VGG-Powered Convolutional Neural Networks in Diabetic Retinopathy Classification: A Comparative Investigation

Diabetic Retinopathy (DR) is a severe and debilitating complication associated with diabetes, often leading to visual impairment in affected adults. Its primary cause is the adverse impact of prolonged hyperglycemia on the retinal vasculature. DR is categorized into Non-Proliferative Diabetic Retinopathy (NPDR) and Proliferative Diabetic Retinopathy (PDR), with early detection and accurate classification being vital, especially given the rising global incidence of Diabetes Mellitus (DM). Effective DR diagnosis involves the identification of retinal lesions and their severity classification. As diabetes cases increase worldwide, there’s a growing need for efficient diagnostic methods. Automated systems, employing deep learning techniques show promise for DR detection and classification. However, manual grading of DR remains time-consuming and expensive. Computer-aided diagnosis (CAD) systems offer a more accessible and efficient alternative. In this paper, we introduce a novel approach using Convolutional Neural Network (CNN) models based on the VGG architecture with transfer learning to address the pressing healthcare challenges posed by DR. Notably, our approach achieves an accuracy rate of 73.01% during transfer learning, which surges to 75.83% after fine-tuning, particularly when classifying Diabetic Retinopathy in our test dataset. VGG16, demonstrates the effectiveness of fine-tuning with a notable increase in accuracy. Future research opportunities include exploring alternative CNN model architectures like ResNet, Inception, and ConvNeXt, as well as optimizing performance through ensemble methods for further enhancements.