Deep Learning and Transformer-Based Architectures for Diabetic Retinopathy Detection and Classification: A Comprehensive Review, Challenges, and Future Directions
DOI:
https://doi.org/10.64764/ijietas.v2.i2.03Keywords:
Diabetic retinopathy; Deep learning; Vision transformers; Cross-attention; Medical image classificationAbstract
Diabetic Retinopathy (DR) continues to be a major global cause of preventable blindness and vision impairment, thus emphasizing the need for timely screening leading to effective clinical solutions. Carrying out the diagnosis by expert ophthalmologists is a time-consuming process which is limited because of the global lack of specialists, particularly in rural areas. Recently, automated computer-aided diagnosis (CAD) systems based on artificial intelligence have changed the retinal image analysis field. The paper provides a thorough analysis of the recent studies in the period from 2024 to 2026 from which 22 ground-breaking papers were included. The classification of the existing methods is suggested, where they are divided into four major groups: pure transformer and dual-attention based approaches, hybrid Convolutional Neural Network (CNN) and Vision Transformer (ViT) models, Swin-transformer methods optimized to different conditions, and multi-modal fusion methods with simple explainable artificial intelligence (XAI) added. Moreover, the paper describes the obstacles that still exist as far as technical issues are concerned, including strong class imbalance at the stages of development, cross-device domain shifts, and computational challenges observed in the existing models.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Author(s)

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.






