Diabetic Retinopathy Detection on Retinal Fundus Images Using Convolutional Neural Network

Abstract

Diabetic retinopathy is a serious eye disease that occurs due to diabetes mellitus, also commonly known as diabetes, and it has grown as the most common cause of blindness in the present world. It is a disease in which the blood vessels behind the retina are damaged. At first, it shows no symptoms, but with time, it eventually leads to blindness. Early diagnosis of diabetic retinopathy can prevent vision loss in patients. The method proposed here for the detection of diabetic retinopathy disease is based on a convolutional neural network that categorizes the fundus images of patients according to the severity level of diabetic retinopathy. The input images are collected from the Kaggle diabetic retinopathy dataset, and various preprocessing steps such as cropping, resizing, grayscaling, CLAHE and Min-Max normalization are performed. Precision, recall and Kappa score values of our model are highly affected, due to the unequal distribution of dataset.