Deep Learning-Based Early Sign Detection Model for Proliferative Diabetic Retinopathy in Neovascularization at the Disc

Abstract

The patients with diabetes for many years are prone to have Diabetic Retinopathy (DR) which is one of the leading causes of blindness. Proliferative Diabetic Retinopathy (PDR) is the advanced stage out of four major progressive stages of DR, where high risk of visual impairments occur. This work shows a deep learning-based automated method for detection of early signs of Proliferative Diabetic Retinopathy at the optic disc area in human retina. Here, we propose the design and implementation of a deep neural network model in replacement of the semi-automated and automated retinal vascular feature extraction methods. Finding the optic disc (OD) center, followed by artery and vein classification from segmented images are essentially important to focus on Neovascularization at the Disc (NVD). A count on the number of major vessels and their width measurement around the OD center are the two indicative parameters for disease diagnosis. Finally, the major vessels are classified as artery and vein sets to differentiate from the newly generated and unwanted blood vessels. This network was trained with the training and testing images of DRIAVE/RITE database for segmentation and artery vein classification. Also, some of our previously published result sets on automated center of optic disc detection on DRIVE dataset have been used to train the model on NVIDIA Titan Xp 8 GB GPU. Finally, the images from MESSIDOR and DIARETDB0 databases were used for testing in detection of Neovascularization at the Disc.