《计算机应用》唯一官方网站 ›› 2024, Vol. 44 ›› Issue (4): 1285-1293.DOI: 10.11772/j.issn.1001-9081.2023040517

• 多媒体计算与计算机仿真 • 上一篇    

基于多任务学习的间质性肺病分割算法

李威1, 陈玲2, 徐修远1, 朱敏2, 郭际香1, 周凯1, 牛颢1, 张煜宸3, 易珊烨3, 章毅1, 罗凤鸣2()   

  1. 1.四川大学 计算机学院,成都 610065
    2.四川大学 华西医院,成都 610044
    3.四川大学 华西临床医学院,成都 610041
  • 收稿日期:2023-05-04 修回日期:2023-07-13 接受日期:2023-07-14 发布日期:2023-08-03 出版日期:2024-04-10
  • 通讯作者: 罗凤鸣
  • 作者简介:李威(1994—),男,四川绵竹人,硕士研究生,CCF会员,主要研究方向:医学图像分割
    陈玲(1989—),女,江西抚州人,助理研究员,博士,主要研究方向:间质性肺病
    徐修远(1992—),男,安徽合肥人,副研究员,博士,主要研究方向:神经网络、智能医学
    朱敏(1989—),女,四川成都人,助理研究员,博士,主要研究方向:间质性肺病
    郭际香(1981—),女,山西吕梁人,副教授,博士,CCF会员,主要研究方向:神经网络、智能医学
    周凯(1995—),男,四川眉山人,博士研究生,主要研究方向:神经网络、智能医学
    牛颢(1983—),男,四川成都人,博士研究生,主要研究方向:神经网络、智能医学
    张煜宸(1996—),女,浙江台州人,硕士研究生,主要研究方向:间质性肺病
    易珊烨(1998—),女,四川乐山人,硕士研究生,主要研究方向:全科医学
    章毅(1963—),男,四川成都人,教授,博士,主要研究方向:人工智能、神经网络
    罗凤鸣(1970—),男,四川成都人,教授,博士,主要研究方向:间质性肺病、高原呼吸慢病、呼吸介入技术。fengmingluo@outlook.com
  • 基金资助:
    国家自然科学基金资助项目(62106163);中国人工智能学会-华为MindSpore学术奖励基金资助项目(21H1235)

Interstitial lung disease segmentation algorithm based on multi-task learning

Wei LI1, Ling CHEN2, Xiuyuan XU1, Min ZHU2, Jixiang GUO1, Kai ZHOU1, Hao NIU1, Yuchen ZHANG3, Shanye YI3, Yi ZHANG1, Fengming LUO2()   

  1. 1.College of Computer Science,Sichuan University,Chengdu Sichuan 610065,China
    2.West China Hospital,Sichuan University,Chengdu Sichuan 610044,China
    3.West China School of Medicine,Sichuan University,Chengdu Sichuan 610041,China
  • Received:2023-05-04 Revised:2023-07-13 Accepted:2023-07-14 Online:2023-08-03 Published:2024-04-10
  • Contact: Fengming LUO
  • About author:LI Wei, born in 1994, M. S. candidate. His research interests include medical image segmentation.
    CHEN Ling, born in 1989, Ph. D., assistant research fellow. Her research interests include interstitial lung disease.
    XU Xiuyuan, born in 1992, Ph. D., associate research fellow. His research interests include neural network, intelligent medicine.
    ZHU Min, born in 1989, Ph. D., assistant research fellow. Her research interests include interstitial lung disease.
    GUO Jixiang, born in 1981, Ph. D., associate professor. Her research interests include neural network, intelligent medicine.
    ZHOU Kai, born in 1995, Ph. D. candidate. His research interests include neural network, intelligent medicine.
    NIU Hao, born in 1983, Ph. D. candidate. His research interests include neural network, intelligent medicine.
    ZHANG Yuchen, born in 1996, M. S. candidate. Her research interests include interstitial lung disease.
    YI Shanye, born in 1998, M. S. candidate. Her research interests include general practice.
    ZHANG Yi, born in 1963, Ph. D., professor. His research interests include artificial intelligence, neural network.
    LUO Fengming, born in 1970, Ph. D., professor. His research interests include interstitial lung disease, plateau respiratory chronic disease, respiratory intervention technology.
  • Supported by:
    National Natural Science Foundation of China(62106163);CAAI-Huawei MindSpore Open Fund(21H1235)

摘要:

间质性肺病(ILD)的分割标签标注成本极高,且现有数据集通常存在样本量较少的问题,导致训练的模型效果较差。针对该问题,提出一种基于多任务学习的ILD分割算法。首先,基于U-Net构建多任务分割模型;其次,使用生成的肺部分割标签作为辅助任务标签进行多任务学习;最后,使用一种自适应调整多任务损失函数权重的方法,平衡主任务和辅助任务的损失。在自构建的ILD数据集上的实验结果表明,多任务分割模型的Dice相似系数(DSC)达到了82.61%,与U-Net相比提升了2.26个百分点。验证了所提算法可以提升ILD的分割性能,协助临床医生进行ILD诊断。

关键词: 间质性肺病, 语义分割, 小样本量, 多任务学习, 自适应多任务损失函数

Abstract:

Interstitial Lung Disease (ILD) segmentation labels are highly costly, leading to small sample sizes in existing datasets and resulting in poor performance of trained models. To address this issue, a segmentation algorithm for ILD based on multi-task learning was proposed. Firstly, a multi-task segmentation model was constructed based on U-Net. Then, the generated lung segmentation labels were used as auxiliary task labels for multi-task learning. Finally, a method of dynamically weighting the multi-task loss functions was used to balance the losses of the primary task and the secondary task. Experimental results on a self-built ILD dataset show that the Dice Similarity Coefficient (DSC) of the multi-task segmentation model reaches 82.61%, which is 2.26 percentage points higher than that of U-Net. The experimental results demonstrate that the proposed algorithm can improve the segmentation performance of ILD and can assist clinical doctors in ILD diagnosis.

Key words: Interstitial Lung Disease (ILD), semantic segmentation, small sample size, multi-task learning, adaptive multi-task loss function

中图分类号: