A broad study of machine learning and deep learning techniques for diabetic retinopathy based on feature extraction, detection and classification

Abstract

Diabetic Retinopathy (DR) is a micro vasculardisorder that affects the eyes and is a long term effectofDiabetesmellitus. The likelihood to develop diabetic retinopathy is directly proportional to the age and duration of the diabetes, as well as increase in the level of blood glucose level and fluctuation in blood pressure levels. A person who has diabetes has more probability to develop diabetic retinopathy. The ration of people with diabetes started to increase from 285 million in 2010 and will reach up to 439 million in the year of 2030.Out of the total number of people with Diabetic Retinopathy, approximately one-fourth of the people have vision-threatening disease. Earlier detection and classificationof Diabetic Retinopathy has to be taken much care in order to sustain a patient’s vision. The diabetic Retinopathy may be classified into various stages like Mild non-proliferative retinopathy, Moderate nonproliferative retinopathy, severe nonproliferative Retinopathy and Proliferative diabetic retinopathy. Theproblem associated with the manual detection of diabetic retinopathy is that the processing time is high, effortconsumingandrequiresanophthalmologist to examine the eye retinal fund us images. The manual analysis includes Visual Acuity testing, Tonometry and Pupil dilation. The vision lost due to Diabetic retinopathy is sometimes irreparable. Hence there is a need for earlier detection and treatment to reduce the risk of blindness.Hence there are various automated methods of diabetic retinopathy screening that have made good progress using image classification, pattern recognition, and machine learning. The input to the automated image classification model can be the color fundus photography or optical Coherence tomography images.