Implementation of deep learning based algorithms for diabetic retinopathy classification from fundus images

Diabetic retinopathy damages the blood vessels in retina. It is a disease that cause vision loss. It is usually seen in two categories of people one is the diabetic patients and other is the working age adult population.No symptoms can be seen during the early stage of this disease, so it progresses and loss of vision occurs. This problem can be solved only by the early detection of disease and its timely treatment. A deep learning model to speed up disease detection is presented in this paper. The datasets consists of thousands of retina images taken using fundus photography under a variety of imaging conditions was assessed from the Kaggle Diabetic Retinopathy Detection database. Gaussian Blur Subtraction and data augmentation were done as a part of image preprocessing. The model were trained using the architectures Resnet,Densenet etc. After testing and retraining the model will be used to develop the system that have a user friendly interference. The model will be able to localize object, detect object and segment the capillaries present in a retina image. The abnormalities in the capillaries are noted to detect blindness in diabetic people and will classify the retinopathy stages as No DR, Mild, Moderate,Severe and Proliferative DR4.