AI model forecasts myopia progression from baseline retinal images

A deep learning model using only fundus images and baseline refraction data was able to accurately predict both myopia progression and the risk of developing high myopia in school-aged children across different populations.

The study analyzed data from a longitudinal, school-based cohort that followed Grade 1 students aged 6 to 9 years from multiple urban primary schools over 6 years. Children who had received myopia control treatments, had amblyopia, or had undergone strabismus surgery were excluded. Two independent external cohorts from different regions and ethnic backgrounds were used for validation.

The researchers developed a deep learning model to predict trajectories of myopia progression and identify individuals at high risk of myopia. Model performance was evaluated using area under the curve (AUC) for risk prediction and mean absolute error (MAE) for spherical equivalent refraction (SER) prediction. Myopia was defined as SER ≤ −0.5 D and high myopia as SER ≤ −6.0 D, based on cycloplegic autorefraction.

Among the 3,048 children included in the primary cohort, the baseline prevalence of myopia was 5.7%, and the prevalence of high myopia was 0.5%. The model achieved an AUC of 0.941 for predicting myopia risk and 0.985 for predicting high myopia risk. The overall MAE for SER prediction was 0.322 D per year.

External validation showed similar performance across different populations.

Reference
Kang MT, Hu Y, Wang N, et al. Deep Learning Prediction of Childhood Myopia Progression Using Fundus Image and Refraction Data. JAMA Netw Open. 2026;9(1):e2553543. doi: 10.1001/jamanetworkopen.2025.53543. PMID: 41587032.