Classification of Diabetic Retinopathy Patients Using Support Vector Machines (svm) Based on Digital Retinal Image

Diabetic retinopathy is a micro vascular complication which is characterized by several changes in the retina. Changes occur in the diameter of the blood vessel, microaneurysm, hemorrhage exudates, and the growth of new blood vessels. These changes need to be detected early so that steps for further handling and treatment can be determined. Laser therapy is one of the common therapies for patients with Diabetic Retinopathy. This therapy is a manual examination of the scanned results of the fundus retinal image. Manual examinations that generate ophthalmologist sight differ from each other. To overcome this problem, a special program is needed to analyze the fundus image of the eye. To create a special program for analyzing the fundus images of the eye required several stages of research. The study begins by preprocessing eye fundus images, getting rid of the optic dick form the fundus of the eye and then separating the vascular tissue of the damaged area of the retina. Damaged areas of the retina consist of dark and bright lesians . Mathematical morphology methods are used to detect the presence of dark lesian. To detect the presence of bright lesian a combination of mathematical morphology, Estimated Background, Colour analysis, Max-tree and attribute filters are used by utilizing a branch filtering approach. Fundus image segmentation results are extracted and classified using Support Vector Machines (SVM) based on microneurysm and exudates features. Eye fundus images are classified into, Mild NonProliferative Diabetic Retinopathy, Moderate Non-Proliferative Diabetic Retinopathy and Severe NonProliferative Diabetic Retinopathy. The novelty of this research using maxtree representation and atribute filtering to enhance image quality for exudate segmentation. From the classification experiments on patients with diabetic retinopathy the following sensitivity level were obtained, specificity and AUC above 90%. This indicates that the research could help opthalmologist in analyzing a retina that is affected by diabetic retinopathy. The results of the study showed 96.9% sensitivity, specificity 100%, positive predictive value(ppv) 100%, negative predictive value(npv) 88.19 and AUC 0.985%.