Journal of Computer Applications ›› 2020, Vol. 40 ›› Issue (5): 1460-1464.DOI: 10.11772/j.issn.1001-9081.2019101744

• Virtual reality and multimedia computing • Previous Articles     Next Articles

Mass and calcification classification method in mammogram based on multi-view transfer learning

XIAO He1, LIU Zhiqin1, WANG Qingfeng1, HUANG Jun1, ZHOU Ying2, LIU Qiyu2, XU Weiyun2   

  1. 1.College of Computer Science and Technology, Southwest University of Science and Technology, MianyangSichuan 621000, China
    2.Radiology Department, Mianyang Central Hospital, MianyangSichuan 621000, China
  • Received:2019-10-15 Revised:2019-12-12 Online:2020-05-10 Published:2020-05-15
  • Contact: LIU Zhiqin, born in 1962, M. S., professor. Her research interests include high performance computing, numerical simulation, artificial intelligence.
  • About author:XIAO He, born in 1994, M. S. candidate. His research interests include artificial intelligence, medical image analysis.LIU Zhiqin, born in 1962, M. S., professor. Her research interests include high performance computing, numerical simulation, artificial intelligence.WANG Qingfeng, born in 1988, Ph. D., lecturer. Her research interests include artificial intelligence, machine learning, medical image analysis.HUANG Jun, born in 1988, Ph. D. candidate, lecturer. His research interests include high performance computing, optimal control theory.ZHOU Ying, born in 1984, Ph. D., attending physician. Her research interests include medical image analysis, radiomics.LIU Qiyu, born in 1963, M. S., chief physician. His research interests include interventional radiology.XU Weiyun, born in 1963, Ph. D., chief physician. Her research interests include breast surgery.
  • Supported by:

    This work is partially supported by the Sichuan Provincial Open Fundation of Civil-Military Integration Research Institute (2017SCII0219, 2017SCII0220), the Sichuan Science and Technology Program (2019JDRC0119).

基于迁移学习的多视角乳腺肿块和钙化簇分类方法

肖禾1, 刘志勤1, 王庆凤1, 黄俊1, 周莹2, 刘启榆2, 徐卫云2   

  1. 1.西南科技大学 计算机科学与技术学院,四川绵阳 621010
    2.绵阳市中心医院 放射科,四川绵阳 621010
  • 通讯作者: 刘志勤(1962—)
  • 作者简介:肖禾(1994—),男,四川峨眉山人,硕士研究生,CCF会员,主要研究方向:人工智能、医学图像分析; 刘志勤(1962—),女,四川绵阳人,教授,硕士,CCF会员,主要研究方向:高性能计算、数值模拟、人工智能; 王庆凤(1988—),女,四川安岳人,讲师,博士,CCF会员,主要研究方向:人工智能、机器学习、医学图像分析; 黄俊(1988—),男,四川富顺人,讲师,博士研究生,CCF会员,主要研究方向:高性能计算、最优化控制理论; 周莹(1984—),女,四川绵阳人,主治医师,博士,主要研究方向:医学图像分析、影像组学方法; 刘启榆(1963—),男,四川广元人,主任医师,硕士,主要研究方向:放射介入医学; 徐卫云(1963—),女,四川绵阳人,主任医师,博士,主要研究方向:乳腺外科。
  • 基金资助:

    四川省军民融合研究院开放基金资助项目(2017SCII0219,2017SCII0220);四川省科技计划项目(2019JDRC0119)。

Abstract:

In order to solve the problem of insufficient available training data in the classification task of breast mass and calcification, a multi-view model based on secondary transfer learning was proposed combining with imaging characteristics of mammogram. Firstly, CBIS-DDSM (Curated Breast Imaging Subset of Digital Database for Screening Mammography) was used to construct the breast local tissue section dataset for the pre-training of the backbone network, and the domain adaptation learning of the backbone network was completed, so the backbone network had the essential ability of capturing pathological features. Then, the backbone network was secondarily transferred to the multi-view model and was fine-tuned based on the dataset of Mianyang Central Hospital. At the same time, the number of positive samples in the training was increased by CBIS-DDSM to improve the generalization ability of the network. The experimental results show that the domain adaption learning and data augmentation strategy improves the performance criteria by 17% averagely and achieves 94% and 90% AUC (Area Under Curve) values for mass and calcification respectively.

Key words: mammogram, Computer Aided Diagnosis (CAD), Convolutional Neural Network (CNN), transfer learning, domain adaptation, multi-view network

摘要:

针对乳腺肿块和钙化簇分类任务中可用训练数据量较少的问题,结合乳腺钼靶图成像特点提出了一种基于二次迁移学习的多视角模型。首先,使用CBIS-DDSM制作乳腺局部组织切片数据集来预训练主干网络,完成主干网络的领域适应性学习,使之具备基本的病理特征捕捉能力;随后,把主干网络二次迁移到多视角网络中,在绵阳市中心医院数据集上进行微调,同时利用CBIS-DDSM增加训练的正样本数量以提升网络的泛化能力。实验结果表明,领域适应性学习和数据扩充策略平均提升了17%性能指标,取得了94%和90%的肿块和钙化簇曲线下面积(AUC)值。

关键词: 乳腺钼靶图像, 计算机辅助诊断, 卷积神经网络, 迁移学习, 领域适应, 多视角网络

CLC Number: