摘要
能谱CT可以利用不同X射线能量下材料衰减特性的差异,区分不同的材料成分。基于投影的材料分解是一种常用的材料分解方法,分为投影域分解和基材料图像重建两个步骤。针对该方法在分解过程中容易受到噪声污染的问题,提出一种基于双正则化的两步分解方法,将四方向总变分(TV4)正则化先验同时引入到材料分解和基图像重建中。TV4在传统总变分(TV)基础上扩展至四个方向梯度,能够更全面地捕捉材料图像中的多方向边缘,联合优化抑制噪声,在低剂量或高噪声数据中更具鲁棒性。实验采用仿真模体与临床前真实小鼠的多个能量通道投影数据进行实验,验证提出算法的有效性。在投影分解步骤,对比分析提出的TV4算法与LS算法、SR-TF算法在分解上的去噪结果。为进一步验证所提方法在材料分解精度上的性能,对比分析不同步骤上的正则化方法与本文提出的双正则化策略获得的基材料图像,并采用均方根误差(RMSE)和峰值信噪比(PSNR)指标对分解结果定量分析。实验结果表明,本文提出的算法能够清晰地分解出不同基材料的图像,对应PSNR值是所有方法中最高的且具有最小的RMSE值,说明本算法有效抑制了分解中噪声和伪影的干扰,提高了基材料图像的质量。
Spectral computed tomography(CT)can distinguish different material compositions by utilizing the differences in material attenuation characteristics under various X-ray energies.Projection-based material decomposition is a commonly used method,which consists of two steps:projection-domain decomposition and basis-material image reconstruction.To address the susceptibility of this method to noise contamination during decomposition,this study proposes a double-regularized two-step decomposition framework that simultaneously incorporates four-directional total variation(TV4)regularization priors into both material decomposition and basis image reconstruction.Extending conventional total variation(TV)to four-directional gradients,TV4 demonstrates enhanced capability in comprehensively capturing multi-directional edges within material images while achieving joint optimization for noise suppression,thereby exhibiting superior robustness in low-dose or high-noise scenarios.Experimental validation was conducted using multi-energy channel projection data from both simulated phantoms and preclinical in vivo mice.In the projection decomposition phase,the proposed TV4 algorithm was compared with conventional LS and SR-TF algorithms in terms of denoising performance.To further evaluate the material decomposition accuracy,basis material images obtained through different regularization strategies were quantitatively compared using root mean square error(RMSE)and peak signal-to-noise ratio(PSNR)metrics.Results demonstrate that the proposed algorithm achieves a clear separation of basis material images,attaining the highest PSNR values and the lowest RMSE values among all compared methods.These findings confirm the method's effectiveness in suppressing noise and artifact interference during decomposition while significantly enhancing basis material image quality.
作者
于鑫丽
孔慧华
张然
YU Xin-li;KONG Hui-hua;ZHANG Ran(School of Mathematics,North University of China,Taiyuan 030051,China;National Key Laboratory of Photoelectric Dynamic Testing Technology and Instrument for Extreme Environments,Taiyuan 030051,China)
出处
《光谱学与光谱分析》
北大核心
2025年第10期2935-2941,共7页
Spectroscopy and Spectral Analysis
基金
国家重点研发计划项目(2023YFE0205800)
国家自然科学基金联合基金项目(U23A20285)
山西省重点研发项目(202302150401011)
山西省基础研究计划项目(202403021223006)
山西省科技成果转化引导专项(202304021301028)
山西省留学基金项目(2023-129)资助。
关键词
能谱CT
双TV4正则化
材料分解
投影分解
基图像重建
Spectral CT
Double-TV4 regularization
Material decomposition
Projection decomposition
Base image reconstruction
作者简介
于鑫丽,女,2000年生,中北大学数学学院硕士研究生,e-mail:1678230834@qq.com;通讯作者:孔慧华,E-mail:huihuak@163.com。
