Sensing Diabetic Retinopathy Using Deep Learning
K. R. Thukaram Rao, Baksam Udaya Keerthana, Sampangi Sri Hari, Mustur Nikitha, Jonnagaddala Vinay Kumar, Bollam Lokesh Yadav
Diabetic Retinopathy (DR) is a serious vision-threatening complication of diabetes, contributing
significantly to global blindness rates. Early detection and accurate classification of DR are essential
for effective treatment and the prevention of vision loss. Traditional diagnostic methods such as manual
inspection of retinal images are time-consuming, labor-intensive, and subject to variability in expert
opinion. Recent advancements in artificial intelligence and deep learning have provided promising
solutions for automating this process. The proposed framework introduces an automated system for
detecting and classifying diabetic retinopathy using Generative Adversarial Networks (GANs) integrated
with Convolutional Neural Networks (CNNs). The system comprises three major stages: pre-processing,
feature extraction, and classification. GANs are utilized for generating high-quality synthetic retinal
images to enhance the dataset and improve model robustness. CNNs are employed to extract deep
features and classify the severity of DR. This method significantly improves detection accuracy and
generalization. Future developments will focus on increasing dataset diversity, optimizing the GAN
architecture, and integrating the system for real-time screening applications in clinical settings.
