Adaptive Dual-Axis Style-Based Recalibration Network With Class-Wise Statistics Loss for Imbalanced Medical Image Classification

doi:10.1109/TIP.2025.3551128

	Adaptive Dual-Axis Style-Based Recalibration Network With Class-Wise Statistics Loss for Imbalanced Medical Image Classification
	Zhang, Xiaoqing 1,2,3,4; Xiao, Zunjie 3,4; Ma, Jingzhe 5; Wu, Xiao 3,4; Zhao, Jilu 3,4; Zhang, Shuai 6,7; Li, Runzhi 8; Pan, Yi 1,2,9; Liu, Jiang 3,4,10,11,12
	2025
发表期刊	IEEE TRANSACTIONS ON IMAGE PROCESSING
ISSN	1057-7149
卷号	34 页码:2081-2096
摘要	Salient and small lesions (e.g., microaneurysms on fundus) both play significant roles in real-world disease diagnosis under medical image examinations. Although deep neural networks (DNNs) have achieved promising medical image classification performance, they often have limitations in capturing both salient and small lesion information, restricting performance improvement in imbalanced medical image classification. Recently, with the advent of DNN-based style transfer in medical image generation, the roles of clinical styles have attracted great interest, as they are crucial indicators of lesions. Motivated by this observation, we propose a novel Adaptive Dual-Axis Style-based Recalibration (ADSR) module, leveraging the potential of clinical styles to guide DNNs in effectively learning salient and small lesion information from a dual-axis perspective. ADSR first emphasizes salient lesion information via global style-based adaptation, then captures small lesion information with pixel-wise style-based fusion. We construct an ADSR-Net for imbalanced medical image classification by stacking multiple ADSR modules. Additionally, DNNs typically adopt cross-entropy loss for parameter optimization, which ignores the impacts of class-wise predicted probability distributions. To address this, we introduce a new Class-wise Statistics Loss (CWS) combined with CE to further boost imbalanced medical image classification results. Extensive experiments on five imbalanced medical image datasets demonstrate not only the superiority of ADSR-Net and CWS over state-of-the-art (SOTA) methods but also their improved confidence calibration results. For example, ADSR-Net with the proposed loss significantly outperforms CABNet50 by 21.39% and 27.82% in F1 and B-ACC while reducing 3.31% and 4.57% in ECE and BS on ISIC2018.
关键词	Imbalanced medical image classification dual-axis style-based recalibration imbalanced learning class-wise statistics loss confidence calibration Imbalanced medical image classification dual-axis style-based recalibration imbalanced learning class-wise statistics loss confidence calibration
DOI	10.1109/TIP.2025.3551128
收录类别	SCI
语种	英语
资助项目	National Key Research and Development Program of China[2024YFC2510800] ; Shenzhen Basic Research Fund[KQTD20200820113106007] ; Shenzhen Key Laboratory of Intelligent Bioinformatics[ZDSYS20220422103800001] ; National Natural Science Foundation of China[82272086] ; National Natural Science Foundation of China[82273493] ; Henan Provincial Science and Technology Research[232102311232]
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Artificial Intelligence ; Engineering, Electrical & Electronic
WOS记录号	WOS:001459526800002
出版者	IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/40657
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Li, Runzhi; Pan, Yi; Liu, Jiang
作者单位	1.Chinese Acad Sci, Shenzhen Inst Adv Technol, Ctr High Performance Comp, Shenzhen 518055, Peoples R China 2.Chinese Acad Sci, Shenzhen Inst Adv Technol, Shenzhen Key Lab Intelligent Bioinformat, Shenzhen 518055, Peoples R China 3.Southern Univ Sci & Technol, Res Inst Trustworthy Autonomous Syst, Shenzhen 518055, Peoples R China 4.Southern Univ Sci & Technol, Dept Comp Sci & Engn, Shenzhen 518055, Peoples R China 5.Shenzhen Polytech Univ, Guangdong Hong Kong Macao Greater Bay Area Artific, Shenzhen 518055, Peoples R China 6.UCL, EPSRC Ctr Intervent & Surg Sci WEISS, London WC1E 6BT, England 7.UCL, Dept Comp Sci, London WC1E 6BT, England 8.Zhengzhou Univ, Cooperat Innovat Ctr Internet Healthcare, Zhengzhou 450001, Peoples R China 9.Shenzhen Univ Adv Technol, Fac Comp Sci & Control Engn, Shenzhen 518107, Peoples R China 10.Univ Nottingham, Sch Comp Sci, Ningbo 315100, Peoples R China 11.Wenzhou Med Univ, Sch Ophthalmol & Optometry, Wenzhou 325035, Peoples R China 12.Changchun Univ, Dept Elect & Informat Engn, Changchun 130022, Peoples R China
推荐引用方式 GB/T 7714	Zhang, Xiaoqing,Xiao, Zunjie,Ma, Jingzhe,et al. Adaptive Dual-Axis Style-Based Recalibration Network With Class-Wise Statistics Loss for Imbalanced Medical Image Classification[J]. IEEE TRANSACTIONS ON IMAGE PROCESSING,2025,34:2081-2096.
APA	Zhang, Xiaoqing.,Xiao, Zunjie.,Ma, Jingzhe.,Wu, Xiao.,Zhao, Jilu.,...&Liu, Jiang.(2025).Adaptive Dual-Axis Style-Based Recalibration Network With Class-Wise Statistics Loss for Imbalanced Medical Image Classification.IEEE TRANSACTIONS ON IMAGE PROCESSING,34,2081-2096.
MLA	Zhang, Xiaoqing,et al."Adaptive Dual-Axis Style-Based Recalibration Network With Class-Wise Statistics Loss for Imbalanced Medical Image Classification".IEEE TRANSACTIONS ON IMAGE PROCESSING 34(2025):2081-2096.