Early Detection of Diabetic Retinopathy using Deep Learning

Abstract

Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness worldwide, particularly among the working-age population. Early diagnosis and timely intervention can prevent more than 90% of vision loss cases; however, traditional screening methods rely on manual inspection by ophthalmologists, which is time-intensive, costly, and subject to inter-observer variability. This study proposes a robust and automated deep learning-based framework for the early detection and classification of diabetic retinopathy using retinal fundus images. The proposed approach integrates Convolutional Neural Networks (CNN) with transfer learning techniques to effectively identify and classify multiple stages of DR. The model is trained and evaluated on benchmark datasets, including EyePACS, APTOS 2019, and Messidor, ensuring diversity and generalization capability. Experimental results demonstrate that the proposed system achieves an accuracy of 97.8%, outperforming several baseline models. Furthermore, explainability is incorporated using Gradient-weighted Class Activation Mapping (Grad-CAM), enabling visualization of pathological regions in retinal images and enhancing clinical trust. The proposed solution provides a scalable, efficient, and reliable tool for automated DR screening, particularly beneficial in resource-constrained healthcare environments.

Introduction

This text discusses Diabetic Retinopathy (DR) and the development of AI-based systems for its early detection and classification using retinal images.

DR is a serious complication of diabetes that can cause permanent vision loss if not detected early. It is caused by damage to retinal blood vessels due to prolonged high blood sugar and can progress from mild symptoms (like microaneurysms and hemorrhages) to severe stages (Proliferative Diabetic Retinopathy), which may lead to blindness. Because early DR is often asymptomatic, regular screening is essential, but manual diagnosis by ophthalmologists is time-consuming and inconsistent.

To address this, the study focuses on deep learning (DL) and machine learning (ML) methods, especially Convolutional Neural Networks (CNNs) and transfer learning models like DenseNet, ResNet, and InceptionV3. These models can automatically extract features from retinal images and classify DR stages with high accuracy (often 95–99%). The goal is to build an automated, efficient, and explainable AI system to assist doctors, improve early detection, and reduce workload.

The literature review shows that many existing studies use different AI approaches for DR detection:

• CNN-based and hybrid models generally perform well in classification tasks.
• Advanced models (DenseNet, InceptionV3, RetNet, etc.) achieve high accuracy but often face issues like class imbalance, limited generalization, or high computational cost.
• Some models lack proper comparison, robustness testing, or handling of real-world variability.
• Explainability techniques like Grad-CAM are used, but interpretability is often limited.

Conclusion

This study presents a robust and automated deep learning-based framework for the early detection and classification of Diabetic Retinopathy using retinal fundus images. The proposed system effectively addresses the limitations of traditional manual screening approaches by providing a scalable, accurate, and time-efficient solution. The integration of Convolutional Neural Networks with transfer learning enables effective feature extraction and high classification performance across multiple DR stages. Additionally, the incorporation of explainable AI techniques, such as Grad-CAM, enhances model transparency by highlighting clinically relevant regions, thereby improving trust and interpretability for medical practitioners. The experimental results demonstrate strong performance across key evaluation metrics, including accuracy, precision, recall, and F1-score, confirming the model’s reliability and generalization capability. Moreover, the modular architecture of the system allows seamless integration with healthcare infrastructures such as EHRs and tele-ophthalmology platforms. Overall, the proposed approach provides a cost-effective and scalable solution for early DR detection, with significant potential to improve patient outcomes and support the advancement of intelligent healthcare systems, particularly in underserved regions.

References

[1] P. Yadav and N. P. Singh, \"Classification of normal and abnormal retinal images using feature-based machine learning approach,\" in *Recent Trends in Communication, Computing, and Electronics*, Springer, 2019, pp. 387–396. [2] P. Saeedi et al., \"Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045,\" *Diabetes Research and Clinical Practice*, vol. 157, 2019. [3] P. Aschner et al., \"The international diabetes federation’s guide for diabetes epidemiological studies,\" *Diabetes Research and Clinical Practice*, vol. 172, 2021. [4] International Diabetes Federation, \"IDF Diabetes Atlas, 10th ed.,\" 2021. [Online]. Available: https://diabetesatlas.org/atlas/tenth-edition/ [5] T. Y. Wong and C. Sabanayagam, \"Strategies to tackle the global burden of diabetic retinopathy: From epidemiology to artificial intelligence,\" *Ophthalmologica*, vol. 243, no. 1, pp. 9–20, 2020. [6] P. P. Choo et al., \"Management of sight-threatening diabetic retinopathy during pregnancy,\" *World Journal of Diabetes*, vol. 12, no. 9, pp. 1386, 2021. [7] E. A. Lundeen et al., \"Prevalence of diabetic retinopathy in the US in 2021,\" *JAMA Ophthalmology*, vol. 141, no. 8, pp. 747–754, 2023. [8] Z. L. Teo et al., \"Global prevalence of diabetic retinopathy and projection of burden through 2045,\" *Ophthalmology*, vol. 128, no. 11, pp. 1580–1591, 2021. [9] P. Salpea et al., \"Call for data contribution to the 11th edition of IDF’s Diabetes Atlas,\" *Diabetes Research and Clinical Practice*, vol. 198, 2023. [10] S. Bhandari et al., \"Early-stage diabetic retinopathy detection using deep learning: A review,\" *Archives of Computational Methods in Engineering*, vol. 30, no. 2, pp. 799–810, 2023. [11] N. M. A. Tajudin et al., \"Deep learning in the grading of diabetic retinopathy: A review,\" *IET Computer Vision*, vol. 16, no. 8, pp. 667–682, 2022. [12] A. M. Dayana and W. R. S. Emmanuel, \"A comprehensive review of diabetic retinopathy detection using deep learning,\" *Archives of Computational Methods in Engineering*, vol. 30, no. 7, pp. 4565–4599, 2023. [13] H. Shakibania et al., \"Dual branch deep learning network for diabetic retinopathy detection,\" *Biomedical Signal Processing and Control*, vol. 93, 2024. [14] D. Muthusamy and P. Palani, \"Deep learning models for diabetic retinopathy detection: An overview,\" *Artificial Intelligence Review*, vol. 57, 2024. [15] S. Akhtar et al., \"Diabetic retinopathy severity grading using transfer learning,\" *International Journal of Engineering and Manufacturing*, vol. 14, 2024. [16] L. Arora et al., \"Ensemble deep learning and EfficientNet for DR diagnosis,\" *Scientific Reports*, vol. 14, 2024. [17] M. Youldash et al., \"Early detection of diabetic retinopathy using deep learning,\" *AI*, vol. 5, no. 4, 2024. [18] J. Yao et al., \"Artificial intelligence algorithms for DR and macular edema detection,\" *Eye and Vision*, vol. 11, 2024. [19] M. S. Harisha et al., \"Deep learning-based segmentation and classification of DR,\" *Artificial Intelligence in Health*, vol. 1, 2024. [20] G. Alwakid et al., \"Deep learning-enhanced DR classification,\" *Digital Health*, vol. 9, 2023. [21] X. Xu et al., \"Computer-aided diagnosis of DR using multi-view learning,\" *Computers in Biology and Medicine*, vol. 174, 2024. [22] A. M. Moustari et al., \"Two-stage deep learning classification for DR using Grad-CAM,\" *Automatika*, vol. 65, no. 3, pp. 1284–1299, 2024. [23] G. Saxena et al., \"Robust deep learning agent for DR detection,\" *Intelligence-Based Medicine*, vol. 3, 2020. [24] S. Yi and Z. Chen, \"Medical image dataset cleaning framework,\" *Heliyon*, vol. 10, 2024. [25] N. Mukherjee and S. Sengupta, \"Deep feature extraction strategies for DR classification,\" *Arabian Journal for Science and Engineering*, vol. 48, 2023. [26] M. Tariq et al., \"Transfer learning-based classification of DR,\" in *Medical Imaging Conference*, Springer, 2022. [27] G. Alwakid et al., \"Enhancement of DR detection using CLAHE and ESRGAN,\" *Diagnostics*, vol. 13, 2023. [28] C. Mohanty et al., \"Deep learning architectures for DR detection,\" *Sensors*, vol. 23, 2023. [29] W. K. Wong et al., \"Transfer learning-based DR detection using ensemble models,\" *IEEE Access*, vol. 11, 2023. [30] A. M. Mutawa et al., \"Transfer learning for DR detection: Dataset combination study,\" *Applied Sciences*, vol. 13, 2023. [31] M. N. Nahiduzzaman et al., \"Parallel CNN-based DR classification,\" *Expert Systems with Applications*, vol. 217, 2023. [32] M. Murugappan et al., \"Few-shot classification framework for DR detection,\" *Measurement*, vol. 200, 2022. [33] E. Abdel Maksoud et al., \"Hybrid deep learning system for DR detection,\" *Medical & Biological Engineering & Computing*, vol. 60, no. 7, 2022 [34] A. Bilal et al., \"Automatic DR detection using U-Net and deep learning,\" *Symmetry*, vol. 14, 2022. [35] A. R. Beham and V. Thanikaiselvan, \"Optimized deep learning for DR detection,\" *Soft Computing*, 2023. [36] M. M. Butt et al., \"Hybrid deep learning features for DR detection,\" *Diagnostics*, vol. 12, 2022.

Copyright

Copyright © 2026 Sameer Gangurde, Tejas Shirsath, Abhishek Sabale, Mrs. Swati Rajput. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.