Leveraging Retinal Images for Detection and Classification of Diabetic Retinopathy Using DenseNet121 and ResNet50

Diabetic retinopathy, which affects up to 80% of patients with diabetes who have the disease for ten years or longer, is the result of diabetic damage to the retina. When it comes to places where diabetic retinopathy diagnosis is most critical, the necessary equipment and knowledge are sometimes inadequate. The majority of research on diabetic retinopathy has focused on manually extracting features or detecting the disease; however, A novel deep neural network based on DenseNet121 and ResNet50 architectures was introduced to predict the severity level of diabetic retinopathy in retinal image scans, with the aim of aiding in early-stage treatment interventions. In this research, deep convolutional neural networks with GPU acceleration are designed and put into practice for autonomously diagnosing and classifying high-resolution retinal pictures into five disease phases based on severity. Training accuracy in DenseNet121 achieved is 98.3% and ResNet50 is 98.47% and Validation Accuracy in DenseNet121 achieved is 96.8% and ResNet50 is 97.1 %. Resnet50 performs slightly better than DenseNet121.