Diabetic Retinopathy Detection: AI Models and Approaches

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of blindness globally, stemming from diabetes-induced retinal blood vessel damage.
• Early detection of DR is crucial for preventing vision loss, but traditional screening methods are time-consuming and require specialized expertise.
• Recent advancements in artificial intelligence (AI), encompassing classical machine learning and deep learning, offer more accurate and efficient DR detection solutions.

Purpose of the Study:

• To provide a comprehensive review of current AI models and approaches for diabetic retinopathy screening.
• To assess the diagnostic performance and clinical utility of various AI algorithms in DR detection.

Main Methods:

• A systematic literature search was conducted across major databases (PubMed, Web of Science, Scopus, ScienceDirect, EBSCOhost).
• Keywords included 'diabetes,' 'retinopathy,' 'screening,' and 'early detection.'
• Studies published in English between 2020 and 2025 were included.

Main Results:

• Deep learning models have substantially improved DR diagnostic performance, surpassing traditional machine learning.
• Specific AI models (URNet, ViT, ResNet-50, EfficientNetB0, DenseNet, ResNet-18) demonstrated high performance on public datasets.
• Screening devices like ADX-DR, EyeArt, and Google AI achieved high sensitivity and specificity, with EyeArt and Google AI matching or exceeding specialist performance.

Conclusions:

• AI, particularly deep learning (e.g., CNNs), has achieved expert-level accuracy in DR classification with real-world validation.
• Semiautonomous systems (IDx-DR, EyeArt) offer clinical scalability, especially in regions with limited ophthalmologists.
• Future AI techniques like ensemble models and federated learning promise further enhancements in accuracy and reliability for global DR prevention.