Early Detection of Diabetic Retinopathy Using a Fine-Tuned VGG19 Convolutional Neural Network

Among those with diabetes, diabetic retinopathy is the main cause of blindness; thus, early identification is vital to avoid significant vision loss. This paper presents a finely tuned VGG19 convolutional neural network (CNN) model to automatically identify Diabetic Retinopathy from retinal fundus images. Kaggle's dataset used for both training and validation of the model comprised labelled images classified as either "Diabetic Retinopathy" or "No Diabetic Retinopathy". The model was maximised for this specific job by means of fine-tuning pre-trained weights. The model achieved an overall accuracy of 94% by precisely 0.99 for the "No Diabetic Retinopathy" class and a recall of 0.99 for the "Diabetic Retinopathy" class, thereby proving its great performance in identifying both diseases. By means of F1-score, precision, and recall, model performance was assessed, therefore generating a consistent and reliable classification over the complete dataset. The results illustrate how well deep learning models perform in medical image processing when early DR detection has a scalable and accurate tool. This approach can enable clinicians to detect DR early on, hence perhaps reducing the global diabetic-related blindness load.