采用细胞形态特征对比学习的肺癌图像分类

doi:10.3778/j.issn.1002-8331.2205-0418

摘要/Abstract

摘要： 针对肺癌图像分类中出现的已标记肺癌病理图像较少且细胞形态复杂的问题，提出一种基于细胞形态特征对比学习的肺癌病理图像分类方法，通过对比学习将置信度较高的未标记数据混入到训练数据中，解决已标记数据不足的问题。在最近邻对比学习的基础上，提出基于最远和最邻近的对比学习，将最远和最近邻图像同时用于对比学习，通过增加对比样本的正样本学习难度和数据的多样性，提升对比学习的性能。将基于可变形卷积和动态卷积的ResNet50作为编码器，增强对细胞形态特征的提取能力。实验结果表明，在已标记数据较少的情况下，与现有的分类方法相比，该分类方法能够充分利用已标记和未标记癌症病理图像中的细胞特征信息，获得较好的分类效果。

关键词: 肺癌, 病理图像分类, 对比学习, 可变形卷积, 动态卷积

Abstract: Aiming at the problems that the labeled pathological images of lung cancer are few and the cell morphology is complex in lung cancer image classification, a lung cancer pathological image classification method based on cell morphological feature contrastive learning is proposed. Through contrastive learning, the unlabeled data with high confidence is mixed into the training data to solve the problem of insufficient labeled data. Based on the nearest-neighbor contrastive learning, a farthest and nearest-neighbors-based contrastive learning is proposed. The farthest and nearest neighbor images are used for the contrastive learning at the same time. The performance of the contrastive learning is improved by increasing the learning difficulty of the positive samples and the diversity of the data. Deformable convolution and dynamic convolution-based ResNet50 are used as an encoder to enhance the extraction ability of cell morphological features. The experimental results show that, in the case of less labeled data, compared with the existing classification methods, the proposed method can make full use of the cell feature information in the labeled and unlabeled cancer pathology images and obtain better classification results.

Key words: lung cancer, pathological image classification, contrastive learning, deformable convolution, dynamic convolution

贾伟, 江海峰, 赵雪芬. 采用细胞形态特征对比学习的肺癌图像分类[J]. 计算机工程与应用, 2023, 59(19): 140-150.

JIA Wei, JIANG Haifeng, ZHAO Xuefen. Lung Cancer Image Classification Using Cell Morphological Feature Contrastive Learning[J]. Computer Engineering and Applications, 2023, 59(19): 140-150.

参考文献

[1] TRAVIS W D.Lung cancer pathology：current concepts[J].Clinics in Chest Medicine，2020，41（1）：67-85.
[2] TRAVIS W D，BRAMBILLA E，NOGUCHI M，et al.The new IASLC/ATS/ERS international multidisciplinary lung adenocarcinoma classification[J].Journal of Thoracic Oncology，2011，5（2）：244-285.
[3] DING R，ZHU D，HE P，et al.Comorbidity in lung cancer patients and its association with medical service cost and treatment choice in China[J].BMC Cancer，2020，20（1）：1-11.
[4] GROSE D，MORRISON D S，DEVEREUX G，et al.Comorbidities in lung cancer：prevalence，severity and links with socioeconomic status and treatment[J].Postgraduate Medical Journal，2014，90：305-310.
[5] MONIRUL K M，JIANG X，ANGGONDOWATI T，et al.Comorbidity and survival in lung cancer patients[J].Cancer Epidemiol Biomarkers Prevention，2015，24（7）：1001-1148.
[6] ZHENG M.Classification and pathology of lung cancer[J].Surgical Oncology Clinics of North America，2016，25（3）：447-468.
[7] TRAVIS W D，BRAMBILLA E，NICHOLSON A G，et al.The 2015 world health organization classification of lung tumors：impact of genetic，clinical and radiologic advances since the 2004 classification[J].Journal of Thoracic Oncology，2015，10（9）：1243-1260.
[8] WANG S，YANG D M，RONG R，et al.Artificial intelligence in lung cancer pathology image analysis[J].Cancer，2019，11：1-16.
[9] GOU Y，SONG Q，JIANG M，et al.Histological subtypes classification of lung cancers on CT images using 3D deep learning and radiomics[J].Academic Radiology，2020，28（9）：258-266.
[10] IBTEHAZ N，RAHMAN M S.MultiResUNet：rethinking the U-Net architecture for multimodal biomedical image segmentation[J].Neural Networks，2020，121：74-87.
[11] ALI M，ALI R.Multi-input dual-stream capsule network for improved lung and colon cancer classification[J].Diagnostics，2021，11（8）：1-18.
[12] VAMBOL A，KHARCHENKO V，POTII O，et al.Dynamic routing between capsules[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems，Long Beach，California，USA，December 4-9，2017：3859-3869.
[13] LI M，MA X，CHEN C，et al.Research on the auxiliary classification and diagnosis of lung cancer subtypes based on histopathological images[J].IEEE Access，2021，9：53687-53707.
[14] URBANOWICZ R J，MEEKER M，LACAVA W，et al.Relief-based feature selection：introduction and review[J].Journal of Biomedical Informatics，2018，85：189-203.
[15] HE K，FAN H，WU Y，et al.Momentum contrast for unsupervised visual representation learning[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition，Seattle，USA，June 13-19，2020.New York：IEEE Press，2020：9726-9735.
[16] CHEN T，KORNBLITH S，NOROUZI M，et al.A simple framework for contrastive learning of visual representations[C]//Proceedings of the 37th International Conference on Machine Learning，Los Angeles，USA，July 14-16，2020.New York：ACM，2020：1-11.
[17] DWIBEDI D，AYTAR Y，TOMPSON J，et al.With a little help from my friends：nearest-neighbor contrastive learning of visual representations[C]//Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision，Montreal，Canada，October 10-17，2021.New York：IEEE Press，2021：9568-9597.
[18] DAI J，QI H，XIONG Y，et al.Deformable convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，Venice，Italy，October 22-29，2017.New York：IEEE Press，2017：764-773.
[19] ZHANG J，WANG Q，ZHANG Y，et al.DyNet：dynamic convolution for accelerating convolutional neural networks[J].arXiv：2004.10694，2000.
[20] CLARK K，VENDT B，SMITH K，et al.The cancer imaging archive（TCIA）：maintaining and operating a public information repository[J].Journal of Digital Imaging，2013，26：1045-1057.
[21] BORKOWSKI A A，BUI M M，THOMAS L B，et al.Lung and colon cancer histopathological image dataset（LC25000）[J].arXiv：1912.12142，2019.