Intelligent Diabetic Retinopathy Detection Using Hybrid Deep Learning and Lesion Analysis

Abstract

Retinal images play an important role in helping ophthalmologists identify different eye diseases at an early stage. Many retinal disorders cause small changes in blood vessels, making early detection essential for timely treatment. In this work, an automated and non-invasive system based on deep learning is developed to diagnose multiple eye conditions using colour fundus images. The RFMiD dataset is used to build the model, where multi-class images are extracted and enhanced using data augmentation techniques to improve performance in real-world scenarios. The images are preprocessed to reduce computational complexity and improve efficiency. A multi-layer deep learning model is designed, where a convolutional neural network extracts key features from the images and uses them for accurate disease prediction. The model’s performance is evaluated using standard metrics, showing better results compared to existing methods, making it effective for reliable eye disease detection.