摘要
一般情况下,计算机断层扫描(computed tomography,CT)重建后的图像与真实物体之间会存在一些差异,这种差异很大程度体现在重建图像上的伪影。受到金属植入物的影响,CT图像中出现了不同程度的金属伪影,因此近40年出现了大量的金属伪影校正(metal artifacts reduction,MAR)方法对CT图像中的金属伪影进行去除。本文首先回顾了产生金属伪影的基本原因,并介绍了CT图像的传统的MAR方法和目前取得较大进展的基于深度学习的MAR方法的发展趋势;接着文中详细介绍了几种基于卷积神经网络的MAR方法;最后对本文进行了总结并对金属伪影校正方法的前景进行了展望。
In general,reconstructed images of computed tomography(CT)differs from ground truth of objects.The main differences is artifacts on reconstructed images.Caused by metal implants,metal artifacts appear often in CT images.Therefore,a large number of metal artifact reduction(MAR)methodshave been published in last forty years.This paper firstly reviews the basic causes of metal artifacts,and introduces the development of MAR methods for CT images,including the traditional MAR methods and the successful MAR methods based on deep learning.Then,several MAR methods based on convolutional neural networks are introduced in detail.Finally,we summarizes with the prospects of MAR methods.
作者
肖文
曾理
XIAO Wen;ZENG Li(College of Mathematics and Statistics,Chongqing University,Chongqing 401331,China;ICT Research Center,Chongqing University,Chongqing University,Chongqing 400044,China)
出处
《中国体视学与图像分析》
2019年第1期29-36,共8页
Chinese Journal of Stereology and Image Analysis
基金
国家自然科学基金面上项目资助(编号61771003)
关键词
计算机断层扫描
金属伪影校正
卷积神经网络
深度学习
computed tomography
metal artifact reduction
convolutional neural network
deep learning
