Study on fundus characteristics of preschool children based on artificial intelligence quantitative technology

The study aimed to establish the first artificial intelligence (AI)-based quantitative reference standard for fundus characteristics in 3–5-year-old preschool children, analyzing microstructural features and their correlations with myopia biomarkers to inform age-stratified standards and early myopia warning. This cross-sectional observational study involved healthy preschoolers (3–5 years) without ocular/systemic disorders affecting refraction. 45° fovea-centered fundus images were captured via handheld camera, and a TransUnet model automated segmentation/quantification of optic disc/cup, retinal vasculature, and fundus tessellated density (FTD). Refractive parameters including spherical equivalent (SE), axial length (AL), and corneal curvature were measured, with statistical analyses exploring age-related variations and correlations. Age-stratified reference intervals were defined. Optic disc/cup parameters (median area: 2.60 mm², Cup-to-Disc Area Ratio: 0.23) showed no age-group differences (P > 0.05). Retinal vascular parameters varied with age, while macular FTD exhibited zonal age differences (P < 0.05) and correlated positively with AL (r = 0.141–0.207, P < 0.05). SE associated positively with vessel density/tortuosity (r = 0.128–0.280, P < 0.05), but FTD did not correlate with SE (P > 0.05). This study integrated AI with fundus imaging to establish quantitative fundus parameter datasets, defining age-stratified references and refractive correlations. Findings provide a scientific basis for pediatric myopia early warning and preventive interventions by identifying anatomical correlations between fundus parameters and refractive development.