘子法(ADMM)进行求解。以峰值信噪比(PSNR)和相对误差(RLNE)作为定量评价指标,结合人眼视觉感受以及局部区域放大,对比分析本文算法与L+S分解、kt FOCUSS、k-t SPARSE SENSE等主流算法的性能优劣。实验结果表明:本文提出的多尺度低秩分解模型,经过ADMM算法重构的效果在性能指标上明显优于其他对比算法,同时图像细节和边缘轮廓成像质量更佳。该方法将推动CMR快速成像技术的发展及其在临床疾病诊疗中的应用。Taking advantages of the sparsity or compressibility inherent in real world signals, compressed sensing (CS) can collect compressed data at the sampling rate much lower than that needed in Shannon’s theorem. The combination of CS and low rank modeling is used to medical imaging techniques to increase the scanning speed of cardiac magnetic resonance (CMR), alleviate the patients* suffering and improve the images quality. The alternating direction method of multipliers (ADMM) algorithm is proposed for multiscale low rank matrix decomposition of CMR images. The algorithm performance is evaluated quantitatively by the peak signal to noise ratio (PSNR) and relative l2 norm error (RLNE), with the human visual system and the local region magnification as the qualitative comparison. Compared to L + S, kt FOCUSS, k-t SPARSE SENSE algorithms, experimental results demonstrate that the proposed algorithm can achieve the best performance indices, and maintain the most detail features and edge contours. The proposed algorithm can encourage the development of fast imaging techniques, and improve the diagnoses values of CMR in clinical applications.

关 键 词：心脏磁共振 多尺度低秩模型 快速成像 压缩感知 

分 类 号：R445.2[医药卫生—影像医学与核医学]