Comparative Evaluation of Machine Learning Techniques for the Detection of Diabetic Retinopathy

Abstract

Diabetic Retinopathy (DR) is a common diabetes disorder that attacks blood vessels in the light-sensitive tissue known as the retina. It is among the most common causes of loss of vision among patients with diabetes, and it is the leading cause of reduced vision and blindness even among aged adults. Naturally, this occurrence begins with no apparent change in vision. For the identification of DR, ophthalmologists use the retinal image of a patient known as the fundus image, and the blood vessels may also be captured explicitly from the retina. This paper presents a comparative study of five commonly used machine learning techniques: K-Nearest Neighbor, Support Vector Machine and Discriminant Analysis, Naïve Bayes, and Ensembles. The texture characteristics of the fundus image were extracted using the Local Binary Pattern (LBP) descriptor. And this feature extracted using LBP was used to train the classifiers. The proposed method classifies the retina's fundus pictures as "no DR" or "current DR." The Ensemble Classifier (EC) technique generated a better DR detection accuracy of 98.31% than the other four classifiers and existing works based on the classifiers' comparative analysis.