摘要
随着计算机断层扫描(Computed Tomography,CT)技术的发展与广泛应用,其重建算法得到了长足的发展。近年来,基于数据驱动的深度学习技术与CT图像重建结合的趋势愈加明显,深度学习应用于影像重建在实验环境中取得了显著的重建效果。本文以CT成像原理、重建技术发展以及存在问题为切入点。首先,围绕深度学习的有监督和无监督学习策略,将两类基于深度学习的重建算法原理与目前国内外研究进展进行结合介绍。主要包括有监督学习对传统解析迭代方法重建过程的网络实现与优化,以及无监督学习中自监督学习和生成模型类重建原理方法。其次,介绍了一些常用的基于像素、区域结构细节、语义特征等层面的损失函数,并对它们的优缺点进行了分析。最后,本文对深度重建领域仍存在的一些问题进行了总结,并介绍了一些新的学习型重建策略。
With the development and wide application of computed tomography technology,the reconstruction algorithm has made great progress.In recent years,the trend of combining data-driven deep learning technology with CT image reconstruction has become more and more obvious,and the reconstruction quality of deep learning applied to image reconstruction in experimental environment is remarkable.Based on the principle of CT imaging,the development of reconstruction technology and existing problems,this paper focuses on the supervised and unsupervised learning strategies of deep learning,and introduces the principles of two kinds of deep reconstruction algorithms.The current research progress at home and abroad is covered.Firstly,it introduces the network realization and optimization of supervised learning in the reconstruction process of traditional analytic and iteration methods,and the principles and methods of self-supervised learning,and generation model reconstruction in unsupervised learning.Secondly,some commonly used loss functions based on pixel-wise metrics,regional structure details and semantic features are introduced.Their advantages and disadvantages are analyzed.Finally,this paper summarizes some problems existing in practical applications of deep learning in image reconstruction,and introduces some new learning strategies for reconstruction.
作者
吴凡
刘进
张意
陈阳
陆志凯
WU Fan;LIU Jin;ZHANG Yi;CHEN Yang;LU Zhikai(School of Computer and Information,Anhui Polytechnic University,Wuhu 24100,China;School of Computer Science and Engineering,Southeast University,Nanjing 210096,China;School of Cyber Science and Engineering,Sichuan University,Chengdu 610207,China;No.906 Hospital,Central Military Commission Joint,Ningbo 315211,China)
出处
《中国体视学与图像分析》
2022年第4期387-404,共18页
Chinese Journal of Stereology and Image Analysis
基金
国家自然科学基金(61801003)
安徽省高等学校科学研究项目(2022AH050968)
高校优秀青年人才支持计划重点项目(gxyqZD2019052)
关键词
CT成像
深度重建
卷积神经网络
有监督学习
无监督学习
CT imaging
deep reconstruction
convolutional neural network
supervised learning
unsupervised learning
