Fundus Fluorescein Angiography Images with Multi-Lesion Annotations

428 standard 55° and 53 ultra-wide-field fundus fluorescein angiography images are annotated for four retinal lesion types. Ziwei Zhao published this dataset in 2026 to support deep learning model development for automated lesion segmentation. The dataset includes performance metrics for a residual U-Net model trained on the images.

Use Cases

• Train segmentation models to identify non-perfusion areas in 55° FFA images using provided annotations.
• Benchmark model performance on microaneurysm segmentation using reported Dice scores of 0.70 ± 0.13.
• Analyze domain shift effects by comparing segmentation results between 55° and ultra-wide-field image subsets.
• Study lesion prevalence differences across diseases like diabetic retinopathy and retinal vein occlusion.
• Evaluate segmentation model performance across different angiographic phases (arteriovenous, venous, late).

Strengths

• 428 annotated standard 55° FFA images provides a primary test set.
• Annotations cover four specific lesion types: non-perfusion areas, microaneurysms, neovascularization, and laser spots.
• Includes preliminary exploratory data from 53 ultra-wide-field FFA images.
• Model performance metrics (Dice, IoU, recall) are provided for each lesion category.

Limitations

• The ultra-wide-field subset has only 53 images, noted as a limited sample size for analysis.
• Dataset scope is limited to 25.6 KB, indicating it contains metadata or annotations, not the raw image files themselves.
• Geographic and demographic source information for the patient images is not provided.

Provenance

Source

figshare, author Ziwei Zhao.

Collection Method

Images were annotated for four retinal vascular lesions to train a deep learning framework.

Time Range

null

Freshness

Last updated in April 2026.

Geography

null