Deep Learning-Based Prediction of Refractive Error Using Photorefraction Images Captured by a Smartphone: Model...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate refractive error prediction in children is vital for early amblyopia detection and treatment.
• Amblyopia can cause permanent visual impairment if not identified early.
• Smartphone-based screening tools offer accessible methods for mass amblyopia risk assessment.

Purpose of the Study:

• To develop an automated deep learning system for predicting refractive error in children.
• To utilize smartphone-captured eccentric photorefraction images for refractive error prediction.
• To establish an accurate prediction algorithm for efficient pediatric vision screening.

Main Methods:

• A deep learning model was trained using 305 eccentric photorefraction images from children (mean age 4.32 years).
• Images were classified into seven refractive error categories based on cycloplegic refraction measurements.
• The system automated the prediction of refractive error ranges.

Main Results:

• The deep learning system achieved an overall accuracy of 81.6% in predicting refractive error.
• Accuracies for specific refractive error classes ranged from 75.0% to 83.3%.
• The system demonstrated sufficient accuracy across various diopter ranges.

Conclusions:

• Smartphone-based deep learning systems show significant potential for precise refractive error prediction in children.
• This study provides a robust dataset of pediatric photorefraction images for future research.
• The developed system can enhance early detection of vision impairments like amblyopia.