Performance Evaluation of Multiple Classifiers for Hemorrhage Severity in Diabetic Retinal Fundus Images

Abstract

Diabetic retinopathy is one of the major diseases caused by diabetes. Diseases considered under diabetic retinopathy are problem in Optic Disc, Blood Vessels, Microaneurysms, exudates and fovea. Among these haemorrhages is one of the dangerous diseases which make vision loss speedily. Hence most of the recent research works are focusing on detecting the availability and analyzing the severity of haemorrhages in retinal images. But the accuracy of the haemorrhages analysis is less and not up to the market. This paper is motivated to provide a best classification approach by comparing the performance among different classification approaches to make use of it in medical industry. It helps to diagnose the severity of the haemorrhages for applying proper treatment in the earlier stage itself and avoid major surgery. Image preprocessing, image enhancement, haemorrhages segmentation, feature extraction and classification are the main steps followed in the proposed approach. KNN, Random Forest, Naïve Bayes and Multi-Class Support Vector Machine are the four different classifiers used in this paper. The experimental results is verified and the performance is evaluated among the above said classification approaches where it increases the accuracy of haemorrhages detection and classification on Diabetic Retinopathy Fundus Images.