Automatic detection of non-proliferative diabetic retinopathy in retinal fundus images using convolution neural network

Abstract

Diabetic retinopathy (DR) is one of the complications of diabetes and a leading cause of blindness in the world. The tiny blood vessels inside the retina are damaged due to diabetes and result in various vision-related problems and it may lead to complete vision loss without early detection and treatment. Diabetic retinopathy may not cause any symptoms during its earlier stage of the disease and many physical tests such as visual acuity tests, pupil dilation, etc., are required to detect diabetic retinopathy disease. So, early detection of diabetic retinopathy disease is required to avoid vision loss. This work aims to automate the detection and grading of non-proliferative Diabetic Retinopathy from retinal fundus images using Convolution Neural Networks. The model was tested on two popular datasets such as MESSIDOR and IDRiD. Before applying the Convolution Neural Network (CNN) layers, the images were pre-processed and resolution was adjusted (256 × 256). The maximum accuracy achieved is 90.89% using MESSIDOR images. The research can be carried forward by applying various preprocessing techniques before putting them through different computational layers.