AI-based oculomics for trajectory-driven risk stratification of pathologic myopia in paediatric high myopia

AIMS: To identify early oculomic biomarkers predictive of pathologic myopia (PM) in children with high myopia (HM) and to develop an artificial intelligence (AI)-based model for individualised risk stratification. METHODS: This prospective longitudinal study included 375 children with bilateral HM (spherical equivalent ≤ -6.00 dioptres (D)) from the Zhongshan High Myopia Cohort, followed for a median of 15.0 years (IQR, 14.9-15.4; range, 14.7-15.7). Deep learning-based automated retinal vascular phenotyping was applied to fundus photographs and integrated with longitudinal ocular biometry and swept-source optical coherence tomography metrics. Multivariable models identified independent predictors of PM onset and predictive performance was assessed using area under the receiver operating characteristic curve (AUROC). A web-based prediction tool was developed for clinical deployment. RESULTS: PM developed in 64 children (17.1%). Independent predictors included lower retinal vessel density (OR, 9.69; 95% CI 4.33 to 21.70), reduced fractal dimension (OR, 5.94; 95% CI 3.07 to 11.50), narrower arteriolar and venular calibres (central retinal arteriolar equivalent: OR, 3.44; 95% CI 1.86 to 6.34; central retinal venular equivalent: OR, 4.42; 95% CI 2.24 to 8.71), thinner subfoveal choroid (OR, 2.64; 95% CI 1.56 to 4.45), faster early axial elongation (OR, 1.72; 95% CI 1.17 to 2.50) and accelerated early choroidal thinning (OR, 2.33; 95% CI 1.41 to 3.84). The AI oculomic model achieved excellent discrimination (AUROC, 0.96; 95% CI 0.92 to 1.00), improving to 0.98 (95% CI 0.95 to 1.00) with biometric variables. These predictors were implemented in a publicly accessible risk prediction platform (System for Myopia AI-based Risk Tracking-PM). CONCLUSIONS: PM risk in paediatric HM is detectable early through retinal microvascular architecture and ocular growth dynamics. AI-enabled oculomic profiling offers a scalable approach to early risk stratification, supporting targeted surveillance and timely prevention of sight-threatening pathology.