Survey on automated detection of referable Diabetic Retinopathy using machine learning

Abstract

Diabetic retinopathy is a complication of diabetes, that results in the rupture of the blood vessels in the light-sensitive region of the eye also known as the retina. This situation can occur in a person who has either type 1 or type 2 diabetes. Excess sugar levels in a diabetic person block the blood vessels that nourish the retina of the eye. So as to compensate this shortcoming the retina grows new blood vessels, however, this new blood vessel can rupture easily. The traditional means to detect diabetic retinopathy is to undergo regular screening and then to consult a doctor. This is a significant time-consuming task as there is a shortage of experienced ophthalmologist, as a result, 45% of the patient suffer from vision loss even before they are diagnosed. Another major problem associated with this method is that there is significant inconsistency among doctors who diagnose diabetic retinopathy, as a result, there are chances that diabetic retinopathy can go undetected at its early stage. The automated detection involves training a machine learning model that can detect new cases of diabetic retinopathy from retinal fundus images which have been graded by experienced ophthalmologists. The decision for predicting the degree of diabetic retinopathy has been done using machine learning algorithms such as deep convolution network, SVM and Naïve Bayes.