Automatic Refractive Error Estimation Using Deep Learning-Based Analysis of Red Reflex Images.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Refractive error is a significant cause of vision impairment.
• Early detection and intervention are crucial for managing conditions like myopia.
• Current screening methods can be costly and inaccessible in certain regions.

Purpose of the Study:

• To evaluate the efficacy of a deep learning model in extracting refractive error metrics from pupillary red reflex images.
• To assess the potential of using low-cost handheld fundus cameras for vision screening.
• To explore a novel, economical approach for early detection of refractive errors.

Main Methods:

• Convolutional neural networks were trained using infrared and color pupillary crescent images.
• Images were sourced from participants at Choithram Hospital, India, and Dargaville Medical Center, New Zealand.
• The model predicted refractive error in terms of spherical and cylindrical power.

Main Results:

• The best-performing model achieved 75% accuracy in predicting spherical power using infrared images.
• A multiclass classifier approach was employed for prediction.
• The study demonstrated the usability of red reflex images for refractive error estimation.

Conclusions:

• The proposed method shows promise for estimating refractive error using red reflex images, despite not achieving superior accuracy.
• This novel approach can guide future research in portable and smartphone-based eye care solutions.
• The technique offers a potential rapid and economical vision-screening method for early intervention.