Smartphone-based offline AI for multi-disease retinal screening: Real-world accuracy

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Diabetic retinopathy (DR), glaucoma, and age-related macular degeneration (AMD) are leading causes of vision loss.
• Early detection and screening are crucial for managing these conditions.
• Smartphone-based fundus cameras offer potential for accessible eye care.

Purpose of the Study:

• To assess the diagnostic accuracy of an offline artificial intelligence (AI) system, Medios-AI (MAI), for simultaneous screening of DR, glaucoma, and AMD.
• To evaluate MAI's performance in a real-world setting using a smartphone-based fundus camera.
• To compare MAI's diagnostic capabilities against expert ophthalmologists.

Main Methods:

• A prospective cross-sectional study enrolled 193 adults (371 eyes).
• Fundus images were captured using smartphone-based and desktop fundus cameras.
• An offline MAI algorithm analyzed images, and results were compared to masked grading by two ophthalmologists.

Main Results:

• MAI demonstrated high sensitivity and specificity for detecting any retinal disease (99.3% and 95.7%, respectively).
• Specific disease detection showed high accuracy: Glaucoma (98.2% sensitivity, 99.0% specificity), AMD (88.9% sensitivity, 97.5% specificity), and DR (84.6% sensitivity, 99.0% specificity).
• Excellent agreement was observed between AI and human graders for vertical cup-to-disc ratio.

Conclusions:

• The Medios-AI system exhibits significant diagnostic accuracy for DR, glaucoma, and AMD.
• Its offline, smartphone-based platform supports scalable, point-of-care retinal screening.
• MAI is a promising tool for improving eye care accessibility, particularly in resource-limited settings.