A Deep Learning Approach in Detecting Diabetic Retinopathy Using Convolutional Neural Network on Gaussian Filtered Retina Scanned images

Abstract

Diabetic retinopathy is a diabetes complication that affects eyes. It is caused by damage to the blood vessels of the light-sensitive tissue at the back of the eye (retina). At first, diabetic retinopathy may cause no symptoms or only mild vision problems however, it can cause blindness. The condition can develop in anyone who has type 1 or type 2 diabetes. It may lead to poor vision and subsequently to complete blindness. This paper presents a Deep Learning approach in detecting Diabetic Retinopathy on Gaussian Filtered Retina Scanned images. We used a Gaussian filtered scan retina image dataset which was downloaded from kaggle.com. This dataset contains five image folders which are Mild folder that contains 370 images of patients with lesser risk to Diabetic Retinopathy (early stage), Moderate Folder contains 999 images of patients having 12%-27% risk of Diabetic Retinopathy, the Severe Folder contains 193 images of patients whose blood vessels have become more blocked, the Proliferate Folder contains 295 images of patients which are on the verge of going on a permanent blindness, the last folder is the No Diabetic Retinopathy folder which contains 1805 images of patients who have no Diabetic Retinopathy. After building and training our convolutional neural network model, the results obtain by the model shows an accuracy of 81.35% at an epoch number of 8. The trained model was saved and tested using flask framework. This model can be deployed to web for detecting and classifying the various categories of diabetic retinopathy.