Related Experiment Video
Updated: Dec 4, 2025
Author Spotlight: Understanding Retinal Vessel Resilience and Disease Progression
Published on: January 12, 2024
Artificial intelligence-enabled screening for diabetic retinopathy: a real-world, multicenter and prospective study.
Yifei Zhang1, Juan Shi1, Ying Peng1
1Department of Endocrine and Metabolic Diseases, Shanghai Institute of Endocrine and Metabolic Diseases, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China; Shanghai National Clinical Research Center for metabolic Diseases, Key Laboratory for Endocrine and Metabolic Diseases of the National Health Commission of the PR China, Shanghai National Center for Translational Medicine, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
An artificial intelligence system effectively screens for diabetic retinopathy (DR) in China, validating its feasibility for early detection and prevention of blindness in diabetes patients.
Area of Science:
- Ophthalmology
- Artificial Intelligence in Healthcare
- Public Health
Background:
- The global diabetes epidemic necessitates efficient and scalable methods for early diabetic retinopathy (DR) screening to prevent blindness.
- Current screening methods face challenges in accessibility and scalability, particularly in large populations like China.
Purpose of the Study:
- To validate an artificial intelligence (AI)-enabled system for DR screening in adult patients with diabetes in China.
- To investigate the prevalence of DR using this AI system across a large, multicenter cohort.
Main Methods:
- A prospective, nationwide, multicenter study involving 47,269 adult diabetes patients across 155 centers in China.
- Collection of non-mydriatic, macula-centered fundus photographs per eye, analyzed by a deep learning (DL)-based, five-stage DR classification system.
- DL algorithm validation using images from a randomly selected one-third of participants, compared against specialist grading.
Main Results:
- The DL-based DR grading algorithms demonstrated high performance: 83.3% sensitivity and 92.5% specificity for detecting referable DR.
- The five-stage DR classification achieved an 83.0% concordance, comparable to interobserver variability among specialists (84.3%).
- Estimated prevalence of any DR was 28.8%, referable DR 24.4%, and vision-threatening DR 10.8%. Prevalence was higher in females, the elderly, and those with longer diabetes duration or higher HbA1c.
Conclusions:
- The study successfully validated a nationwide, multicenter, AI-enabled DR screening system in China.
- The findings highlight the importance and feasibility of deploying this AI system in clinical practice at diabetes centers for efficient DR screening.
