Early Detection of Diabetic Retinopathy from Retinal Fundus Images Using Deep Learning Models

This paper presented an end-to-end deep learning framework for early detection and severity grading of Diabetic Retinopathy from retinal fundus images. Five models—a custom CNN, VGG-16, ResNet-50, EfficientNet-B3 and ViT-B/16—were trained and compared on the APTOS-2019 and EyePACS datasets. The Vision Transformer achieved the highest accuracy of 95.6 % and AUC of 97.7 %, while EfficientNet-B3 offered the best accuracy-to-parameter trade-off. CLAHE pre-processing, aggressive augmentation and transfer learning were shown to be essential components of the pipeline through a quantitative ablation study.   Future research directions include: ·        Extending the framework to multi-disease classification (DR, glaucoma, age-related macular degeneration). ·        Incorporating Grad-CAM and other explainable-AI methods to provide visual interpretability of model predictions for clinicians. ·        Federated learning across hospitals to leverage larger and more diverse datasets while preserving patient privacy. ·        Self-supervised pre-training on millions of unlabelled fundus images to further reduce dependence on labelled data. ·        Deployment of a lightweight EfficientNet-B0 / MobileViT variant on smartphones for low-cost field screening in remote regions. ·        Integration of multimodal data (Optical Coherence Tomography, patient metadata) to improve diagnostic robustness. ·        Real-time tele-ophthalmology platforms connected to nationwide diabetic registries.

[1]  International Diabetes Federation, IDF Diabetes Atlas, 10th Edition, Brussels, 2021. [2]  World Health Organization, "World Report on Vision," Geneva, 2019. [3]  American Academy of Ophthalmology, "Diabetic Retinopathy Preferred Practice Pattern," Ophthalmology, vol. 127, no. 1, pp. 66–145, 2020. [4]  T. Y. Wong, J. Sun, R. Kawasaki, et al., "Guidelines on Diabetic Eye Care: The International Council of Ophthalmology Recommendations for Screening, Follow-up, Referral, and Treatment Based on Resource Settings," Ophthalmology, vol. 125, no. 10, pp. 1608–1622, 2018. [5]  A. Krizhevsky, I. Sutskever and G. E. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks," NIPS, 2012. [6]  K. Simonyan and A. Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," ICLR, 2015. [7]  K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," CVPR, pp. 770–778, 2016. [8]  C. Szegedy, V. Vanhoucke, S. Ioffe, et al., "Rethinking the Inception Architecture for Computer Vision," CVPR, pp. 2818–2826, 2016. [9]  G. Huang, Z. Liu, L. van der Maaten and K. Q. Weinberger, "Densely Connected Convolutional Networks," CVPR, pp. 4700–4708, 2017. [10]  M. Tan and Q. V. Le, "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks," ICML, 2019.