Detection Of Diabetic Retinopathy Using Deep Convolutional Neural Network On Mobile Devices

Abstract: Diabetic Retinopathy (DR) is an ophthalmic disease that damages retinal blood vessels.DR causes impaired vision and may even lead to blindness if it is not diagnosed in the early stages. Early eye examination and detection of DR is the best solution to control the growth of this disease. DR has five stages or classes, namely normal, mild, moderate, severe, and PDR (Proliferative Diabetic Retinopathy). Normally, highly trained experts need to examine the colored fundus images to diagnose this fatal disease. This manual diagnosis  (by clinicians) is tedious, error-prone, and extortionate. Therefore, various computer vision-based techniques have been proposed to detect DR and its different stages automatically. However, more research is still needed to encode the complicated underlying features. Most of the systems so far can only classify DR’s different stages with very low accuracy, particularly for the early stages. Further, the diagnosis process is done only in clinics or hospitals. In this research, we are aiming to develop a mobile-based classifier model using a deep neural network, that allows the person with diabetes to have the initial analysis of  DR made locally, in order to decrease the load of the eye health care professionals (opticians) when it comes to detect and classify anomalies in the retina image, and also support the professional in analysing the retina image.