Feasibility and acceptance of artificial intelligence-based diabetic retinopathy screening in Rwanda

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) screening requires practical, evidence-based applications of artificial intelligence (AI).
• Current screening methods need validation with AI interpretation in diverse clinical settings.

Purpose of the Study:

• To assess the practical application and effectiveness of AI-based diabetic retinopathy screening in Rwandan diabetes clinics.
• To evaluate participant satisfaction and preference for AI screening compared to human grading.

Main Methods:

• Retinal imaging and AI interpretation were used for DR screening in 827 participants across four Rwandan diabetes clinics.
• AI detection included DR, optic nerve anomalies, and macular anomalies, with referable criteria defined.
• Human grading by a UK National Health System-certified grader served as a benchmark for AI performance.

Main Results:

• AI screening resulted in 33.2% of participants being referred; 99.5% reported high satisfaction, with 63.7% preferring AI over human grading.
• AI demonstrated 92% sensitivity and 85% specificity for referable DR compared to human grading.
• Referral reasons varied, with the highest adherence (53.4%) observed for participants referred for DR.

Conclusions:

• AI-based diabetic retinopathy screening is accurate and practical in Rwandan diabetes clinics.
• High participant satisfaction and preference for AI indicate its acceptability in clinical settings.
• AI screening shows promise for improving DR detection and management in resource-limited areas.