Fundus Fluorescein Angiography Images with Multi-Lesion Annotations

A dataset of 428 standard 55° and 53 ultra-wide-field fundus fluorescein angiography images annotated for four retinal vascular lesions. The dataset was created by Ziwei Zhao for developing a deep learning segmentation framework and was last updated in April 2026.

Use Cases

• Train a model to segment non-perfusion areas in 55° FFA images using annotated pixel masks.
• Develop a classifier to detect microaneurysms from image features for diabetic retinopathy assessment.
• Benchmark neovascularization segmentation performance across different angiographic phases (arteriovenous, venous, late).
• Study domain shift effects by comparing model performance on standard 55° versus ultra-wide-field FFA images.
• Analyze lesion segmentation differences between retinal vein occlusion and diabetic retinopathy conditions.

Strengths

• 428 annotated standard 55° FFA images provides a primary test set for model evaluation.
• Annotations cover four specific lesion types: non-perfusion areas, microaneurysms, neovascularization, and laser spots.
• Model performance metrics (Dice scores) are provided for each lesion type on the 55° test set.

Limitations

• Ultra-wide-field subset is very small with only 53 images, limiting analysis to preliminary observation.
• Dataset size of 1.5 MB indicates it contains metadata or documentation, not the raw image files themselves.
• Performance on UWF images was lower for some lesions, indicating potential domain shift issues.

Provenance

Source

figshare, author Ziwei Zhao.

Collection Method

Images annotated for four retinal vascular lesions to train a residual U-Net framework.

Time Range

null

Freshness

Last updated April 2026.

Geography

null