A Deep Learning Fusion Approach for Automatic Diagnosis and Grading of Diabetic Retinopathy

Diabetic retinopathy, a severe complication arising from diabetes, exerts its impact on the retina. Timely identification plays a pivotal role in facilitating successful treatment. In this regard, this study explores the utilization of 3 state-of-the-art CNN architectures and proposes a fusion based approach to enhance the accuracy of diabetic retinopathy classification. The CNN models, namely EfficientNet-B0, MobileNet, and ResNet50 are explored on a public dataset of retinal images, MESSIDOR. The proposed approach achieves better performance than previous methods. On MESSIDOR, it achieves 82% accuracy for referral/non-referral binary classification and 63% accuracy for multi-class classification.