Exudates Detection from Digital Fundus Images Using GLCM Features with Decision Tree Classifier

Abstract

Diabetes affects a number of human organs, the most common organ being the human eye. Diabetic Retinopathy, Glaucoma, Macular Edema are some of the common ophthalmic disorders found in diabetic patients. Ophthalmologists diagnose Diabetic Retinopathy in a digital fundus image with the presence of exudates. The proposed algorithm consolidates morphological operations for blood vessel removal, segmentation and optic disc removal followed by exudates detection. In this experiment the GLCM features are extracted. These features enhance the detection of affected regions in a retinal image as it depicts how frequently various combinations of gray levels co-exist in an image section. This experiment also explores the use of SVM, k-NN and Decision tree classifiers to distinguish between diseased and healthy retinal images. It is observed from experimentation that the Decision tree classifier yields best results of classifying exudates in digital fundus images. The PPV of the proposed algorithm with decision tree classifier is 100% for DIARETDB0, 97.6% for DIARETDB1, 97% for e-Ophtha EX and 100% for Messidor databases. The sensitivity of the proposed algorithm is 100% for DIARETDB0, 100% for DIARETDB1, 91.6% for e-Ophtha EX and 94.5% for Messidor databases. The proposed algorithm also exhibits 100%, 97.6%, 94.5% and 94.5% accuracy values for DIARETDB0, DIARETDB1, e-Ophtha EX and Messidor databases, respectively using GLCM features with decision tree classifier. Thus, Decision tree classifier is proposed as robust means for detecting exudates and examining the presence of Diabetic Retinopathy in digital fundus images.