Diabetic Retinopathy Diagnosis using VGG-19 Deep Learning Model

Diabetic retinopathy (DR) ranks as the leading cause of eye problems and blindness among people with diabetes globally. Early identification and accurate classification of DR will help to considerably slow down the development of DR and thereby determine the timeliness of intervention and treatment. This work, notably with reference to the VGG19 convolutional neural network architecture, targets the automatic categorization of diabetic retinopathy from retinal images using deep learning approaches. A collection of retinal images that have been categorized as DR and non-DR (No-Dris) was employed in the study to leverage transfer learning to improve the pre-trained VGG19 model. Performance criteria include accuracy, precision, recall, and the confusion matrix help in assessing the model’s effectiveness. Results with a test accuracy of $94.37 \%$ reveal the tremendous capacity of the model to distinguish between DR and No-DR situations. The results show the possibilities of deep learning for supporting medical practitioners with early identification and management of diabetic retinopathy, therefore indicating opportunities for further development and use in clinical environments to improve patient outcomes.