An Automatic Diabetic Retinal Disease Classification Model Using Hybrid Convolutional Neural Networks Based on Fundus Images

Abstract

Diabetic Retinopathy (DR) is a severe complication of diabetes that can cause retinal damage and lead to blindness if not properly managed. Traditional DR screening, conducted by ophthalmologists, is time-consuming, highlighting the need for more efficient detection methods. This project leverages Deep Learning (DL) to automate DR detection using Mobile Net, a lightweight model trained on 3,662 high-resolution fundus images from the APTOS 2019 dataset. The Mobile Net model is designed to classify DR into five stages, ranging from zero to four. To further enhance performance, we investigate hybrid models that combine Mobile Net with other architectures: Mobile Net + Graph Neural Network (GNN) and Mobile Net + Recurrent Neural Network (RNN). The Mobile Net + GNN hybrid aims to capture complex spatial relationships within retinal images, potentially improving classification accuracy by interpreting structural patterns. The Mobile Net + RNN hybrid focuses on addressing sequential dependencies and temporal patterns, thereby enhancing the model\'s capability to detect subtle changes across different DR stages. Our results demonstrate that the Mobile Net + GNN model achieved an accuracy of 92%, while the Mobile Net + RNN model reached an accuracy of 81%. These findings suggest that the hybrid approaches offer a promising path to optimize DR detection, improving both accuracy and efficiency compared to traditional methods.

Introduction

Diabetic Retinopathy (DR) is a serious complication of diabetes and one of the leading causes of blindness worldwide. It occurs when prolonged high blood sugar damages the retinal blood vessels, causing leakage, bleeding, and progressive vision loss. DR develops through five stages, from no diabetic retinopathy to proliferative diabetic retinopathy (PDR). Early detection is essential to prevent irreversible blindness, but conventional diagnosis relies on manual examination by ophthalmologists, making the process time-consuming, expensive, and prone to human error.

Recent advances in artificial intelligence (AI) and deep learning (DL) have enabled automated analysis of retinal fundus images for accurate DR detection. This project proposes a lightweight and efficient deep learning-based screening system using MobileNet, trained on the APTOS Kaggle dataset containing 3,662 labeled retinal images across five DR severity stages. To further improve classification performance, the study integrates MobileNet with Graph Neural Networks (GNN) to capture spatial relationships within retinal images and MobileNet with Recurrent Neural Networks (RNN) to model sequential dependencies.

The literature review demonstrates that deep learning techniques, particularly convolutional neural networks (CNNs), have significantly improved diabetic retinopathy detection compared to traditional image-processing methods. Previous studies achieved high sensitivity and accuracy using architectures such as GoogLeNet, AlexNet, ResNet, Xception, and MobileNet, while transfer learning, data preprocessing, and feature aggregation further enhanced performance. However, existing systems often require high computational resources and have difficulty detecting subtle early-stage retinal abnormalities.

The proposed methodology begins with collecting anonymized retinal images from Kaggle and public medical databases. Images are preprocessed by resizing them to 224 × 224 pixels, normalizing pixel values, and converting labels into numerical form. Data augmentation techniques—including random rotation, zooming, and horizontal flipping—are applied to improve model robustness. The MobileNet, MobileNet+GNN, and MobileNet+RNN models are trained using the Adam optimizer, binary cross-entropy loss, a batch size of 20, and 15 training epochs. The dataset is divided into 70% training, 15% validation, and 15% testing, and model performance is evaluated using accuracy, precision, recall, and F1-score.

Experimental results show that the MobileNet+GNN model achieved the best overall performance, with 92% accuracy, 90% precision, 86% recall, and an 86% F1-score, outperforming both the standalone MobileNet and MobileNet+RNN models. The system also provides an easy-to-use interface for uploading retinal images and automatically predicting the stage of diabetic retinopathy. Overall, the proposed hybrid MobileNet-GNN approach offers a highly accurate, computationally efficient, and practical solution for automated diabetic retinopathy screening, supporting early diagnosis and reducing the workload of ophthalmologists.

Conclusion

In conclusion, the application of deep learning in detecting diabetic retinopathy represents a significant advancement in the field of medical diagnostics. This innovative approach has shown promising results in automating the early detection of diabetic retinopathy, thereby enabling timely interventions and reducing the risk of vision loss in diabetic patients. With further refinement and integration into healthcare systems, deep learning models can contribute to more efficient and accessible screening, ultimately improving the quality of life for individuals with diabetes. In this Project we are using Mobile Net Architecture, Mobile Net along with GNN and Mobile Net with RNN in this GNN hybrid model is giving Best accuracy compared to others.

References

[1] VarunGulshan, Subhashini Venugopalan, Rajiv Raman, “Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs,” JAMA. 2016;316(22):24022410. doi:10.1001/jama.2016.17216. [2] Kele Xu, Dawei Feng, and Haibo Mi, “Deep Convolutional Neural Network-Based Early Automated Detection of Diabetic Retinopathy Using Fundus Image,” Received: 10 November 2017; Accepted: 22 November 2017; Published: 23 November 2017. [3] SheikhMuhammad Saiful Islam, Md Mahedi Hasan, and Sohaib Abdullah, “Deep Learning based Early Detection and Grading of Diabetic Retinopathy Using Retinal Fundus Images,” arXiv:1812.10595v1 [cs.CV] 27 Dec 2018. [4] Lam C, Yi D, Guo M, Lindsey T., “Automated Detection of Diabetic Retinopathy using Deep Learning,” AMIA Jt Summits Transl Sci Proc. 2018 May 18;2017:147-155. PMID: 29888061; PMCID: PMC5961805. [5] SaraHosseinzadeh Kassani, Peyman Hosseinzadeh Kassani, Reza Khazaeinezhad, Michal J. Wesolowski et al. \"Diabetic Retinopathy Classification Using a Modified Xception Architecture\", 2019 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT), 2019 [6] Anuj Jain, Arnav Jalui, Jahanvi Jasani, Yash Lahoti, Ruhina Karani. \"Deep Learning for Detection and Severity Classification of Diabetic Retinopathy\", 2019 1st International Conference on Innovations in Information and Communication Technology (ICIICT), 2019 [7] R Borys Tymchenko, Philip Marchenko and Dmitry Spodarets, “Deep Learning Approach to Diabetic Retinopathy Detection”. [8] Weiguo Fan, Edward A. Chandan K. Reddy, “A Deep Learning Based Pipeline for Image Grading of Diabetic Retinopathy”. [9] Eswar Kumar Kilari, Swathi Putta. \" Delayed progression of diabetic cataractogenesis and retinopathy by in STZ-induced diabetic rats \", Cutaneous and Ocular Toxicology, 2016 [10] N. Yalin, S. Alver and N. Uluhatun, ”Classi?cation of retinal images with deep learning for early detection of diabetic retinopathy disease,” 2018 26th Signal Processing and Communications Applications Conference (SIU), Izmir, 2018, pp. 520. [11] S. H. Kassani, P. H. Kassani, M. J. Wesolowski, K. A. Schneider, and R. Deters, ““Breast cancer diagnosis with transfer learning and global pooling,” arXiv preprint arXiv:1909.11839, 2019.

Copyright

Copyright © 2026 Utchila Gopi , Bindu Priya. 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.