Project Report on A Machine Learning Approach To Detect Diabetic Retinopathy

Abstract

Diabetic Retinopathy is an eye disease that affects mainly the retina. It is the consequence of long-standing Diabetes which results as blood leaks from the retinal blood vessels onto the retina. It is one of the most leading causes of blindness in today’s world. Therefore early diagnosis and treatment are a must to save people’s eyesight. It is very difficult to identify the disease at each earliest stage since the symptoms do not develop until there is significant damage to the retina. Therefore, regular monitoring of the retina is a must to prevent the disease. A significant advancement in the field of biomedicine has made it possible for early detection and diagnosis of this disease. In this paper, I have proposed a machine learning approach using DenseNet-121 to detect five stages of Diabetic retinopathy namely no-DR, mild-DR, moderate-DR, severe DR, and proliferative-DR with an accuracy of 94%. For the experiment, I have also used some other models for example Inception V3, VGG 19, DenseNet-201, XceptionNet, and traditional CNN to find the best model for my project. I have used 3113 retinal fundus images for the training purpose and 549 images for the validation purpose. A web application was developed to easily check for disease simply by uploading the retinal fundus image. Flask web framework was used for the deployment of the model.