Related Experiment Video
Updated: Jun 16, 2026
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
Ethnicity-Stratified Normative Retinal Vascular Features from the UK Biobank Using Deep Learning
Ranjit J Injety1,2, Callum Hunt1, Ha-Jun Yoon1
1The University of Leicester Ulverscroft Eye Unit, School of Psychology and Vision Sciences, University of Leicester, Leicester, UK.
Purpose:
Retinal vascular features provide noninvasive biomarkers of systemic vascular health, and deep learning tools such as AutoMorph now enable their large-scale quantification. However, researchers still lack normative data across diverse ethnic populations.
Design:
A prospective cohort study.
Participants:
We analyzed 6843 UK Biobank participants who reported no disease or medical condition according to the World Health Organization's International Classification of Diseases and Related Health Problems.
Methods:
We analyzed retinal images from the UK Biobank using the AutoMorph deep learning pipeline, which extracted retinal morphometric features such as vessel caliber, fractal dimension, vessel density, and tortuosity. We then used these data to define ethnicity-stratified normative ranges for retinal morphometric features in healthy UK Biobank participants. We examined associations with demographic covariates (age, sex, and ethnicity) using multivariate regression.
Main Outcome Measures:
We determined normal ranges for retinal morphometric features in the healthy UK Biobank population and assessed associations with age, sex, and ethnicity.
Results:
The cohort (mean age 53.5 ± 7.9 years; 50.1% male) was predominantly White participants (91.1%). Retinal vascular complexity declined with age, reflected by lower fractal dimension, vessel density, and tortuosity metrics. Zone-specific analyses confirmed age-related reductions in the central retinal artery equivalent and central retinal vein equivalent. Vessel density (0.050 ± 0.005) and fractal dimension (1.052 ± 0.09) were higher among Chinese participants as compared to White participants (P < 0.0001 for both). Across all metrics, ethnicity, followed by sex, exerted the strongest influence on vascular morphometrics.
Conclusions:
We established the first ethnicity-stratified normative dataset of retinal vascular features derived from the UK Biobank using deep learning. These data provide a reference framework for oculomic biomarkers in multi-ethnic populations and could support precision-medicine approaches to systemic disease risk assessment.
Financial Disclosures:
Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.
