摘要
医学图像分割作为语义分割中的重要一环关系人类健康,一直以来备受关注和重视。其中,视网膜血管分割任务是对眼底视网膜图像中的血管像素进行分割提取,能够帮助医生快速诊断眼部疾病。但是,视网膜血管形态复杂、结构细小,分割难度大。随着深度学习领域研究不断深入,技术的不断进步使得图像分割精度大幅提升。为了更好地了解视网膜血管分割方法的发展,全面总结了近年来基于深度学习的视网膜血管分割研究成果。首先介绍了视网膜血管分割常用数据集,讨论了关键评价指标和损失函数;然后将成果按照基于网络结构设计的方法(例如U型网络变体)、基于模块设计的方法(例如注意力模块)、基于生成对抗模型的方法和基于Transformer的方法进行分类总结,分析各方法的优缺点并比较模型性能优劣;最后,针对目前视网膜血管分割领域存在的几大问题和挑战,讨论了对应的解决方案和思路,并对未来发展方向进行展望,以期进一步推动视网膜血管分割技术的进步,具有较好参考价值。
As an important part of semantic segmentation related to human health,medical image segmentation has always been highly concerned and valued.Among them,retinal vessel segmentation is to segment and extract vascular pixels from retinal images in the fundus,which can help doctors quickly diagnose eye diseases.However,the morphology of retinal blood vessels is complex and the structure is small,making segmentation difficult.With the continuous development of research in the field of deep learning,the continuous advancement of technology has greatly improved the accuracy of image segmentation.To better understand the development of retinal vessel segmentation methods,this article comprehensively summarized the research results of deep learning-based retinal vessel segmentation in recent years.Firstly,it introduced the commonly used datasets for retinal vessel segmentation,and discussed the key evaluation indicators and loss function.Then it classified and summarized the results according to methods based on network structure design(such as U-shaped network variants),module design(such as attention modules),generative adversarial learning,and Transformer,analyzed the advantages and disadvantages of each method,and compared the performance of the models.Finally,it discussed the corresponding solutions and ideas for several major problems and challenges in retinal vessel segmentation,provided prospects for future development to promote retinal vessel segmentation technology’s progress further,and had good reference value.
作者
张文豪
瞿绍军
颜美丽
Zhang Wenhao;Qu Shaojun;Yan Meili(College of Information Science&Engineering,Hunan Normal University,Changsha 410081,China)
出处
《计算机应用研究》
北大核心
2025年第5期1299-1311,共13页
Application Research of Computers
基金
国家自然科学基金资助项目(12071126)
湖南省教育厅科学研究重点资助项目(23A0081)。
