Artificial intelligence (AI) is increasingly used to support the grading diagnosis of diabetes-related ocular diseases such as diabetic retinopathy and diabetic macular edema. A systematic review published in Diabetes, Obesity and Metabolism evaluated recent advances in AI-based grading systems and examined challenges related to clinical implementation.
The review followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and included 74 English-language studies published between 2021 and 2025. Literature searches were conducted in the Web of Science database using Boolean operators. The analysis examined AI algorithm types, performance evaluation methods, clinical validation approaches, and translational research related to the grading diagnosis of diabetes-related ocular diseases.
Across the included studies, AI systems formed a technical framework primarily based on supervised learning algorithms combined with multiple integrated models. Compared with manual grading, these systems demonstrated advantages in screening efficiency, diagnostic consistency, and accessibility of eye disease screening.
However, several limitations were identified. These included reduced performance in detecting early-stage lesions, diagnostic challenges in patients with multiple ocular conditions, and limited generalization across imaging devices. Emerging strategies to address these gaps include data augmentation using generative adversarial networks, multimodal data fusion, and improvements in model interpretability. The review indicates that AI may improve screening efficiency and grading consistency for diabetes-related ocular diseases, particularly in settings with limited ophthalmology resources, while further work is required to support large-scale clinical implementation.
