AutoMorph: automated retinal vascular morphology quantification via a deep learning pipeline

AI Generated Summary*

An open source deep learning pipeline, AutoMorph, derives retinal vascular morphology metrics from color fundus photographs via four modules: preprocessing, quality grading, segmentation of vessels, arteries/veins and optic disc/cup, and feature measurement (density, tortuosity, fractal dimension, caliber). Grading tested internally on EyePACS-Q reached an F1 of 0.86, and an eight model ensemble with confidence thresholds cut wrongly accepted ungradable images by 76%, at the cost of rejecting some usable ones. Externally, all ungradable DDR images were caught; vessel segmentation F1 was 0.78 (DR HAGIS) and 0.73 on ultra-widefield AV-WIDE, artery/vein 0.66 (IOSTAR-AV), disc 0.94 (IDRID). Agreement with expert-based features was good to excellent, with runtime near 20 seconds per image on one GPU.

*This summary was generated by AI and is published unedited. Oku does not alter these summaries. It may contain errors or omissions and is provided for general informational purposes only. Oku does not guarantee its accuracy, completeness, or reliability. For authoritative information, please refer to the original, peer-reviewed article.