基于生成对抗网络的CT图像金属伪影校正  

Metal artifact correction of industrial CT images based on generative adversarial networks

作　　者：姜世博 孙跃文[1] 许硕 吴志芳[1] JIANG Shibo;SUN Yuewen;XU Shuo;WU Zhifang(Institute of Nuclear Energy and New Energy Technology,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学核能与新能源技术研究院,北京100084

出　　处：《哈尔滨工程大学学报》2022年第12期1766-1771,共6页Journal of Harbin Engineering University

摘　　要：为了能在更好地保留工业CT图像特征细节的情况下进行金属伪影校正,本文提出一种基于生成对抗网络的去金属伪影校正方法。本文通过解析仿真的方法生成不同金属材质不同类型工件的CT图像作为训练集,训练得到含金属伪影与无金属伪影图像的映射关系。经过仿真和实际扫描的CT数据集验证,与多项式拟合方法对比,本文提出的方法在去除金属伪影的同时,还较好地保留了CT图像尺寸、边界等特征信息。若将经过本方法处理的CT图像用于后续的CT图像分割中,将显著提升CT图像分割质量。To effectively perform metal artifact reduction while preserving industrial CT image features,we propose a metal artifact reduction method based on generative adversarial networks.The proposed method generates CT images of different workpieces made of different metal materials through analytical simulations,obtaining the mapping relationship between the image containing metal artifacts and that free of metal artifacts through training.The simulation and dataset of actual CT images verify that compared with the polynomial fitting method,the method proposed in this paper can correct metal artifacts and retain feature information,such as CT image size and boundary.The quality of CT image segmentation will be significantly improved using the proposed method for subsequent CT image segmentation.

关 键 词：工业CT 金属伪影 伪影校正 神经网络 生成对抗网络 CT仿真 数据集 重投影 

分 类 号：TL99[核科学技术—核技术及应用]