Performance Comparison Using Thresholding Based Method for Diabetic Retinopathy

Abstract

Patients with diabetes need annual screening to circumvent vision loss which may lead to blindness. Diabetic Retinopathy (DR) is a diabetic complication that causes structural changes in the retina. Non proliferative diabetic retinopathy (NPDR) is a common, usually mild form of retinopathy that generally does not interfere with vision. However, the diabetic retinopathy can progress from non-proliferative to proliferative retinopathy (PDR) if left untreated. To prevent this situation, the automatic computer system is introduced to identify the early stages of DR. There are a lot of studies and research of DR but yet to achieve the accurate result. In order to achieve the target, numerous image segmentation methods were used for comparison performance. In this paper, three datasets namely of DRIVE, E-Optha and Messidor were used as input images. There are three methods from thresholding-based category were used in order to identify the microaneurysms (MAs) and the blood vessel. For DRIVE database, Otsu obtained an accuracy of 92.09%, 93.38% in sensitivity followed by specificity of 64.82%. While Entropy method obtained an accuracy of 92.03%, 94.65% in term of sensitivity followed by 62.38% in specificity. For Fuzzy C Mean (FCM) the accuracy was 92.42%, 94.46% in term of sensitivity and 63.09% in specificity.