工业X射线CT中基于深度学习的射束硬化伪影抑制方法（英文）  

Deep Learning Based Beam Hardening Artifact Reduction in Industrial X-ray CT

作　　者：周丽平 孙怡[1] 程凯 余建桥 ZHOU Li-ping, SUN Yi, CHENG Kai, YU Jian-qiao(School of Information and Communication Engineering, Dalian University of Technology, Dalian 116024, Chin)

机构地区：[1]大连理工大学信息与通信工程学院,辽宁大连116024

出　　处：《CT理论与应用研究（中英文）》2018年第2期227-240,共14页Computerized Tomography Theory and Applications

基　　金：National Key scientific Instrument and Equipment Development Projects(2014YQ240445)

摘　　要：利用工业CT进行无损检测时,由于实际X射线源的宽能谱特性,目前现有的大部分重建算法得到的图像含有射束硬化伪影。射束硬化伪影降低了图像的质量,影响了CT图像应用,如CT图像诊断等。本文提出一种基于深度学习的减少硬化伪影的方法,用大量含有硬化伪影的断层图像作为输入,用相应的在固定能量下重建的不含硬化伪影的图像作为输出来训练卷积神经网络。通过建立含有硬化伪影的断层图像与不含硬化伪影的断层图像之间的映射关系,来抑制硬化伪影。实验结果证明了本文所提方法在降低CT图像硬化伪影上的有效性。In the nondestructive detection with industrial CT, due to the fact that the actual X-ray source has a wide spectrum, slices reconstructed by most existing reconstruction algorithms will suffer from beam hardening artifacts. It will degrade image quality greatly, affecting important CT image task such as CT diagnosis and so on. In this study, we propose a method to suppress beam hardening artifacts based on deep learning. We train a convolutional neural network using a large number of images with beam hardening artifacts as input and the corresponding artifact-free images reconstructed at a fixed energy as output to establish the mapping between image with beam hardening artifacts and artifact-free image for suppressing beam hardening artifacts. Experimental results show the effectiveness of the proposed method in the beam hardening artifact reduction of CT images.

关 键 词：射束硬化 计算机断层成像 深度学习 卷积神经网络 

分 类 号：O242[理学—计算数学]