Multiclass classification of Diabetic Retinopathy using ResNet50 transfer learning model

Diabetic retinopathy (DR) is the fundamental cause of permanent eyesight loss among working-age individuals. The primary cause of this condition is the difficulties of diabetes affecting the blood vessels and lesions in the retina. Deep learning (DL) has gained attention for classifying retinal fundus images to diagnose DR. It addresses challenges like high costs, time consumption, and errors in manual diagnosis. Consequently, a computer-based, automatic, and robust system for diagnosing DR is essential in the early stage. This study uses a ResNet50 transfer learning model to classify DR based on severity levels, achieving higher accuracy and precision than existing methods. The proposed model works effectively with smaller datasets, enabling faster training and requiring fewer computational resources. By fine-tuning the network, the approach enhances the evolution of DR severity, achieving 96% accuracy and other performance metrics like precision, recall, and F1-score.