Token pyramid pooling-driven style adapter learning with dual-view balanced loss for imbalanced diabetic retinopathy grading

doi:10.1016/j.patcog.2025.112194

	Token pyramid pooling-driven style adapter learning with dual-view balanced loss for imbalanced diabetic retinopathy grading
	Zhao, Jilu 1; Zhang, Xiaoqing 1,2,3; Zhang, Jiawei 4; Sun, Hanxi 5; Nie, Qiushi 1; Xiao, Zunjie 1; Xiao, Linxia 2,3; Zhang, Fengyun 6; Hu, Yan 1; Liu, Jiang 1,7,8
	2026-03-01
发表期刊	PATTERN RECOGNITION
ISSN	0031-3203
卷号	171 页码:13
摘要	Precise diabetic retinopathy (DR) grading is essential for developing personalized and effective treatment plans. Although deep neural networks (DNNs) have achieved promising DR grading results, constructing a precise and trustworthy DR grading model remains challenging due to limited high-quality medical image data, high computational costs, and imbalanced data distributions. To tackle these challenges, we explore the transferability of feature representations from pre-trained vision foundation models (VFMs) to fundus images through adapter learning, aiming to build an efficient imbalanced DR grading model. Unlike classical full-tuning, which fine-tunes all pre-trained parameters of VFMs, adapter learning achieves competitive performance by adding negligible finetuned parameter number. Motivated by the above analysis, we develop a Token Pyramid Pooling-Driven Style Adapter Learning (TPDSAL) to better capture task-specific feature representations from VFMs, which fully exploits pathological distribution prior of DR and the inherent fundus imaging characteristics. Besides, we propose a novel dual-view balanced loss (DVB) to improve imbalanced DR grading performance and trustworthiness, which explores the potential of training class frequencies in sample-wise predicted logit space and sample-wise loss value space simultaneously. Extensive experiments on four public fundus image datasets manifest the superiority of our TPDSAL with DVB over competitive transfer tuning and loss methods in terms of imbalanced grading performance and trustworthiness. Further analysis suggests that clinical prior knowledge utilization is beneficial for adapter learning in capturing task-specific feature representations from VFMs.
关键词	Imbalanced DR grading Adapter learning Pre-trained vision foundation models Dual-view balanced loss Trustworthniess Transferability
DOI	10.1016/j.patcog.2025.112194
收录类别	SCI
语种	英语
资助项目	National Key R&D Program of China[2024YFC2510800] ; General Program of National Natural Science Foundation of China[82272086] ; General Program of National Natural Science Foundation of China[62303442] ; Guangdong Basic, Applied Basic Research Foundation[2025A1515010423] ; Shenzhen Medical Research Fund[D2402014]
WOS研究方向	Computer Science ; Engineering
WOS类目	Computer Science, Artificial Intelligence ; Engineering, Electrical & Electronic
WOS记录号	WOS:001543234100006
出版者	ELSEVIER SCI LTD
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/41991
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Liu, Jiang
作者单位	1.Southern Univ Sci & Technol, Res Inst Trustworthy Autonomous Syst, Dept Comp Sci, Shenzhen, Peoples R China 2.Chinese Acad Sci, Ctr High Performance Comp, Shenzhen, Peoples R China 3.Chinese Acad Sci, Shenzhen Inst Adv Technol, Shenzhen Key Lab Intelligent Bioinformat, Shenzhen, Peoples R China 4.Anhui Med Univ, Affiliated Hosp 1, Dept Gen Surg, Hefei, Anhui, Peoples R China 5.Georgia Inst Technol, H Milton Stewart Sch Ind & Syst Engn, Atlanta, GA USA 6.Southwest Univ, Coll Artificial Intelligence, Chongqing, Peoples R China 7.Univ Nottingham Ningbo China, Sch Comp Sci, Ningbo, Zhejiang, Peoples R China 8.Wenzhou Med Univ, Sch Ophthalmol & Optometry, Wenzhou 325035, Peoples R China
推荐引用方式 GB/T 7714	Zhao, Jilu,Zhang, Xiaoqing,Zhang, Jiawei,et al. Token pyramid pooling-driven style adapter learning with dual-view balanced loss for imbalanced diabetic retinopathy grading[J]. PATTERN RECOGNITION,2026,171:13.
APA	Zhao, Jilu.,Zhang, Xiaoqing.,Zhang, Jiawei.,Sun, Hanxi.,Nie, Qiushi.,...&Liu, Jiang.(2026).Token pyramid pooling-driven style adapter learning with dual-view balanced loss for imbalanced diabetic retinopathy grading.PATTERN RECOGNITION,171,13.
MLA	Zhao, Jilu,et al."Token pyramid pooling-driven style adapter learning with dual-view balanced loss for imbalanced diabetic retinopathy grading".PATTERN RECOGNITION 171(2026):13.