Classification of Diabetic Retinopathy Using Image Pre-processing Techniques

Abstract

Most diseases that affect humans cause an effect on the eyes which is observed first by a physician while treating a patient. Eye diseases are conditions that affect the functioning of the eye and lead to loss of vision. Globally, at least 2.2 billion people have near or distant vision impairment. Most of these eye diseases have long-term effects so, early detection of the eye disease is very important to avoid the consequences. Some eye diseases include cataracts, glaucoma, diabetic retinopathy, and retinal detachment, etc. Almost 10% of people are facing an eye disease called diabetic retinopathy (DR). In this paper, multiclassification of DR images has been done after performing a few preprocessing techniques like erosion, histogram equalization, and comparing the accuracy of the CNN model before and after the pre-processing. The classification model yielded an accuracy of 86.4% on retinal fundus images filtered from the Kaggle DR detection database. The obtained results show that the proposed preprocessing methods are best suitable for diagnosing diabetic retinopathy from retinal fundus images.