Automated AI quantification of retinal parameters associated with axial length progression in children: a cohort study

Purpose This study investigated whether quantitative retinal parameters derived from colour fundus photography are associated with axial length (AL) progression in school-age children. Methods In this prospective cohort, 556 school-age children underwent comprehensive baseline evaluations, including cycloplegic refraction, AL measurement and colour fundus photography (CFP) at baseline and follow-up. Quantitative retinal features were automatically extracted from fundus images using an artificial intelligence (AI)-assisted platform. Children were classified into Group 0 and Group 1 according to an annual AL progression of 0.2 mm. Univariable screening was followed by multivariable logistic regression adjusted for sex and baseline AL. Sensitivity analyses were adjusted for baseline SE or simultaneously for baseline AL and SE and included standardised effect estimates and Firth bias-reduced logistic regression. Results Among 556 children, 276 were classified as Group 0 and 280 as Group 1. After adjustment for sex and baseline AL, greater average venous curvature between 0.5 and 1.0 PD remained associated with lower odds of AL progression (OR per SD increase=0.79, 95% CI 0.66 to 0.95, p=0.012), whereas the association with the macula–optic disc angle was attenuated (OR per SD increase=0.85, 95% CI 0.71 to 1.01, p=0.068). Venous curvature remained significant in sensitivity analyses adjusting for baseline SE or both baseline AL and SE. Conclusion Selected quantitative fundus features, particularly peripapillary venous curvature, were associated with 1 year AL progression in school-age children. These findings suggest that CFP-derived parameters may have potential value as indicators of AL progression, although further validation in independent cohorts is needed.