Dual encoding feature filtering generalized attention UNET for retinal vessel segmentation  

双编码特征过滤泛化注意力UNET视网膜血管分割

作　　者：ISLAM Md Tauhidul WU Da-Wen TANG Qing-Qing ZHAO Kai-Yang YIN Teng LI Yan-Fei SHANG Wen-Yi LIU Jing-Yu ZHANG Hai-Xian 陶西洛;吴达文;唐青青;赵凯洋;尹腾;李彦霏;尚文一;刘晶玉;张海仙(四川大学计算机学院机器智能实验室,成都610064;四川大学华西医院眼科,成都610041;四川大学华西临床医学院,成都610041)

机构地区：[1]Machine Intelligence Laboratory,College of Computer Science,Sichuan University,Chengdu 610064,China [2]Department of Ophthalmology,West China Hospital,Sichuan University,Chengdu 610041,China [3]West China School of Medicine,Sichuan University,Chengdu 610041,China

出　　处：《四川大学学报(自然科学版)》2025年第1期79-95,共17页Journal of Sichuan University(Natural Science Edition)

基　　金：四川省科技厅重点研发项目(2022ZDZX0023)。

摘　　要：Retinal blood vessel segmentation is crucial for diagnosing ocular and cardiovascular diseases.Although the introduction of U-Net in 2015 by Olaf Ronneberger significantly advanced this field,yet issues like limited training data,imbalance data distribution,and inadequate feature extraction persist,hindering both the segmentation performance and optimal model generalization.Addressing these critical issues,the DEFFA-Unet is proposed featuring an additional encoder to process domain-invariant pre-processed inputs,thereby improving both richer feature encoding and enhanced model generalization.A feature filtering fusion module is developed to ensure the precise feature filtering and robust hybrid feature fusion.In response to the task-specific need for higher precision where false positives are very costly,traditional skip connections are replaced with the attention-guided feature reconstructing fusion module.Additionally,innovative data augmentation and balancing methods are proposed to counter data scarcity and distribution imbalance,further boosting the robustness and generalization of the model.With a comprehensive suite of evaluation metrics,extensive validations on four benchmark datasets(DRIVE,CHASEDB1,STARE,and HRF)and an SLO dataset(IOSTAR),demonstrate the proposed method’s superiority over both baseline and state-of-the-art models.Particularly the proposed method significantly outperforms the compared methods in cross-validation model generalization.视网膜血管的形态与功能改变对临床中眼部及心血管疾病的诊断具有重要指导意义.依靠人工观察和经验判断眼底图像中视网膜血管状态不仅效率低下且主观性强,因此智能视网膜血管分割技术应运而生.随着深度学习技术不断发展,深度神经网络模型显著提高了视网膜血管的分割性能,但该领域依旧面临着有限的训练数据、数据分布不平衡和特征提取不充分等问题,阻碍了分割性能和模型泛化能力的提升.为了解决这些关键问题,本文提出了DEFFA-Unet,其通过增加专用编码器来处理域不变的预处理输入,以增强特征编码的深度和模型的泛化能力.开发了一个特征过滤融合模块,以确保精确的特征过滤和稳定的特征融合.针对高精度、低误诊率要求的临床应用场景,传统的跳跃连接被替换为注意力引导的特征重构融合模块.此外,本文提出了一种基于目标特征相似性的独特平衡方法JESB,以应对数据分布不平衡的问题;并提出了一种创新的数据增强技术SOTA-CSA,以解决数据限制和跨域问题.实验在4个基准数据集(DRIVE、CHASEDB1、STARE和HRF)和一个SLO数据集(IOSTAR)上进行,采用全面的评估指标进行验证.实验结果表明了DEFFA-Unet相较于基线模型和最先进模型的优越性.

关 键 词：Vessel segmentation Data balancing Data augmentation Dual encoder Attention Mechanism Model generalization 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]