Classification of retinal images in stages of diabetic retinopathy using deep learning

Diabetic retinopathy (DR) is an advanced stage of retinal disease with increasing prevalence and the major cause of blindness. In DR, blood vessels get damaged over time at the back of the retina. Large numbers of retinal images are generated as diabetes patients all over the world and it is increasing every year. This increases the workload of ophthalmologists which may result in delayed diagnosis and treatment. In this paper, automatic classification of normal eye and DR eye using convolutional neural network (CNN) is presented. This computerized system also further classifies DR eye in one of the four stages as Microaneurysms, Hemorrhages, Hard Exudates, and Soft Exudates (Cotton Wool Spots). The framework is trained and validated on 413 images and tested on 187 images of IDRiD database with 98.930% average classification accuracy. The system also achieves average 97.316% of sensitivity and 99.338% of specificity. This system will assist ophthalmologists to screen retinal images for abnormality from a large database within reduced time span which may leads to timely treatment to the patient.