基于全变分展开的低剂量CT重建网络  

A Total Variation Prior Unfoldment Based Low Dose CT Reconstruction Network

作　　者：张鹏程[1,2] 侯瑞峰 刘祎[1,2] 桂志国[1] ZHANG Pengcheng;HOU Ruifeng;LIU Yi;GUI Zhiguo(State Key Laboratory of Dynamic Testing Technology,North University of China,Taiyuan 030051,China;School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学省部共建动态测试技术国家重点实验室,太原030051 [2]中北大学信息与通信工程学院,太原030051

出　　处：《电子科技大学学报》2024年第4期617-628,共12页Journal of University of Electronic Science and Technology of China

基　　金：山西省基础研究计划基金(20210302124403,202303021211148);山西省回国留学人员科研资助项目(2021-111)。

摘　　要：针对CT迭代展开重建网络仅对数据保真项进行神经网络展开降低了重建网络计算性能的问题,通过对基于全变分的CT迭代重建算法进行神经网络展开,提出一种对数据保真项和全变分正则项全部进行神经网络展开的重建网络,从而改善了CT重建图像的视觉质量。首先,采用原始–对偶算法求解基于全变分的CT重建问题,得到易于神经网络展开的迭代重建算法。然后,对该迭代重建算法进行神经网络展开,尤其是对正则项部分的算法进行神经网络展开,得到迭代展开CT重建网络。在模拟的低剂量CT数据集上验证了该算法的有效性。实验结果表明,与6种低剂量CT重建算法相比,该算法在抑制低剂量CT图像噪声的同时,很好地保留了图像中的结构和细节纹理。重建图像的定量评价分析显示,该算法取得了良好的峰值信噪比和归一化均方误差指标值,验证了提出的低剂量CT重建算法具有较好的噪声抑制能力和较强的鲁棒性。Without unrolling the prior terms,most unrolling approaches for Computed Tomography(CT)reconstruction primarily unroll the fidelity term of iterative reconstruction methods to neural networks,which may reduce the computational efficiency of the reconstruction network.To overcome this drawback,a new CT reconstruction network is formed by unrolling a CT iterative reconstruction algorithm based on total variation,especially for the unfoldment of the total variation prior.The unfoldment of the prior improves the visual quality of CT reconstructed images.Firstly,the primal-dual algorithm is utilized to solve the CT reconstruction problem based on the total variation prior,to obtain an iterative reconstruction algorithm which can be easily unrolled to the neural network.Then,the unrolling approach for CT reconstruction is obtained by unrolling this iterative reconstruction algorithm.The effectiveness of the proposed algorithm is tested on a simulated low dose CT dataset.The experimental results show that,compared with six kinds of low-dose CT reconstruction algorithms,the new algorithm effectively preserves the structure and texture details of the image while removing noise in low-dose CT images.The quantitative analysis shows that the proposed algorithm scored the highest PSNR and the lowest NMSE,which indicates that the proposed algorithm is good at noise suppression in the low-dose CT reconstruction.

关 键 词：CT重建 全变分 神经网络 迭代展开 原始-对偶算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]