《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (5): 1410-1414.DOI: 10.11772/j.issn.1001-9081.2024060856

• 第十届中国数据挖掘会议 • 上一篇    

基于知识蒸馏双分支结构的视网膜病变辅助诊断方法

牛四杰1,2(), 刘昱良1,2   

  1. 1.济南大学 信息科学与工程学院,济南 250022
    2.山东省网络环境智能计算技术重点实验室(济南大学),济南 250022
  • 收稿日期:2024-06-08 修回日期:2024-09-05 接受日期:2024-09-10 发布日期:2024-10-08 出版日期:2025-05-10
  • 通讯作者: 牛四杰
  • 作者简介:牛四杰(1984—),男,山东临沂人,教授,博士,CCF会员,主要研究方向:模式识别、医学影像分析
    刘昱良(1999—),女,吉林长春人,硕士,主要研究方向:医学图像分类。
  • 基金资助:
    国家自然科学基金资助项目(62471202);山东省高等学校人才引育创新团队发展计划项目(鲁教科函[2021]51号);山东省科技型中小企业创新能力提升工程项目(2022TSGC1048)

Auxiliary diagnostic method for retinopathy based on dual-branch structure with knowledge distillation

Sijie NIU1,2(), Yuliang LIU1,2   

  1. 1.School of Information Science and Engineering,University of Jinan,Jinan Shandong 250022,China
    2.Shandong Provincial Key Laboratory of Network-based Intelligent Computing (University of Jinan),Jinan Shandong 250022,China
  • Received:2024-06-08 Revised:2024-09-05 Accepted:2024-09-10 Online:2024-10-08 Published:2025-05-10
  • Contact: Sijie NIU
  • About author:NIU Sijie, born in 1984, Ph. D., professor. His research interests include pattern recognition, medical image analysis.
    LIU Yuliang, born in 1999, M. S. Her research interests include medical image classification.
  • Supported by:
    National Natural Science Foundation of China(62471202);Development Program of Youth Innovation Team of Institutions of Higher Learning in Shandong Province (Lujiaokehan [2021] No. 51);Shandong Province Small- and Medium-sized Science and Technology Enterprises Innovation Capability Building Engineering Project(2022TSGC1048)

摘要:

利用传统模型对糖尿病肾病(DN)高风险患者的视网膜疾病进行早期诊断时,由于糖尿病患者的视网膜图像数据少且类别不平衡,诊断精度不高。因此,提出一种基于知识蒸馏双分支结构的视网膜病变辅助诊断方法,以提高对少数类别的识别能力。该方法首先使用在大型医学数据集上训练的教师网络指导学生网络学习,将教师网络所学得的信息传递给学生网络,以提升学生网络的泛化能力,缓解数据少的问题。其次,在学生网络中提出一种双分支结构:分支一使用重平衡策略,引入Focal Loss函数,通过调节损失函数的权重使模型更关注难分样本;分支二利用类别注意力模块(CAM)学习每个类别的判别性特征,使模型在训练中不会偏向数据多的类别。这2个分支分别促进分类器学习和特征学习,可缓解类别不平衡。使用临床上收集的视网膜图像数据对所提方法进行评估,实验结果表明,所提方法在66例(89眼) DN高风险患者筛查任务上的准确率和特异度比病变感知注意力模型(LAM)分别提高了1.05和1.53个百分点。所提方法可以提高DN识别精度,实现视网膜疾病的辅助诊断。

关键词: 深度学习, 知识蒸馏, 类别不平衡, 视网膜病变图像分类

Abstract:

When using traditional models for the early diagnosis of retinopathy in high-risk patients with Diabetic Nephropathy (DN), the diagnostic accuracy is often compromised due to limited and category imbalanced retinal images of diabetic patients. To address this issue, an auxiliary diagnostic method for retinopathy based on dual-branch structure with knowledge distillation was proposed to improve the recognition capability for minority categories. Firstly, a teacher network pre-trained on large medical datasets was employed to guide the student network's learning process, transferring acquired knowledge to improve the student network's generalization ability and mitigate data scarcity. Secondly, a dual-branch structure was proposed in the student network. Branch 1 utilized a rebalancing strategy with Focal Loss function to emphasize challenging samples by adjusting loss function weights, while Branch 2 employed a Category Attention Module (CAM) to learn discriminative features for each category, preventing model bias towards majority categories. These two branches respectively promoted classifier learning and feature learning to alleviate category imbalance. Evaluated on clinically collected retinal image data, experimental results demonstrate that the proposed method achieves 1.05 and 1.53 percentage points improvements in accuracy and specificity respectively compared with Lesion-aware Attention Model (LAM) in screening tasks involving 66 cases (89 eyes) of high-risk patients with DN. The proposed method improves the recognition accuracy of DN and realizes the auxiliary diagnosis of retinal diseases.

Key words: deep learning, knowledge distillation, category imbalance, retinopathy image classification

中图分类号: