Detection and Classification of Diabetic Retinopathy Using Inception V3 and Xception Architectures

Abstract

Patients with diabetes usually develop a condition called diabetic retinopathy (DR), resulting from retinal damage. This impairment usually happens when the glucose levels in the blood are elevated, finally causing a blockage in the blood vessels that feed a part of the eye called the retina and finally severing it from the blood supply. Therefore, the eye attempts to produce fresh blood cells. But these cells are either poorly developed or weak. So, it can be leaked out easily. Hence, to lessen the severe effects of this disease, these patients must be diagnosed as soon as possible. Earlier, a number of approaches were put forth to recognise this illness using machine learning algorithms, image processing, and other techniques. The diagnosis process of this disease involves pre-processing of coloured images of the fundus, extraction of clinical features and classification of retinopathy. In this research, fundus photography of the retina is utilised to accelerate the detection of various kinds of retinopathy caused by diabetes based on convolutional neural network (CNN) pre-trained transfer learning algorithm. Inception V3 and Xception are used in this model to determine and categorise diabetic retinopathy, respectively. As a result, people with this disease can lower their risk of exposure to permanent blindness.