Hybrid Deep Learning Framework For Accurate Diabetic Retinopathy Classification

Abstract

Diabetic Retinopathy (DR) is a severe complication of diabetes that can lead to vision loss if not de- tected and treated early. Traditional diagnosis in- volves manual examination of retinal fundus images by ophthalmologists, which is often time-consuming and prone to subjectivity. This paper presents an automated solution for DR detection and severity clas- sification by employing multiple advanced deep learn- ing models—DenseNet, InceptionV3, ResNet, and Mo- bileNet—trained and evaluated independently. Each model is developed and tested separately to analyze its individual effectiveness in detecting DR and classifying its severity into four categories: Mild, Moderate, Severe, and Proliferative. The approach focuses on utilizing transfer learning to improve performance on limited datasets, while techniques such as SMOTE and focal loss are applied to address class imbalance and enhance prediction accuracy. A web-based interface is developed using Django, enabling easy access for healthcare professionals to test and view predictions from each model. This modular system allows flexible analysis and benchmarking of different CNN architec- tures for DR diagnosis, providing a straightforward yet effective support tool for early detection.