Deep Learning Framework for Diabetic Retinopathy Diagnosis

Abstract

Diabetic Retinopathy (DR) is one of the foremost causes for the presence of blindness in the recent times. Ophthalmologists usually diagnose the presence and severity of DR through visual assessment of the retinal fundus images by manual examination. This process of manual diagnosis of DR is a very laborious and time consuming task. With the increasing rate of diabetic retinopathy patients in the world, the number of color fundus images generated has increased exponentially. Due to this large number, there is a huge delay in recognizing the early symptoms of DR and providing timely treatment. Hence, to address this unmet and increasing need, there is a need for developing an automated framework of Diabetic Retinopathy diagnosis. Hence, in this study, we have proposed a Deep Learning framework for DR diagnosis. The study uses a modified version of one of the standard Convolutional Neural Network (CNN) for solving DR fundus image classification problems. The proposed framework efficiently and quickly report whether the person has DR or not and if present, reports the severity of the disease. The framework implemented helps in giving timely treatment to the patients irrespective of geographical and economic constraints.