Diabetic Retinopathy Classification Using Deep Learning Architectures

Diabetes is the main cause for the blindness in the working stage of adults. It causes to blurry vision and loss of eye sight in diabetic patients. Early detection and treatment for this condition leads to prevention of Diabetic Retinopathy (DR). DR is an art and science of recording fundus images on the diabetic patient. Based on the severity of diabetes, DR can be classified into five stages- no DR, mild DR, moderate DR, severe DR, proliferate DR. Before classifying fundus retinal images of diabetic patient, images must be preprocessed. For improving and testing we can use the Kaggle dataset. Our experiments have been performed by using deep learning networks – multilayer perception and convolutional neural network. We report the results based on the accuracy such as validation accuracy and training accuracy. By using different optimizers and different activation functions we report the comparison charts of these two methods – multilayer perception and convolutional neural network