QUARTZ: Cross-Dataset Artificial Intelligence-Enabled Retinal Vasculometry

QUARTZ, an AI-enabled retinal vasculometry system, achieves external EPIC-Norfolk segmentation F1 scores up to 0.9532 and image-quality classification F1 of 0.9197.

• Why it matters: Retinal vessel morphology is linked to ocular and systemic diseases, but large epidemiological studies need automated analysis that works across varied image quality and acquisition settings.
• What they did: QUARTZ was trained using 26 annotated retinal image datasets spanning demographics, diseases, camera types, and protocols, and was evaluated on EPIC-Norfolk for segmentation and image-quality classification.
• The result: Cross-dataset generalisation improved arteriole/venule and optic disc segmentation, but did not improve optic cup segmentation or image-quality classification, and an online tool plus CHASE-AV database were released.

AI summary of the abstract · may contain errors · not medical advice · read the paper.