Related Experiment Video
Updated: May 3, 2026
Inducement and Evaluation of a Murine Model of Experimental Myopia
Published on: January 22, 2019
A Novel Time-Aware Deep Learning Model Predicting Myopia in Children and Adolescents
Ana Maria Varošanec1,2, Leon Marković1,2, Zdenko Sonicki3
1University Eye Department, University Hospital "Sveti Duh", Reference Center of The Ministry of Health of The Republic of Croatia for Pediatric Ophthalmology and Strabismus, Reference Center of The Ministry of Health of The Republic of Croatia for Inherited Retinal Dystrophies, Zagreb, Croatia.
This study predicts children's myopia progression using advanced AI, achieving high accuracy for spherical equivalent (SE) forecasting. The novel method offers a clinically acceptable error rate for myopia management.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Data Science
Background:
- Myopia is a growing global health concern in children and adolescents.
- Accurate prediction of refractive error progression is crucial for timely intervention and management.
- Existing methods may not fully capture the complexities of longitudinal vision data.
Purpose of the Study:
- To quantitatively predict the spherical equivalent (SE) in children and adolescents.
- To leverage variable-length historical vision records for myopia prediction.
- To assess the efficacy of a novel time-aware LSTM model for SE forecasting.
Main Methods:
- Retrospective analysis of 895 myopic children and adolescents (aged 4-18).
- Utilized a novel modification of time-aware long short-term memory (LSTM) network.
- Incorporated extended gate mechanisms to capture temporal features in irregular time-series data.
Main Results:
- The model achieved a mean absolute prediction error (MAE) of 0.10 ± 0.15 diopters (D) for SE.
- Prediction accuracy was influenced by sequence length, prediction duration, age, and myopia severity.
- MAE ranged from 0.03 ± 0.04 D to 0.45 ± 0.24 D based on these factors.
Conclusions:
- Extended gate time-aware LSTM effectively predicts SE in young individuals within a 7-year timeframe.
- The overall prediction error (0.10 ± 0.15 D) is significantly below the clinical acceptability threshold (0.75 D).
- This AI-driven approach aids in the early identification and management of myopia progression.
