[1]林洁,李彩霞,黎海霞,等.运用循环一致生成对抗网络对肥胖肺结核患者超低剂量CT图像进行降噪[J].中国医学物理学杂志,2025,42(12):1601-1608.[doi:DOI:10.3969/j.issn.1005-202X.2025.12.009]
　LIN Jie,LI Caixia,LI Haixia,et al.Denoising ultra-low-dose CT images of obese patients with pulmonary tuberculosis using cycle-consistent generative adversarial network[J].Chinese Journal of Medical Physics,2025,42(12):1601-1608.[doi:DOI:10.3969/j.issn.1005-202X.2025.12.009]
点击复制

运用循环一致生成对抗网络对肥胖肺结核患者超低剂量CT图像进行降噪()

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

卷:
42
期数:
2025年第12期
页码:
1601-1608
栏目:
医学影像物理
出版日期:
2025-12-29

文章信息/Info

Title:
Denoising ultra-low-dose CT images of obese patients with pulmonary tuberculosis using cycle-consistent generative adversarial network
文章编号:
1005-202X（2025）12-1601-08
作者:
林洁1; 李彩霞1; 黎海霞2; 严承功1; 许乙凯1; 乔文俊1
1.南方医科大学南方医院影像诊断科， 广东 广州 510515； 2.拜尔医药保健有限公司， 广东 广州 510000
Author(s):
LIN Jie1;  LI Caixia1;  LI Haixia2;  YAN Chenggong1;  XU Yikai1;  QIAO Wenjun1
1. Department of Radiology, Nanfang Hospital, Southern Medical University, Guangzhou 510515, China 2. Bayer Healthcare Company Limited, Guangzhou 510000, China
关键词:
计算机断层扫描; 生成对抗网络; 肥胖; 降噪; 辐射剂量
Keywords:
Keywords: computed tomography generative adversarial network obesity denoising radiation dose
分类号:
R318；R811.1
DOI:
DOI:10.3969/j.issn.1005-202X.2025.12.009
文献标志码:
A
摘要:
目的：基于循环一致生成对抗网络（Cycle-GAN）的深度学习方法，对肥胖肺结核患者胸部超低剂量CT图像进行去噪并对合成图像进行评估。方法：收集129例肺结核患者资料。连续使用常规剂量与超低剂量对相同范围进行CT扫描（常规剂量：120 kVp，自动曝光系统；超低剂量：80 kVp，10 mAs）。随机将70名患者的扫描图像用于训练Cycle-GAN网络中的超低剂量图像到常规剂量图像的转换，30名正常体质量患者与29名肥胖患者用于测试。将合成图像与全模型迭代重建算法（IMR）、混合迭代算法（iDose）、滤波反投影算法（FBP） 3种迭代算法图像从客观测量数据与主观图像质量评分进行比较。结果：对肺结核肥胖患者使用超低剂量CT扫描，平均有效辐射剂量为0.1 mSv，相较于常规剂量CT显著减少96%。优化后的Cycle-GAN去噪网络，峰值信噪比和结构相似性指数分别提高到（32.7±3.3） dB和0.80±0.05。Cycle-GAN去噪网络能够重建出低噪声图像，图像噪声为（41.4±5.2） HU，均低于iDose［（90.2±9.7） HU］和FBP［（188.6±31.6） HU］（P<0.001）。Cycle-GAN去噪网络重建图像较iDose、FBP与IMR算法，有更高的信噪比。结论：运用Cycle-GAN对肥胖肺结核患者超低剂量CT图像去噪，合成图像可用于影像医生评估患者肺结核病况。
Abstract:
Abstract: Objective To denoise chest ultra-low-dose computed tomography (CT) images of obese patients with pulmonary tuberculosis using a deep learning approach based on the cycle-consistent generative adversarial network (Cycle-GAN), and to evaluate the quality of the reconstructed images. Methods Clinical data of 129 patients with pulmonary tuberculosis were collected. Consecutive CT scans of the same anatomical range were conducted using a standard-dose protocol (120 kVp, automated exposure control system) and an ultra-low-dose protocol (80 kVp, 10 mAs). CT images from 70 randomly selected patients were used to train the Cycle-GAN network for converting ultra-low-dose images to standard-dose images and the images from 30 normal-weight patients and 29 obese patients were used for testing. The reconstructed images were compared with those reconstructed using iterative model reconstruction (IMR), hybrid iterative reconstruction (iDose), filtered back projection (FBP) in terms of objective measurement metrics and subjective image quality scores. Results For obese patients with pulmonary tuberculosis undergoing ultra-low-dose CT scanning, the average effective radiation dose was 0.1 mSv, showing a significant 96% reduction as compared with standard-dose CT scanning. The optimized Cycle-GAN denoising network achieved a peak signal-to-noise ratio of (32.7±3.3) dB and a structural similarity index of 0.80±0.05. The Cycle-GAN denoising network effectively reconstructed the low-noise images with an image noise of (41.4±5.2) HU, which was remarkably lower than that of iDose ［(90.2±9.7) HU］ and FBP ［(188.6±31.6) HU］ (P<0.001). Additionally, the images reconstructed by the Cycle-GAN denoising network exhibited a higher signal-to-noise ratio than those generated by iDose, FBP, and IMR algorithms. Conclusion Using Cycle-GAN to denoise ultra-low-dose CT images of obese patients with pulmonary tuberculosis can generate reconstructed images that are suitable for radiologists to assess the disease status of patients with pulmonary tuberculosis.

相似文献/References:

[1]郭广威,莫贤海,陈炯,等. 低剂量与标准剂量CT对慢性鼻窦炎手术高风险患者图像质量和辐射剂量的影响[J].中国医学物理学杂志,2017,34(11):1137.[doi:DOI:10.3969/j.issn.1005-202X.2017.11.011]
　GUO Guangwei,MO Xianhai,CHEN Jiong,et al. Evaluation of low-dose versus standard-dose CT examination in surgical high-risk patients with chronic rhinosinusitis[J].Chinese Journal of Medical Physics,2017,34(12):1137.[doi:DOI:10.3969/j.issn.1005-202X.2017.11.011]
[2]齐泽瑶,王远军. 基于全变分模型的CT不完全角度重建算法研究进展[J].中国医学物理学杂志,2019,36(2):180.[doi:1005-202X（2019）02-0180-05]
　QI Zeyao,WANG Yuanjun. Research progress on limited-angle CT reconstruction algorithm based on total variation model[J].Chinese Journal of Medical Physics,2019,36(12):180.[doi:1005-202X（2019）02-0180-05]
[3]张巧莹,黄晓宇,梁小红,等. PACS系统三维测量方法与多田公式计算方法在脑出血测量的准确性比较[J].中国医学物理学杂志,2019,36(3):296.[doi:DOI:10.3969/j.issn.1005-202X.2019.03.010]
　ZHANG Qiaoying,,et al. Comparison of accuracy between Picture Archiving and Communication Systems three-dimensional measuring method and Tada formula in intracerebral hemorrhage volume measurement[J].Chinese Journal of Medical Physics,2019,36(12):296.[doi:DOI:10.3969/j.issn.1005-202X.2019.03.010]
[4]于楠,沈聪,贾永军,等.肺密度三维定量分析对急性肺栓塞的诊断价值[J].中国医学物理学杂志,2020,37(2):180.[doi:DOI:10.3969/j.issn.1005-202X.2020.02.009]
　YU Nan,SHEN Cong,JIA Yongjun,et al.Diagnostic value of lung density evaluated by three-dimensional quantitative CT in acute pulmonary embolism[J].Chinese Journal of Medical Physics,2020,37(12):180.[doi:DOI:10.3969/j.issn.1005-202X.2020.02.009]
[5]张白霖,戴振晖,朱琳,等.不同设备不同扫描参数对影像组学特征可再现性影响分析[J].中国医学物理学杂志,2021,38(1):61.[doi:DOI:10.3969/j.issn.1005-202X.2021.01.011]
　ZHANG Bailin,DAI Zhenhui,ZHU Lin,et al.An analysis of influence of different equipment and different scanning parameters on reproducibility of radiomics features[J].Chinese Journal of Medical Physics,2021,38(12):61.[doi:DOI:10.3969/j.issn.1005-202X.2021.01.011]
[6]弥佳,周宇佳,冯前进.基于生成对抗网络的无监督动态对比增强磁共振分解方法[J].中国医学物理学杂志,2021,38(2):186.[doi:DOI:10.3969/j.issn.1005-202X.2021.02.011]
　MI Jia,ZHOU Yujia,FENG Qianjin.Unsupervised image decomposition of DCE-MRI based on generative adversarial networks[J].Chinese Journal of Medical Physics,2021,38(12):186.[doi:DOI:10.3969/j.issn.1005-202X.2021.02.011]
[7]郭翌,吴香奕,吴茜,等.基于循环一致生成对抗网络的多模态影像刚性配准[J].中国医学物理学杂志,2021,38(2):198.[doi:DOI:10.3969/j.issn.1005-202X.2021.02.013]
　GUO Yi,WU Xiangyi,WU Qian,et al.Rigid registration of multimodal images based on CycleGAN[J].Chinese Journal of Medical Physics,2021,38(12):198.[doi:DOI:10.3969/j.issn.1005-202X.2021.02.013]
[8]杭声琪,张文韬,王泰淇,等.基于CT和MRI膝关节有限元模型建立与不同载荷下生物力学特性分析[J].中国医学物理学杂志,2021,38(3):370.[doi:DOI:10.3969/j.issn.1005-202X.2021.03.019]
　HANG Shengqi,ZHANG Wentao,WANG Taiqi,et al.Establishment of finite element model of knee joint based on CT and MRI and analysis of biomechanical characteristics under different loads[J].Chinese Journal of Medical Physics,2021,38(12):370.[doi:DOI:10.3969/j.issn.1005-202X.2021.03.019]
[9]张白霖,黄泽裕,戴振晖,等.直肠充盈对直肠壁CT影像组学特征的影响[J].中国医学物理学杂志,2022,39(2):194.[doi:DOI:10.3969/j.issn.1005-202X.2022.02.011]
　ZHANG Bailin,Huang Zeyu,DAI Zhenhui,et al.Effects of rectal filling on CT radiomics features of rectal wall[J].Chinese Journal of Medical Physics,2022,39(12):194.[doi:DOI:10.3969/j.issn.1005-202X.2022.02.011]
[10]李焕根,黄鸣恋,黄幼佩,等.CT评估急性重症胰腺炎及肝脏损害的应用价值[J].中国医学物理学杂志,2022,39(3):311.[doi:DOI:10.3969/j.issn.1005-202X.2022.03.009]
　LI Huangen,HUANG Minglian,HUANG Youpei,et al.Role of CT in assessing severe acute pancreatitis and liver damage[J].Chinese Journal of Medical Physics,2022,39(12):311.[doi:DOI:10.3969/j.issn.1005-202X.2022.03.009]

备注/Memo

备注/Memo:
【收稿日期】2025-09-15 【基金项目】国家自然科学基金（82271939） 【作者简介】林洁，硕士，医疗器械工程师（主管技师），研究方向：CT图像质量控制与人工智能，E-mail: 305430180@qq.com 【通信作者】乔文俊，研究方向：医学影像技术及图像质控，E-mail: 1257826529@qq.com
更新日期/Last Update: 2025-12-29