A Fundus Image Dataset for AI-based Artery-Vein Vessel Segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinal artery-vein vessel analysis is crucial for diagnosing systemic chronic and cardiovascular diseases.
• Current artificial intelligence (AI) methods for vessel segmentation rely on data-driven approaches.
• Existing public datasets for retinal vessel segmentation suffer from unsatisfactory data quality.

Purpose of the Study:

• To introduce Fundus-AVSeg, a novel, high-quality dataset for AI-based retinal artery-vein segmentation.
• To provide a resource that addresses the limitations of existing datasets in terms of data quality.
• To facilitate advancements in the automated analysis of retinal vasculature.

Main Methods:

• Established a new fundus image dataset named Fundus-AVSeg.
• Included 100 high-resolution fundus images.
• Utilized pixel-wise manual annotation performed by professional ophthalmologists.

Main Results:

• Developed Fundus-AVSeg, a curated dataset specifically for retinal artery-vein segmentation.
• Ensured high data quality through expert manual annotation.
• Created a valuable resource for training and evaluating AI models.

Conclusions:

• The Fundus-AVSeg dataset is expected to significantly benefit the development of AI algorithms for retinal artery-vein segmentation.
• High-quality annotated data is essential for improving the accuracy of AI in medical image analysis.
• This dataset will support further research in automated diagnosis of systemic diseases via retinal imaging.