Deep Learning for Diabetic Retinopathy Detection from Retinal Images

Abstract

Diabetic retinopathy (DR) is a progressive microvascular complication of diabetes and is today consid- ered a major cause of avoidable blindness on a global scale. This should be detected at a young age, but the conventional approach to screening uses manual interpretation of the retina fundus images, which is laborious, subjective, and hard to automatically do on a large population. The proposed project presents an automated deep learning model based on the use of sophisticated convo- lutional neural networks (CNNs) to identify stage of DR precisely using the retinal images. It starts with the creation of a heterogeneous dataset and proceeds to a powerful preprocessing step comprised of normalization, contrast enhancement, image re-scaling, and illumination removal to enhance the appearance of lesions. More data augmentation methods including rotation, flipping, zooming, noise infusion also improve the generalization of the models and deal with the issue of class imbalance. A discriminative retinal feature extractor with a small CNN architecture, built on the latest models, like EfficientNet or ResNet, is trained to recognize images according to various levels of severity of DR. The diagnostic reliability is evaluated with rigor by accuracy, precision, recall, F1-score, sensitivity, specificity, and ROCAUC on model performance to be sure that the model performance is reliable. The resulting trained model is embedded into a user-friendly web-based application which enables clinicians to provide retinal images and get quick and evidence-based predictions with heatmap visualizations to read between the lines. This system provides better screening of DR before symptoms, encourages diagnosing ophthalmology at scale, and shows a promising influence of deep learning on medical image processing.

Introduction

The text provides a detailed review of Diabetic Retinopathy (DR), its challenges, and the role of deep learning in improving early detection through retinal image analysis.

DR is a serious complication of diabetes and a major cause of preventable blindness. Since early stages are often symptomless, timely detection is crucial. Traditional screening using manual fundus image analysis is slow, expensive, and prone to errors, especially in regions with limited medical resources. This has led to growing interest in automated, AI-based diagnostic systems.

Recent advances in deep learning (especially CNNs like AlexNet, ResNet, DenseNet, and EfficientNet) have significantly improved DR detection by learning complex retinal features such as microaneurysms and hemorrhages. Studies like those by Gulshan et al. and Ting et al. show that AI systems can achieve high accuracy and even perform well in real-world clinical screening.

Research has evolved from simple binary classification to more advanced tasks such as multi-class grading, lesion detection, disease progression modeling, and real-world deployment. Large-scale systems like DeepDR combine image quality assessment and severity grading, showing that AI can support clinical decision-making effectively.

However, challenges remain, including data imbalance, poor generalization across datasets, lack of interpretability (black-box models), and high computational cost. To address these issues, researchers explore techniques like transfer learning, hybrid models, data augmentation, federated learning, and lightweight architectures.

EfficientNet, especially EfficientNetB3, is highlighted as a strong model due to its balance of accuracy and efficiency, making it suitable for medical applications where subtle features must be detected with limited computing resources. Explainable AI techniques like Grad-CAM are also important for visualizing model decisions and improving clinical trust.

The proposed work focuses on building an automated DR detection system using EfficientNetB3 on the APTOS 2019 dataset, combined with preprocessing, augmentation, and Grad-CAM visualization. The goal is to create a system that is accurate, efficient, scalable, and interpretable for real-world clinical use.

Conclusion

In this paper, an EfficientNetB3 deep learning model was proposed. to diagnose and categorize diabetic retinopathy in retinal fundus. sample images of the APTOS 2019 Blindness Detection dataset. The suggested system integrated image pre- processing, data transfer learning and Grad-CAM-based inter augmentation. pretability in order to introduce an effective and trustworthy multi-class. severe diabetic retinopathy severity classification model assess. ment. The experimental findings showed that the proposed model has been shown to do well in terms of classification accuracy, having an overall accuracy at 91At 0.92, macro and weighted F1-scores at 0.92. The model demon had a high performance in the classification of No DR and Severe. DR categories, as well as retaining a high level of success in the remaining phases of diabetic retinopathy. Despite the fact that the precision of the Moderate DR class was rather low due to the following reason the levels of severity were visually similar to each other, the model was able to recall many cases of the category, which means that it was able to capture the vast majority of clinically meaningful cases.

References

[1] V. Gulshan et al., “Development and validation of a deep learning algo- rithm for detection of diabetic retinopathy in retinal fundus photographs,” JAMA, vol. 316, no. 22, pp. 2402–2410, 2016. [2] M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convo- lutional neural networks,” in Proc. International Conference on Machine Learning (ICML), 2019. [3] APTOS 2019 Blindness Detection Dataset, Kaggle, 2019. [Online]. Available: https://www.kaggle.com/competitions/ aptos2019-blindness-detection [4] R. R. Selvaraju et al., “Grad-CAM: Visual explanations from deep networks via gradient-based localization,” in Proc. IEEE International Conference on Computer Vision (ICCV), 2017, pp. 618–626. [5] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” in Advances in Neural Informa- tion Processing Systems (NeurIPS), 2012. [6] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778. [7] G. Huang, Z. Liu, L. van der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” in Proc. IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), 2017, pp. 4700–4708. [8] T. Voets, K. Møllersen, and L. A. Bongo, “Reproduction study using public data of: Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs,” PLOS ONE, vol. 14, no. 6, 2019. [9] D. S. W. Ting et al., “Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes,” JAMA, vol. 318, no. 22, pp. 2211–2223, 2017. [10] “Deep learning frameworks for diabetic retinopathy detection with smartphone-based retinal imaging systems,” Pattern Recognition Letters, vol. 137, pp. 409–417, 2020. [11] “Diabetic retinopathy detection through deep learning,” MethodsX, 2020. [12] “A deep learning system for detecting diabetic retinopathy across the disease spectrum,” Nature Communications, vol. 12, no. 1, 2021. [13] “Deep learning for diabetic retinopathy detection and classification based on fundus images: A review,” Computers in Biology and Medicine, 2021. [14] “Deep learning for the detection and classification of diabetic retinopathy with an improved activation function,” Healthcare, vol. 11, no. 1, 2022. [15] “Diabetic retinopathy detection from fundus images of the human retina using deep learning,” 2022. [16] “Predicting of diabetic retinopathy development stages of fundus images using deep learning based on combined features,” PLOS ONE, 2023. [17] “Deep learning and medical image processing techniques for diabetic retinopathy: A survey of applications, challenges, and future trends,” 2023. [18] “A new hybrid deep learning model for diabetic retinopathy detection,” 2023. [19] “A systematic review on diabetic retinopathy detection and classification based on deep learning techniques using fundus images,” PeerJ Computer Science, 2024. [20] “Comprehensive survey of the state-of-the-art deep learning models for diabetic retinopathy detection and grading using retinal fundus photogra- phy,” 2024. [21] “Diabetic retinopathy detection using deep learning with optimized feature selection,” 2024. [22] “A deep learning system for predicting time to progression of diabetic retinopathy,” Nature Medicine, 2024. [23] “Enhancing diabetic retinopathy diagnosis: A lightweight CNN for efficient classification,” arXiv preprint, 2024. [24] “Deep learning-based detection of referable diabetic retinopathy and diabetic macular edema in ultra-widefield fundus images,” arXiv preprint/ MICCAI Challenge paper, 2024. [25] “Systematic review and meta-analysis of regulator-approved deep learn- ing systems for fundus diabetic retinopathy detections,” npj Digital Medicine, 2025. [26] “Diabetic retinopathy screening using machine learning,” 2025. [27] “Deep learning-based optical coherence tomography and retinal imaging for diabetic retinopathy: A systematic review and meta-analysis,” 2025. [28] “New accurate deep learning model for diabetic retinopathy detection utilizing sequential pre-processing and transfer learning,” Biomedical Sig- nal Processing and Control, 2025. [29] “A hybrid deep learning framework for early detection of diabetic retinopathy using retinal fundus images,” Scientific Reports, 2025. [30] “Detection and diagnosis of diabetic retinopathy in retinal fundus images using agentic AI approaches,” Scientific Reports, 2025. [31] “Deep learning-based joint analysis of diabetic retinopathy and glau- coma in retinal fundus images,” Scientific Reports, 2025. [32] “A novel deep learning model for diabetic retinopathy detection,” 2025. [33] “Enhancing diabetic retinopathy classification accuracy using deep learn- ing,” arXiv preprint, 2025. [34] “Development and validation of a deep learning model for early diabetic retinopathy detection,” 2025. [35] “Performance of a deep learning diabetic retinopathy algorithm in India,” JAMA Network Open, 2025. [36] “Deep learning for enhanced prediction of diabetic retinopathy: A comparative study on the diabetes complications data set,” Frontiers in Medicine, 2025. [37] “Early prediction of diabetic retinopathy using a multimodal deep learning model,” 2026.

Copyright

Copyright © 2026 Maku Akhila, Maanas Kaniti, Induri Aishwarya, Dr. S. Sreekanth . 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.