Automated classification of diabetic retinopathy in fundus images using a hybrid CNN

Diabetic retinopathy (DR) is a major vision-threatening complication of diabetes, and early detection is critical to prevent irreversible damage. This research presents an advanced deep learning-based system for early and accurate detection of DR using a hybrid convolutional neural network (CNN) approach. The model integrates multiple CNN architectures—ResNet50 and InceptionV3—to extract complex features from retinal fundus images. The study aims to improve diagnostic performance particularly in the early stages of DR, where clinical signs are often subtle and dispersed, posing challenges to traditional automated methods. Additional architectures, including DenseNet201 and Xception, are incorporated to enhance accuracy, with the proposed model achieving up to 99.7% accuracy. The system is deployed through a user-friendly web interface built with the Flask framework, featuring secure user authentication for data privacy and accessibility. This comprehensive approach not only enables precise classification of DR but also emphasizes timely clinical intervention, ultimately aiming to reduce the risk of vision loss due to this progressive condition.