数据与计算发展前沿 ›› 2021, Vol. 3 ›› Issue (6): 35-49.

doi: 10.11871/jfdc.10-1649.2021.06.003

• 专刊:科学大数据挖掘与知识发现 • 上一篇    下一篇

面向临床需求的CT图像降噪综述

蒲晓蓉1,*(),黄佳欣1(),刘军池1(),孙家瑜2(),罗纪翔1(),赵越1(),陈柯成1(),任亚洲1()   

  1. 1. 电子科技大学,计算机科学与工程学院,大数据研究中心,四川 成都 611731
    2. 四川大学华西医院,四川 成都 610044
  • 收稿日期:2021-11-15 出版日期:2021-12-20 发布日期:2022-01-26
  • 通讯作者: 蒲晓蓉
  • 作者简介:蒲晓蓉, 电子科技大学计算机科学与工程学院,博士,教授,主要研究方向包括人工神经网络、机器学习、计算机视觉、医学图像处理、计算机辅助诊断 (CAD)、智能健康医疗。
    负责制定论文框架,撰写“引言”与“结论与展望”,论文修改、审定。
    PU Xiaorong, Ph.D., is currently a professor with the School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China. Her current research interests include neural networks, computer vision, computer aided diagnosis (CAD), and e-Health.
    In this paper, she is responsible for drawing up the paper framework, writing “Introduction” and “Conclusion and Outlook”, and paper revision and approval. E-mail: puxiaor@uestc.edu.cn;|黄佳欣,电子科技大学计算机科学与工程学院,硕士研究生,研究方向为计算机视觉、医学图像处理。
    负责论文初稿撰写,并研究有关从LDCT图像中学习噪声的降噪方法。
    HUANG Jiaxin is pursuing the M.Sc. degree in computer science and technology with the School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China, under the supervision of Prof. X. Pu. Her research interests include computer vision and medical image processing.
    In this paper, she is responsible for the paper drafting and research on noise reduction methods for learning noise from LDCT images. E-mail: jiaxinhuang011403@gmail.com;|刘军池,电子科技大学计算机科学与工程学院,硕士研究生,研究方向为医学图像分析。
    参与撰写“2 面向临床需求的LDCT图像降噪方法”。
    LIU Junchi is pursuing the M.Sc. degree in computer science and technology with the School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China, under the supervision of Prof. X. Pu. His research interests include medical image processing.
    In this paper, he is responsible for writing “2 LDCT image denoising method clinical-oriented LDCT image denoising”. E-mail: jchiliu@163.com;|孙家瑜,四川大学华西临床医学院(华西医院),教授,2021年获四川大学医学技术博士学位,主要研究方向包括医学成像技术和医学图像深度挖掘。
    参与撰写“3 医学图像质量评价”部分。
    SUN Jiayu is currently a professor of School of clinical medicine of West China (West China Hos-pital) of Sichuan University. He received the MD.D. degree in medical technology, from Sichuan University, Chengdu, China in 2021. His current research interests include medical imaging technology and deep mining of medical images.
    In this paper, he is responsible for writing “3 Medical image quality evaluation”. E-mail: sjy080512@163.com;|罗纪翔,电子科技大学计算机科学与工程学院,硕士研究生,主要研究方向为计算机视觉、医学图像处理。
    参与撰写“1.1 传统低剂量CT图像降噪研究方法”部分。
    LUO Jixiang is pursuing the M.Sc. degree in computer science and technology with the School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China, under the supervision of Prof. X. Pu. His main research interests include computer vision and medical image processing.
    In this paper, he is responsible for writing “1.1 Traditional low-dose CT image denoising research method”. E-mail: galaxyluo@outlook.com;|赵越,电子科技大学计算机科学与工程学院,硕士研究生,主要研究方向为计算机视觉和医学图像处理。
    参与撰写“1.2 基于深度学习的LDCT图像降噪方法”部分。
    ZHAO Yue is pursuing the M.Sc. degree in computer science and technology with the School of Com-puter Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China, under the supervision of Prof. X. Pu. His main research interests include computer vision and medical image processing.
    In this paper, he is responsible for writing “1.2 Denoising me-thod of LDCT image based on deep learning”. E-mail: kennyZ96@hotmail.com;|陈柯成,电子科技大学计算机科学与工程学院,硕士研究生,主要研究方向为机器学习、医学图像处理。
    参与撰写“2 面向临床需求的LDCT图像降噪方法”,并研究基于协同训练的LDCT图像降噪。
    CHEN Kecheng is pursuing the M.Sc. degree in computer science and technology with the School of Computer Science and Engineering, University of Electronic Science and Tech-nology of China, Chengdu, China, under the supervision of Prof. X. Pu. His research interests include machine learning, medical image processing and big data.
    In this paper, he is responsible for writing “2 LDCT image denoising method clinical-oriented LDCT image denoising”, and research on denoising of LDCT image based on co-training. E-mail: cs.ckc96@gmail.com;|任亚洲,电子科技大学计算机科学与工程学院,副教授,分别于2009年和2014年在华南理工大学获信息与计算科学学士学位以及计算机应用技术博士学位,2012年至2014年访问美国乔治梅森大学的数据挖掘实验室。共发表50多篇同行评议的学术论文,目前的研究方向包括机器学习和医学数据分析。
    参与制定论文框架与论文修改、审定。
    REN Yazhou is currently an associate professor with the School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China. He recei-ved the B.Sc. degree in information and computation science and the Ph.D. degree in computer science from the South China University of Technology, Guangzhou, China, in 2009 and 2014, respectively. He visited the Data Mining Laboratory, George Mason University, USA, from 2012 to 2014. He has published more than 50 peer-reviewed research articles. His current research interests include machine learning and medical data analysis.
    In this paper, he is responsible for drawing up the paper frame-work and paper revision and approval. E-mail: yazhou.ren@uestc.edu.cn
  • 基金资助:
    四川省科技计划项目重点项目(2020YFS0119);四川省科技计划项目重点项目(2021YFS0172);国家自然科学基金项目(61806043);2021年成都市技术创新研发项目(2021-YF05-01583-SN);四川大学华西医院学科卓越发展1·3·5工程临床研究孵化项目(2021HXFH004)

A Survey on Clinical Oriented CT Image Denoising

PU Xiaorong1,*(),HUANG Jiaxin1(),LIU Junchi1(),SUN Jiayu2(),LUO Jixiang1(),ZHAO Yue1(),CHEN Kecheng1(),REN Yazhou1()   

  1. 1. School of Computer Science and Engineering, Big Data Research Center, University of Electronic Science and Technology of China, Chengdu, Sichuan 611731, China
    2. West China Hospital, Sichuan University, Chengdu, Sichuan 610044, China
  • Received:2021-11-15 Online:2021-12-20 Published:2022-01-26
  • Contact: PU Xiaorong

摘要:

【目的】低剂量CT扫描技术已广泛用于肺结节和肺癌早期筛查等临床诊断。然而,由辐射剂量降低所导致的成像噪声制约着诊断精度的进一步提高。【方法】本文综合研究了低剂量CT图像降噪技术的发展脉络,从基于迭代优化等传统方法出发,分析了当前基于机器学习技术的低剂量CT图像降噪等CT图像降噪方法。【结果】现有基于机器学习的CT图像降噪方法,一方面采用人为假设噪声分布并构造人工数据集进行模型训练和测试,忽略了临床应用中噪声特点和强度的多样性问题;另一方面,由于深度神经网络的“黑盒”特性,导致现有基于深度神经网络的低剂量CT图像降噪模型的可解释性不足。【结论】CT图像降噪应面向临床实际需求,充分考虑噪声形成的机理和图像噪声的真实分布,结合病灶检测等高阶任务和临床医生的阅片行为,探索面向临床需求的低剂量CT图像降噪新范式。

关键词: CT图像, 深度学习, 图像降噪, 图像质量评价

Abstract:

[Objective] Low-dose CT scanning technology has been widely used in clinical diagnoses such as early screening of lung nodules and lung cancer. However, the imaging noise caused by the reduction of radiation dose has restricted the further improvement of diagnostic accuracy.[Methods] This article comprehensively studies the development of low-dose CT image noise reduction technology. Starting with the traditional methods such as iterative optimization, it analyzes the current methods of low-dose CT image noise reduction based on machine learning. [Results] The existing methods based on machine learning, on the one hand, use artificially assumed noise distribution and construct artificial data sets to train and test the model, ignoring the diverse characteristics and intensity of noise in clinical applications; on the other hand, due to the "black box" characteristics of deep neural networks, the existing low-dose CT image noise reduction models based on deep neural networks are insufficient in interpretability. [Conclusions] CT image noise reduction should be oriented to the requirements of clinical scenarios, fully consider the mechanism of noise formation and the true distribution of image noise, and combine high-level tasks such as lesion detection and the reading behavior of clinicians to explore the brand-new paradigm of low-dose CT image noise reduction for clinical demands.

Key words: CT image, deep learning, image denoising, image quality evaluation