2013 : Segmentation of Hard Exudates in Retinal Fundus Images Using Fuzzy C-Means Clustering with Spatial Correlation

Prof. Ir. Handayani Tjandrasa M.Sc., Ph.D.


Abstract

Diabetic retinopathy is one of severe complication diseases caused by diabetes mellitus. This disease may cause blindness. One sign that a patient has diabetic retinopathy is the existence of hard exudates in his/her retina. Hard exudates are bright spots that consist of leaked material of lipids and protein in retinal blood vessels. Generally, an ophthalmologist measures and labels hard exudates in a retinal image manually to diagnose the severity of the disease. This manual way sometimes is subjective and time consuming. This research realizes the segmentation of hard exudates in retinal images automatically. The result is expected to provide a more accurate and faster segmentation process. The applied method is based on Fuzzy C-Means Clustering (FCM) with spatial correlation. This method is combined with several image processing methods.The main method is consist of two phases. The first one is the preprocessing phase and the second one is the segmentation phase. In the preprocessing phase the contrast of retina fundus image is improved using histogram equalization. Subsequently, disc optic is eliminated using midpoint circle algorithm. The image of retinal fundus without optic disc becomes an input to the segmentation process as the second phase. The segmentation process is done using FCM algorithm with spatial correlation. With this modification, the local spatial information also contributes to the clustering of the pixels. According to the experiment result, the proposed method shows the capability in detecting hard exudates with the average accuracy of 79%.