引用本文: | 段聪颖,顾敏杰,李雪,陈思光.基于联邦半监督学习的皮肤病变智能识别[J].生物信息学,2025,23(1):61-70. |
| DUAN Congying,GU Minjie,LI Xue,CHEN Siguang.Federated semi-supervised learning-based intelligent recognition for skin lesion[J].Chinese Journal of Bioinformatics,2025,23(1):61-70. |
|
摘要: |
近年来,深度学习技术已广泛应用于皮肤病变识别。然而在实际应用场景中,单一医疗机构存在训练数据有限、有标签样本不足,以及集中式学习易泄露隐私等问题。针对上述问题,本文提出了一种基于联邦半监督学习的皮肤病变智能识别机制。具体地,设计了一个基于联邦学习的云边协同皮肤病变智能识别模型,该模型在保护用户隐私的前提下,协同训练各个医疗机构数据,可为用户提供准确便捷的诊断服务。接着,设计了一种面向数据异构的半监督损失函数,以有效控制局部模型与全局模型之间的差异。此外,通过融合多重随机采样与准确率加权方法,明确各个本地模型的贡献,并将各个不均匀的本地模型聚合成一个全局共识模型,以进一步降低数据异构的影响。最后,实验结果表明,相较于近期提出的几种机制,所提机制在性能和可扩展性方面表现更优。 |
关键词: 皮肤病变 深度学习 图像识别 联邦学习 |
DOI:10.12113/202306005 |
分类号:TP391 |
文献标识码:A |
基金项目:国家自然科学基金(No.61971235); 中国博士后科学基金项目(No.2018M630590); 江苏省“333高层次人才培养工程”、江苏省博士后科研资助计划(No.2021K501C); 南京市妇幼保健院青年人才和南京邮电大学‘1311’人才计划资助. |
|
Federated semi-supervised learning-based intelligent recognition for skin lesion |
DUAN Congying1, GU Minjie1, LI Xue2, CHEN Siguang1
|
(1. School of Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing 210003, China; 2. Department of Dermatology, Womens Hospital of Nanjing Medical University (Nanjing Maternity and Child Health Care Hospital), Nanjing 210004, China)
|
Abstract: |
In recent years, deep learning technology has been widely applied in skin lesion recognition. However, in practical applications, single medical institution faces problems such as limited training data, insufficient labeled samples, and the risk of privacy leakage in centralized learning. To address the above problems, a federated semi-supervised learning-based intelligent recognition mechanism for skin lesion is proposed. Specifically, a federated learning-based cloud-edge collaborative intelligent recognition model for skin lesion is designed, which collaboratively trains data from various medical institutions while protecting the privacy of users. This model can provide users with accurate and convenient diagnostic services. Then, a semi-supervised loss function for heterogeneous data is designed to effectively control the difference between local models and global model. In addition, by combining multiple random samplings and accuracy-based weighting method, the contribution of each local model is clarified, and all uneven local models are aggregated into a global consensus model to further reduce the impact of data heterogeneity. Finally, experimental results show that the proposed mechanism has better performance and scalability than several recently proposed mechanisms. |
Key words: Skin lesion Deep learning Image recognition Federated learning |