Artificial intelligence for early detection of diabetic retinopathy: A vision transformer-based approach

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Overall classification accuracy of 97% on benchmark datasets

• Primary Outcome: Multi-class classification of diabetic retinopathy severity (No DR, Mild, Moderate, Severe, Proliferative DR) using CCT-based Vision Transformer on EyePACS and APTOS 2019 datasets

• Nnt Or Sensitivity: Sensitivity and specificity not reported. F1-scores reported as >0.95 across all DR severity classes. No 2×2 contingency table data provided. Positive and negative likelihood ratios cannot be calculated from reported data.

• Confidence Interval: Not reported

Clinical Application

Technical feasibility of the CCT model is demonstrated on benchmark data. Real-world clinical feasibility is undemonstrated. Integration into existing retinal camera systems, electronic health records, or telehealth platforms is not described. Computational requirements, inference time, and deployment infrastructure are not reported. Regulatory approval pathways (e.g., FDA, CE marking, TGA) are not discussed. In Australia, diabetic retinopathy screening is recommended annually for all people with diabetes, as per RACGP and Diabetes Australia guidelines. The National Diabetes Services Scheme (NDSS) supports access to screening. Telehealth-based retinal screening programmes exist in remote and rural Australia (e.g., the KeepSight programme). AI-assisted DR screening tools require TGA approval as a Class IIa or IIb medical device software (SaMD) under the TGA's Software as a Medical Device framework. This study does not meet the evidentiary standards required for TGA regulatory submission or RACGP guideline endorsement. No Australian dataset validation has been performed. The model's performance on images from Australian community screening settings — which may differ in camera type, image quality, and patient demographics — is entirely unknown. Theoretically applicable to adults with diabetes undergoing fundus photography-based DR screening. However, the study population is defined only by benchmark dataset membership, not by clinical characteristics. Applicability to specific populations (type 1 vs. type 2 diabetes, duration of disease, comorbidities, image acquisition device type) cannot be determined from this paper.

Abstract

BACKGROUND: Early identification of diabetic retinopathy (DR), which is a primary cause of vision impairment globally, is a crucial phasis for effective intervention and treatment. Traditional screening workflows rely on manual diagnosis by ophthalmologists, which remains the gold standard but can be time-consuming and subject to variability due to human factors. To support and enhance the screening process, artificial intelligence (AI)-based tools have shown promise in automating DR detection, particularly with recent advances in deep learning. However, medical images with long-range dependencies and spatial linkages can be challenging for CNN-based algorithms to handle. METHODS: This paper proposes a Vision Transformer (ViT)-based model, specifically using a Compact Convolutional Transformer (CCT), for early automated detection of DR. The model uses self-attention techniques to improve feature extraction and classification performance; combining three main stages: the CCT tokenizer, transformer encoder, and sequence pooling. The proposed approach was trained on public datasets (EyePACS and APTOS 2019) and evaluated against state-of-the-art deep learning architectures. RESULTS: Our experimental findings demonstrate that ViT performs among the best in the current state of the art with an overall accuracy of 97% and F1-scores above 0.95 across all DR severity levels. Our system is primarily designed for the pre-screening stage of diabetic retinopathy workflows, enabling rapid and reliable identification of potential DR cases for further clinical evaluation. CONCLUSION: These results highlight the potential of transformer-based designs in medical picture analysis, as well as the implications for telemedicine and e-health solutions in real-time, especially in cases of low-resource settings.

References

1. 1.ElAdel, A., Filali, I., & Zaied, M. (2026). Artificial intelligence for early detection of diabetic retinopathy: A vision transformer-based approach. PLOS ONE. https://doi.org/10.1371/journal.pone.0350854