Abstract: Diabetes is one of the increasing serious health issues, affecting millions of people. Consequently, the greater portion of individual is affected by diabetes, some of the related issues such as cardiovascular disease, kidney failure, nerve damage, foot problems and vision problems are growing rapidly. Diabetic retinopathy (DR) is a severe complication of diabetes and a leading cause of preventable blindness worldwide. Early detection requires expert evaluation of retinal fundus images, but limited access to specialists delays diagnosis. This work presents an automated DR screening pipeline using pertained Convolutional Neural Networks (CNNs), including VGG16, VGG19, and ResNet50, for five-class DR severity classification. Using 3,662 Kaggle fundus images with systematic preprocessing, augmentation, and class balancing, the proposed approach achieves a highest accuracy of 96.3% and F1-score of 0.898. The model supports improved early diagnosis by enabling scalable, automated retinal image screeningto enhance primary diagnosis and minimizes blindness risk.

Keywords: Diabetic retinopathy, eye fundus images, Classification, CNN, VGG16, VGG19, ResNet50 component

Pages: 72-79

WSEAS Transactions on