Scholarlink

Retinal vessel characteristics serve as crucial biomarkers for screening and diagnosing various diseases. Retinal vessel segmentation, particularly arteriovenous (A/V) segmentation, is a key step in enabling AI-assisted disease screening and diagnosis. Fundus photography, a non-invasive retinal imaging technique, is widely accessible and cost-effective. To advance AI applications in screening and diagnosing conditions such as diabetes and cardiovascular diseases, we collaborated with Medical Image Computing and Computer Assisted Intervention (MICCAI) 2025 to launch the Generalized Analysis of Vessels in Eye (GAVE) Challenge. This initiative provides a dataset with expert annotations for three research tasks: vessel segmentation, A/V segmentation, and quantitative biomarker measurement in fundus images. In the annotation process of the dataset, the fluorescein fundus angiography(FFA) paired with color fundus photos are introduced, which can provide a clearer and more accurate annotation reference than relying on color fundus photos alone. This is the first and biggest dataset to incorporate paired FFA into the vascular annotation of the artery/vein and provides arteriovenous ratio (AVR) parameter annotations. This paper describes the released dataset of 150 color fundus images with corresponding annotations, baseline methods for the three subtasks, and evaluation protocols. The GAVE Challenge is accessible at https://aistudio.baidu.com/competition/detail/1315 .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dataset, Baseline and Evaluation Design for GAVE Challenge

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Zhiwei Liu,
  • Huihui Fang,
  • Shandi Liu,
  • Ping Zhang,
  • Wenqun Xi,
  • Changjian Wu,
  • Zihao Zhong,
  • Xinyu Fu,
  • Qifan Yang,
  • Xiangyu Chen,
  • Mengxiong Luo,
  • Xinya Hu,
  • Jiongning Zhao,
  • Mingkui Tan,
  • Weihua Yang,
  • Yanwu Xu

摘要

Retinal vessel characteristics serve as crucial biomarkers for screening and diagnosing various diseases. Retinal vessel segmentation, particularly arteriovenous (A/V) segmentation, is a key step in enabling AI-assisted disease screening and diagnosis. Fundus photography, a non-invasive retinal imaging technique, is widely accessible and cost-effective. To advance AI applications in screening and diagnosing conditions such as diabetes and cardiovascular diseases, we collaborated with Medical Image Computing and Computer Assisted Intervention (MICCAI) 2025 to launch the Generalized Analysis of Vessels in Eye (GAVE) Challenge. This initiative provides a dataset with expert annotations for three research tasks: vessel segmentation, A/V segmentation, and quantitative biomarker measurement in fundus images. In the annotation process of the dataset, the fluorescein fundus angiography(FFA) paired with color fundus photos are introduced, which can provide a clearer and more accurate annotation reference than relying on color fundus photos alone. This is the first and biggest dataset to incorporate paired FFA into the vascular annotation of the artery/vein and provides arteriovenous ratio (AVR) parameter annotations. This paper describes the released dataset of 150 color fundus images with corresponding annotations, baseline methods for the three subtasks, and evaluation protocols. The GAVE Challenge is accessible at https://aistudio.baidu.com/competition/detail/1315 .