Rotterdam Artery-Vein Segmentation in 53 Fundus Images

53 color fundus images from the Department of Ophthalmology at Erasmus MC, Rotterdam, are annotated with semi-manual artery-vein segmentations. The annotations were created by refining outputs from a vessel segmentation model and checking for correct connectivity of artery and vein trees. A private set of 155 additional images is available under a separate agreement.

Use Cases

• Train a segmentation model to distinguish artery and vein trees using the semi-manual annotations as ground truth.
• Validate automated vessel classification algorithms against the connectivity-checked artery and vein segmentations.
• Analyze retinal vascular patterns in color fundus images using the provided artery-vein labels.
• Develop image processing techniques for overlapping artery and vein tree coloring based on the annotation methodology.

Strengths

• 53 publicly available annotated images provide a defined benchmark set.
• Annotations were manually refined and checked for correct anatomical connectivity of artery and vein trees.
• Dataset originates from a specialized ophthalmology department at Erasmus MC, Rotterdam.

Limitations

• The public dataset is relatively small with only 53 images, which may limit model training without the private set.
• Access to the larger private set of 155 images requires signing an End User License Agreement (EULA).
• No specific details on image resolution, patient demographics, or associated clinical variables are provided.

Provenance

Source

Department of Ophthalmology, Erasmus MC, Rotterdam

Collection Method

Semi-manual annotation via refinement of model outputs and connectivity checking.

Time Range

null

Freshness

null

Geography

Rotterdam, Netherlands