Deep Learning Prediction of Childhood Myopia Progression Using Fundus Image and Refraction Data.

Area of Science:

• Ophthalmology and Computational Science
• Artificial Intelligence in Healthcare
• Pediatric Vision Research

Background:

• Childhood myopia is a growing global health issue with significant risks of irreversible vision impairment.
• Early prediction of myopia progression is vital for timely interventions to prevent high myopia and its complications.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for predicting myopia progression and high myopia risk in schoolchildren.
• The model utilizes only fundus images and baseline refraction data for quantitative prediction.

Main Methods:

• A longitudinal school-based cohort study (Anyang Childhood Eye Study) involved 3048 children aged 6-9 years.
• A DL model combining convolutional neural network (CNN) and recurrent neural network (RNN) was developed.
• Model performance was assessed using Area Under the Curve (AUC) for risk prediction and Mean Absolute Error (MAE) for spherical equivalent refraction (SER) prediction, with external validation in diverse cohorts.

Main Results:

• The DL model achieved high AUC scores: 0.941 for myopia risk and 0.985 for high myopia risk.
• The model predicted SER progression with an overall MAE of 0.322 D/year.
• External validation in Beijing and Lhasa cohorts demonstrated robust cross-ethnicity performance (MAE 0.355 D/year and 0.261 D/year, respectively).

Conclusions:

• A DL model using minimal baseline data offers highly accurate prediction of myopia and high myopia risk.
• This AI-driven approach shows promise for large-scale screening and early intervention strategies, particularly in underserved areas.