基于噪声鲁棒学习的胸部X射线尘肺鉴别方法

doi:10.19678/j.issn.1000-3428.0068379

计算机工程 ›› 2024, Vol. 50 ›› Issue (11): 350-359. doi: 10.19678/j.issn.1000-3428.0068379

基于噪声鲁棒学习的胸部X射线尘肺鉴别方法

崔锦莹¹, 梁立河¹, 任雪婷¹, 强彦¹^,², 赵涓涓³^,⁴, 孔晓梅⁵^,*(), 尉骁⁵, 张华⁶

1. 太原理工大学计算机科学与技术学院(大数据学院), 山西太原 030000
2. 中北大学软件学院, 山西太原 030000
3. 太原理工大学软件学院, 山西太原 030000
4. 晋中信息学院大数据学院·信息工程学院, 山西晋中 030600
5. 国家卫生健康委尘肺病重点实验室/呼吸疾病山西省重点实验室/山西医科大学第一医院呼吸与危重症医学科, 山西太原 030000
6. 山西医科大学第一医院放射科, 山西太原 030000

收稿日期:2023-09-12 出版日期:2024-11-15 发布日期:2024-02-22
通讯作者: 孔晓梅
基金资助:
国家自然科学基金重点项目(U21A20469); 中央级公益性科研院所基本科研业务费专项资金(N2020-PT320-005); 国家卫生健康委尘肺病重点实验室开放课题(YKFKT004)

Method of Noise Robust Learning Based Chest X-ray Discrimination of Pneumoconiosis

CUI Jinying¹, LIANG Lihe¹, REN Xueting¹, QIANG Yan¹^,², ZHAO Juanjuan³^,⁴, KONG Xiaomei⁵^,*(), YU Xiao⁵, ZHANG Hua⁶

1. College of Computer Science and Technology (College of Data Science), Taiyuan University of Technology, Taiyuan 030000, Shanxi, China
2. School of Software, North University of China, Taiyuan 030000, Shanxi, China
3. School of Software, Taiyuan University of Technology, Taiyuan 030000, Shanxi, China
4. School of Data Science and Information Engineering, Jinzhong College of Information, Jinzhong 030600, Shanxi, China
5. NHC Key Laboratory of Pneumoconiosis/Shanxi Key Laboratory of Respiratory Diseases/Department of Respiratory and Critical Care Medicine, The First Hospital of Shanxi Medical University, Taiyuan 030000, Shanxi, China
6. Department of Radiology, The First Hospital of Shanxi Medical University, Taiyuan 030000, Shanxi, China

Received:2023-09-12 Online:2024-11-15 Published:2024-02-22
Contact: KONG Xiaomei

摘要/Abstract

摘要：

在医学图像噪声标注数据的训练中, 目前常用的方法是根据训练损失对噪声标签数据集进行划分, 以过滤掉噪声标签样本。然而, 这种方法面临两个需要解决的问题, 即如何在筛选出噪声样本的同时尽可能地保留与其损失分布相似的困难样本, 以及如何提高样本利用率, 挖掘隐藏在噪声样本中的有用信息以减轻模型过拟合的问题。为了解决上述问题, 提出一种由样本分布引导的噪声鲁棒学习策略(SGRL), 包括样本划分与半监督对比分类。为了更可靠地区分信息量大的困难样本与有害噪声样本, 介绍一种噪声滤波器样本选择方法。此外, 提出了一种增强匹配对比网络, 使用所有样本进行训练, 从而得到一个具有噪声鲁棒性的分类模型。在此基础上, 利用对比学习作为补充, 进一步对抗对噪声标签的记忆, 提高筛查准确率。实验结果表明, 该方法在5%、10%、20%和40%噪声比的尘肺胸片数据集上均取得了显著的性能提升。与现有的先进方法相比, 该方法的筛查准确率分别平均提升了5.88、7.05、7.59和6.19个百分点, 验证了改进方法的有效性。

关键词: 噪声标签, 尘肺筛查, 困难样本感知, 弱监督学习, 医学图像分类

Abstract:

In training medical image noise annotation data, the prevailing approach involves partitioning the noise-labeled dataset based on training loss to filter out the noise-labeled samples. However, this method faces two pressing issues that require resolution: first, filtering out noise samples while retaining difficult samples with similar loss distributions as much as possible, and second, enhancing sample utilization and uncovering valuable information embedded in noise samples to alleviate model overfitting. This study proposes a Sample Distribution Guided Noise Robust Learning strategy (SGRL) comprising sample partitioning and semi-supervised contrastive classification to address these challenges. A straightforward yet effective sample selection method called a noise filter method is introduced to distinguish informative, difficult samples from detrimental noise samples more accurately. Additionally, an enhanced matching contrastive network is proposed to train using all samples, yielding a noise-robust classification model. Contrastive learning is utilized as a supplement to counter the memorization of noise labels and improve screening accuracy. The experimental results demonstrate significant performance improvement of the proposed method across dust-induced pneumoconiosis chest X-ray datasets with noise ratios of 5%, 10%, 20%, and 40%. Compared with existing state-of-the-art methods, the screening accuracy of this method increased by an average of 5.88, 7.05, 7.59, and 6.19 percentage points, validating the effectiveness of the proposed improvement method.

Key words: noise labels, pneumoconiosis screening, hard sample aware, weakly supervised learning, medical image classification

崔锦莹, 梁立河, 任雪婷, 强彦, 赵涓涓, 孔晓梅, 尉骁, 张华. 基于噪声鲁棒学习的胸部X射线尘肺鉴别方法[J]. 计算机工程, 2024, 50(11): 350-359.

CUI Jinying, LIANG Lihe, REN Xueting, QIANG Yan, ZHAO Juanjuan, KONG Xiaomei, YU Xiao, ZHANG Hua. Method of Noise Robust Learning Based Chest X-ray Discrimination of Pneumoconiosis[J]. Computer Engineering, 2024, 50(11): 350-359.

https://www.ecice06.com/CN/Y2024/V50/I11/350

图/表 11

图1 SGRL总体框架

Fig.1 Overall framework of SGRL

图2 20%噪声比尘肺数据集中样本平均预测概率分布直方图

Fig.2 Mean prediction probability histogram of the samples in 20% noise ratio pneumoconiosis dataset

图3 NF模块的结构

Fig.3 Structure of NF module

图4 数据增强示意图

Fig.4 Schematic diagram of data enhancement

图5 不同超参数设置下20%噪声比尘肺胸片数据集上的ACC

Fig.5 ACC on the 20% noise ratio pneumoconiosis chest radiograph dataset at different hyperparameters settings

图6 模型在不同损失权重设置下的ACC

Fig.6 ACC of the model with different loss weight settings

图7 不同方法在不同噪声比的尘肺胸片数据集上的准确率

Fig.7 Accuracy of different methods on pneumoconiosis chest radiograph dataset with different noise ratios

图8 不同消融模型设置下的特征热力图

Fig.8 Characteristic heat map for different ablation model settings

参考文献 27

1	ZHANG C M, HE J, SHANG L. An X-ray image classification method with fine-grained features for explainable diagnosis of pneumoconiosis. Personal and Ubiquitous Computing, 2024, 28(2): 403- 415. doi: 10.1007/s00779-023-01730-3
2	WANG Y, CUI F T, DING X P, et al. Automated identification of the preclinical stage of coal workers' pneumoconiosis from digital chest radiography using three-stage cascaded deep learning model. Biomedical Signal Processing and Control, 2023, 83, 104607. doi: 10.1016/j.bspc.2023.104607
3	王峥, 钱青俊, 张建芳, 等. 计算机辅助诊断在尘肺病诊断中应用价值. 中国职业医学, 2020, 47(4): 428- 431. URL
	WANG Z, QIAN Q J, ZHANG J F, et al. Application value of computer-aided diagnosis in diagnosing pneumoconiosis. China Occupational Medicine, 2020, 47(4): 428- 431. URL
4	ARPIT D, JASTRZEBSKI S, BALLAS N, et al. A closer look at memorization in deep networks[C]//Proceedings of International Conference on Machine Learning. Washington D. C., USA: IEEE Press, 2017: 233-242.
5	SONG H, KIM M, PARK D, et al. Learning from noisy labels with deep neural networks: a survey. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(11): 8135- 8153. doi: 10.1109/TNNLS.2022.3152527
6	HAN J F, LUO P, WANG X G. Deep self-learning from noisy labels[C]//Proceedings of IEEE/CVF International Conference on Computer Vision. Washington D. C., USA: IEEE Press, 2019: 5138-5147.
7	GOLDBERGER J, BEN-REUVEN E. Training deep neural-networks using a noise adaptation layer[C]//Proceedings of IEEE International Conference on Learning Representations. Washington D. C., USA: IEEE Press, 2016: 368-367.
8	YAO Y, LIU T, HAN B, et al. Dual T: reducing estimation error for transition matrix in label-noise learning[C]//Proceedings of NIPS'20. Cambridge, USA: MIT Press, 2020: 7260-7271.
9	王学刚, 王玉峰. 基于多轮修正噪声标签的神经网络分类框架. 计算机技术与发展, 2023, 33(8): 151- 158. URL
	WANG X G, WANG Y F. A neural network classification framework based on calibrating noisy labels in multi-round. Computer Technology and Development, 2023, 33(8): 151- 158. URL
10	YI L, LIU S, SHE Q, et al. On learning contrastive representations for learning with noisy labels[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2022: 16682-16691.
11	郭亚庆. 基于噪声特性分析的正则化鲁棒回归建模方法[D]. 太原: 山西大学, 2023.
	GUO Y Q. Regularized robust regression modeling method based on noise characteristic analysis[D]. Taiyuan: Shanxi University, 2023. (in Chinese)
12	NGUYEN T, MUMMADI C, NGO T, et al. SELF: learning to filter noisy labels with self-ensembling[C]// Proceedings of IEEE International Conference on Learning Representations. Washington D. C., USA: IEEE Press, 2020: 467-478.
13	HAN B, YAO Q, YU X, et al. Robust training of deep neural networks with extremely noisy labels[C]// Proceedings of the 34th International Conference on Neural Information Processing Systems. Washington D. C., USA: IEEE Press, 2020: 458-466.
14	YU X, HAN B, YAO J, et al. How does disagreement help generalization against label corruption?[C]//Proceedings of IEEE International Conference on Machine Learning. Washington D. C., USA: IEEE Press, 2019: 7164-7173.
15	暴恒, 邓理睿, 张良, 等. 基于检索增强的噪声标签细粒度图像分类方法. 北京航空航天大学学报, 2024, 50(7): 2284- 2292. doi: 10.13700/j.bh.1001-5965.2022.0589
	BAO H, DENG L R, ZHANG L, et al. Retrieval-based augmentation for refined image classification with noisy labels. Journal of Beijing University of Aeronautics and Astronautics, 2024, 50(7): 2284- 2292. doi: 10.13700/j.bh.1001-5965.2022.0589
16	TANAKA D, IKAMI D, YAMASAKI T, et al. Joint optimization framework for learning with noisy labels[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE Press, 2018: 5552-5560.
17	YI K, WU J X. Probabilistic end-to-end noise correction for learning with noisy labels[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE Press, 2019: 7017-7025.
18	LI J, SOCHER R, HOI S C H. DivideMix: learning with noisy labels as semi-supervised Learning[C]// Proceedings of International Conference on Learning Representations. Washington D. C., USA: IEEE Press, 2019: 268-277.
19	ISCEN A, VALMADRE J, ARNAB A, et al. Learning with neighbor consistency for noisy labels[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE Press, 2022: 4672-4681.
20	MALLEM S, HASNAT A, NAKIB A. Efficient meta label correction based on meta learning and bi-level optimization. Engineering Applications of Artificial Intelligence, 2023, 117, 105517. doi: 10.1016/j.engappai.2022.105517
21	HOU C Q, YANG C H, REN F J, et al. A noise robust batch mode semi-supervised and active learning framework for image classification. Berlin, Germany: Springer, 2019: 541- 552.
22	古楠楠. 针对数据标签噪声的自步半监督降维. 计算机工程, 2023, 49(11): 131- 142. doi: 10.19678/j.issn.1000-3428.0067397
	GU N N. Self-paced semi-supervised dimensionality reduction for data with noisy labels. Computer Engineering, 2023, 49(11): 131- 142. doi: 10.19678/j.issn.1000-3428.0067397
23	CUBUK E D, ZOPH B, SHLENS J, et al. Randaugment: practical automated data augmentation with a reduced search space[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE Press, 2020: 702-703.
24	ZHANG C Y, BENGIO S, HARDT M, et al. Understanding deep learning (still) requires rethinking generalization. Communications of the ACM, 2021, 64(3): 107- 115. doi: 10.1145/3446776
25	SOHN K, BERTHELOT D, CARLINI N, et al. Fixmatch: simplifying semi-supervised learning with consistency and confidence[C]//Proceedings of NIPS'20. Cambridge, USA: MIT Press, 2020: 596-608.
26	CHEN T, KORNBLITH S, NOROUZI M, et al. A simple framework for contrastive learning of visual representations[C]//Proceedings of International Conference on Machine Learning. Washington D. C., USA: IEEE Press, 2020: 1597-1607.
27	KINGMA D P, BA J L. Adam: a method for stochastic optimization[EB/OL]. [2023-08-10]. https://arxiv.org/pdf/1412.6980.

[1]	周炫余, 吴莲华, 郑勤华, 肖天星, 王紫璇, 张思敏. 联合语义提示和记忆增强的弱监督跳绳视频异常检测方法[J]. 计算机工程, 2024, 50(7): 87-95.
[2]	张慧妍, 梁勇, 兰景宏, 赵强. 基于记忆模块与过滤式生成对抗网络的入侵检测方法[J]. 计算机工程, 2024, 50(6): 197-207.
[3]	邵良杉, 赵松泽. 基于多模型融合的不完整数据分数插补算法[J]. 计算机工程, 2023, 49(9): 79-88, 98.
[4]	张驰名, 王庆凤, 刘志勤, 黄俊, 陈波, 付婕, 周莹. 基于深度学习的胸部常见病变诊断方法[J]. 计算机工程, 2020, 46(7): 306-311,320.
[5]	殷佳豪, 刘世杰, 鲍宇, 杨轩, 朱紫维. 基于一维卷积神经网络的实时心脏按压评估[J]. 计算机工程, 2020, 46(5): 298-304,311.
[6]	景庄伟, 管海燕, 彭代峰, 于永涛. 基于深度神经网络的图像语义分割研究综述[J]. 计算机工程, 2020, 46(10): 1-17.
[7]	张驰名, 王庆凤, 刘志勤, 黄俊, 周莹, 刘启榆, 徐卫云. 基于深度迁移学习的肺结节辅助诊断方法[J]. 计算机工程, 2020, 46(1): 271-278.

选择文件类型/文献管理软件名称

选择包含的内容

基于噪声鲁棒学习的胸部X射线尘肺鉴别方法

Method of Noise Robust Learning Based Chest X-ray Discrimination of Pneumoconiosis

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 27

相关文章 7

编辑推荐

Metrics

本文评价

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

基于噪声鲁棒学习的胸部X射线尘肺鉴别方法

Method of Noise Robust Learning Based Chest X-ray Discrimination of Pneumoconiosis

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 27

相关文章 7

编辑推荐

Metrics

本文评价