A deep learning system for fully automated retinal vessel measurement in high throughput image analysis

AI Generated Summary*

A deep learning pipeline for segmenting arteries, veins and the optic disc in fundus photographs and quantifying vessel morphology was built and tested. Development drew on 420 newly labeled images covering diabetic retinopathy, glaucoma, macular degeneration, pathologic myopia and hypertension, plus 20 public datasets, using a multi-branch U-Net. Across seven test sets, artery AUCs ran 0.91 to 0.96 and vein AUCs 0.93 to 0.96, sensitivities 0.68 to 0.87. Caliber equivalents agreed excellently with manual tracing internally (ICC 0.93 arteries, 0.97 veins) but only moderately or worse on macula-centered public sets. Repeat images gave ICCs of 0.78 to 0.98 in the Standard zone and 0.54 to 0.82 for whole fundus measures, with batch processing near 2 seconds per image.

*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.