Generalized Analysis of Vessels in Eye (GAVE) Challenge 2025

This challenge focuses on analyzing the vascular structure in color fundus photographs, specifically extracting retinal blood vessels and distinguishing between arteries and veins. Retinal blood vessel characteristics, such as caliber and structure, serve as important biomarkers for diagnosing and monitoring various medical conditions, including glaucoma, age-related macular degeneration, diabetic retinopathy, and hypertension. These biomarkers can be obtained through color fundus photography, a non-invasive and cost-effective imaging technique. Due to its affordability and ease of use, this technique has been widely adopted in clinical practice, research, and national screening programs.

To conduct a comprehensive analysis of the retinal vascular system, it is essential to first segment the blood vessels and then classify them as arteries or veins. The resulting artery-vein segmentation maps enable the quantification of diagnostically relevant features such as blood vessel width, diameter, and tortuosity. Accurate measurement of these features further facilitates the calculation of complex biomarkers, such as the arteriolar-to-venular diameter ratio.

Beyond blood vessel and artery-vein segmentation in color fundus photographs, this challenge introduces a novel task: the automatic measurement of the arteriovenous ratio. This measurement is closely linked to various cardiovascular conditions, including arteriosclerosis and hypertensive lesions, adding both depth to the challenge and significant implications for real-world medical applications. The core objective of this challenge is to achieve automated and precise blood vessel segmentation, artery-vein classification, and arteriovenous ratio measurement using color fundus photographs.

The challenge dataset consists of 150 color fundus photographs collected from different hospitals using various imaging devices, along with corresponding segmentation and arteriovenous ratio annotations. In the preliminary stage, 50 datasets with labels will be provided for model training, followed by the release of another 50 datasets
for validation. An online evaluation platform will be available, allowing teams to refine their models based on leaderboard rankings. In the final stage, the remaining 50 datasets will be released for model evaluation. From a technical perspective, this challenge involves research on image segmentation in computer vision, a critical aspect of computer-aided clinical diagnosis. From a biomedical perspective, it aims to accurately extract the retinal vascular structure, contributing significantly to the understanding and identification of various diseases.