[Using artificial intelligence as an initial triage strategy in diabetic retinopathy screening program in China]

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Early detection and screening are crucial for managing DR.
• AI offers potential for improving the efficiency of DR screening programs.

Purpose of the Study:

• To evaluate the diagnostic accuracy and efficiency of an AI triaging model in a DR screening program.
• To assess the AI model's ability to identify patients requiring manual grading for DR.

Main Methods:

• A DR screening program included 8,005 diabetic patients.
• An AI algorithm triaged fundus images, directing cases for immediate AI reports or manual grading.
• Accuracy and sensitivity were assessed by comparing AI and manual grading against a specialist panel.

Main Results:

• AI triaging directed 65.8% of patients to immediate AI reports, reducing manual grading workload by 60%.
• In the AI-reported group (Group A), AI achieved 100% accuracy and specificity for referral DR.
• In the manually graded group (Group B), AI sensitivity was 100%, with 75.8% accuracy.

Conclusions:

• The AI triaging model effectively reduces manual grading workload in DR screening.
• The AI model demonstrates high accuracy and sensitivity, ensuring no referral patients are missed.
• AI integration can enhance the efficiency and scalability of DR screening programs.