Diabetic Retinopathy Classification using a Combination of EfficientNets

Diabetic Retinopathy (DR) is a diabetes complication that affects vision. It is caused by damage to the blood vessels of retina. Early and accurate detection of DR is crucial to reduce likelihood of progression to proliferative retinopathy and blindness. This paper proposes a method for classifying the severity of DR using deep learning. Experiments were conducted by blending the members of EfficientNet for classification of the diabetic retinopathy image as no DR, mild, moderate, severe, or proliferative DR. The models have been trained using different datasets and best model achieved a quadratic kappa score of 0.924377 on the APTOS test dataset. The results are promising and warrant further investigation. The presented model has the potential aid in fast diagnosis for better early detection of DR.