Clinically Explainable Diabetic Retinopathy Detection with VGG Features, SVM, and GAN

Diabetic retinopathy (DR) is very common among the world and it is the most common cause of preventable vision loss. Prompt and precise screening is vital for stopping sight threats. The use of artificial intelligence to enhance the grading of diabetic retinopathy will be examined. A hybrid VGG-SVM method is tested. The image features are extracted with the help of VGG networks and a support vector machine does the final classification. This setup aims to raise diagnostic accuracy. To enhance performance and mitigate class imbalance, datasets can be increased using a GAN approach. As the clarity of the model is important, LIME is used to show which retinal spots matter the most for predictions. It shines light on certain eye abnormalities. This makes it easier for doctors to trust the system. The method was trained on large retinal datasets. It reached 96% accuracy, precision, and F1-score. Our outcome beats strong CNN and earlier top-performing methods. Research has shown that combining strategies like deep feature learning, margin-based classification, and focused data growth can help achieve high precision along with clear, accurate results for diabetic retinopathy screening. We plan to conduct more tests in various clinics, adapt to new devices, and connect to actual screening systems. The objective is make it perform optimally for a variety of groups and imaging setups.