Implementation of a Deep Learning-Based Approach for Diabetic Retinopathy Diagnosis and Severity Grading Using ResNet-50 Architecture

One of the most prevalent causes of blindness that can be prevented is diabetic retinopathy (DR). Because of this, automated screening tools that can identify early disease indicators in retinal fundus images are crucial. The severity levels of DR may now be correctly classified thanks to recent advancements in deep learning, particularly Convolutional Neural Networks (CNNs). With an emphasis on the ResNet-50 and Retinopathy Severity Grading (RSG-Net) designs, this paper examines the most effective methods for identifying DR. The study demonstrates how DR grading may be made more dependable across several datasets, including the APTOS 2019 Blindness Detection dataset, by preprocessing photos, adding new data, and utilising transfer learning. Several stages of DR, from No DR to Proliferative DR, are examined to determine how well current CNN models can diagnose. The benefits and drawbacks of binary and multi-class classification methods, as well as their applicability in the clinic, are also thoroughly examined in this work. This study examines CNN-based techniques for early detection of DR and discusses how they can support widespread, real-time eye screening.