计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (5): 213-221.DOI: 10.3778/j.issn.1002-8331.2111-0081

• 图形图像处理 • 上一篇    下一篇

改进DenseNet的乳腺癌病理图像八分类研究

赵晓平,王荣发,孙中波,魏旭全   

  1. 1.南京信息工程大学 计算机与软件学院,南京 210044
    2.南京信息工程大学 数字取证教育部工程研究中心,南京 210044
  • 出版日期:2023-03-01 发布日期:2023-03-01

Research on Eight Classifications of Breast Cancer Pathological Images Based on Improved DenseNet

ZHAO Xiaoping, WANG Rongfa, SUN Zhongbo, WEI Xuquan   

  1. 1.School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing 210044, China
    2.Engineering Research Center of Digital Forensics, Ministry of Education, Nanjing University of lnformation Science and Technology, Nanjing 210044, China
  • Online:2023-03-01 Published:2023-03-01

摘要: 目前,在医学图像领域存在乳腺癌组织病理图像自动分类难以应用于临床诊断的现象,究其根源是当前没有大型公开的数据集或数据集数据不均衡。针对上述问题,提出一种结合密集卷积神经网络(dense convolutional network,DenseNet)、注意力机制(attention mecheanism)和焦点损失函数(Focal loss)的乳腺癌组织病理图像的多分类模型,即DAFLNet。DAFLNet在乳腺癌组织病理图像数据集BreaKHis上进行训练、验证与测试,最终实验结果显示,该模型对良恶性二分类的识别准确率达到99.1%,对乳腺亚型八分类的识别准确率达到95.5%。证明在数据不均衡的条件下,DAFLNet模型能够准确地对乳腺组织病理图像进行八分类。

关键词: 乳腺癌病理图像, DenseNet, 八分类, 注意力机制, Focal loss

Abstract: At present, in the field of medical image, the automatic classification of breast cancer histopathology is difficult to be applied to clinical diagnosis. The root cause is that there is no large-scale open dataset or data set is unbalanced. In view of the above problems, this paper proposes a multi classification model of breast cancer histopathological images combined with dense convolutional network(DenseNet), attention mechanism and Focal loss, that is DAFLNet. DAFLNet is trained, validated and tested on the breast cancer histopathological image dataset BreaKHis. The final experimental results show that the recognition accuracy of the model for benign and malignant secondary classification is 99.1%, and the recognition accuracy of eight classification of breast subtypes is 95.5%. It is proved that DAFLNet model can accurately classify breast histopathological images under the condition of unbalanced data.

Key words: breast cancer pathological image, dense convolutional network(DenseNet), eight categories, attention mecha-nism, Focal loss