吉林大学学报(信息科学版) ›› 2022, Vol. 40 ›› Issue (2): 213-218.

• • 上一篇    下一篇

基于 DenseNet 和迁移学习的乳腺癌图像识别

杨雨航1 , 刘 铭1 , 王新民1 , 肖志成1 , 蒋 扬2   

  1. 1. 长春工业大学 数学与统计学院, 长春 130012; 2. 中电文思海辉技术有限公司 汽车制造数字化事业部, 辽宁 大连 116000
  • 收稿日期:2021-09-06 出版日期:2022-06-11 发布日期:2022-06-11
  • 通讯作者: 刘铭(1979— ), 男, 吉林白山人, 长春工业大学教授,硕士生导师, 主要从事机器学习、 大数据分析与数据挖掘研究, (Tel)86-15843108878(E-mail)jlcclm@ 163. com.
  • 作者简介:杨雨航(1998— ), 女, 吉林公主岭人, 长春工业大学硕士研究生, 主要从事机器学习、 大数据分析与数据挖掘研究, (Tel)86-17843359498(E-mail)913081026@ qq. com.
  • 基金资助:
    吉林省自然科学基金资助项目(2020021157JC); 吉林省教育厅科学技术基金资助项目(JJKH20191295KJ)

Breast Cancer Image Recognition Based on DenseNet and Transfer Learning

YANG Yuhang 1 , LIU Ming 1 , WANG Xinmin 1 , XIAO Zhicheng 1 , JIANG Yang 2   

  1. 1. College of Mathematics and Statistics, Changchun University of Technology, Changchun 130012, China; 2. Department of Vehicle Manufacturing Digital, Pactera Technology International Limited Company, Dalian 116000, China
  • Received:2021-09-06 Online:2022-06-11 Published:2022-06-11

摘要: 目前病理组织学分析是诊断乳腺癌最广泛的方法, 而将众多的病理图像诊断分析分类, 需要病理学专家 通过显微镜对组织学样本进行目测识别, 采取人为方法识别特征费时费力。 为此, 针对乳腺癌病理图像数据集BreaKHis的不同放大倍数分为良性、恶性样本的二分类进行识别, 提出了基于DenseNet网络和迁移学习的 乳腺癌图像识别方法, 并将该方法过程与VGGNet、ResNet、DenseNet等深度学习模型进行了对比实验。实验 结果表明, 该模型学习能力强, 识别效果最好, 在此数据集上准确率可达到98%以上。

关键词: 乳腺癌图像;  , DenseNet 网络;  , 迁移学习;  , 深度学习

Abstract: Breast cancer is the most common malignant disease in women endangering women's life and health. Histopathological analysis is the most extensive method for the diagnosis of breast cancer. The diagnosis and classification of numerous pathological images require pathological experts to visually identify histological samples by microscopy, which takes time and effort to identify features. According to the different magnification of BreaKHis data set of breast cancer pathological images, it can be divided into benign and malignant samples for recognition. A breast cancer image recognition method based on DenseNet network and transfer learning is proposed to compare with VGGNet, ResNet, DenseNet and other deep learning models. The proposed model has strong learning ability and the best recognition effect, and the accuracy of the model can reach more than 98% on this dataset.

Key words: breast cancer images; , DenseNet network; , transfer learning; , deep Learning

中图分类号: 

  • TP183