基于噪声估计与引导的低剂量CT图像盲去噪方法  

Low-dose CT image blind denoising method based on noise estimation and guidance

作　　者：张怀天 李金宝[1,2] ZHANG Huai-Tian;LI Jin-Bao(College of computer science and technology,Heilongjiang University, Harbin 150080,China;Shandong Artificial Intelligence Institute, Qilu University of Technology (Shandong Academy of Sciences), Ji'nan 250014, China)

机构地区：[1]黑龙江大学计算机科学技术学院,哈尔滨150080 [2]齐鲁工业大学(山东省科学院)山东省人工智能研究院,济南250014

出　　处：《黑龙江大学工程学报》2021年第3期131-140,共10页Journal of Engineering of Heilongjiang University

基　　金：国家重点研发计划项目(2020YFB1710200);黑龙江省自然科学基金重点项目(ZD2019F003)。

摘　　要：提出了一种基于噪声特征估计与引导的低剂量CT盲去噪方法。首先,采用有监督特征学习的方式对输入图像的辐射剂量进行等级评估,并估计出图像潜在的噪声特征图。其次,提出了一种基于噪声引导的低剂量CT图像盲去噪模型,通过显性噪声特征引导的方式将噪声特征与原始图像进行融合,并采用残差编码-解码卷积神经网络实现CT图像噪声去除。实验结果表明,在真实数据集上噪声估计网络及特征融合网络能够大幅提升去噪网络的性能,并且在未知剂量CT图像去噪任务上取得了较好的去噪效果。A low-dose CT blind denoising method based on noise feature estimation and guidance is proposed.Firstly,the supervised feature learning method is used to evaluate the radiation dose of the input image and estimate the potential noise feature map of the image.Secondly,a low-dose CT image blind denoising model based on noise guidance is proposed.The noise features are fused with the original image by means of explicit noise feature guidance,and a residual encoder-decoder convolutional neural network is used to remove noise from CT images.Experimental results show that the noise estimation network and feature fusion network on real data sets can greatly improve the performance of the denoising network,and achieve better denoising effects in the task of denoising CT images with unknown doses.

关 键 词：图像盲去噪 深度学习 噪声估计 感知损失 未知剂量CT 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]