Segmentation of Retinal Anatomical Structures using Morphological-based Thresholding

Abstract

A Normal retina has anatomical structures namely optic disc and retinal vessels. An abnormal retina consists of these anatomical structures as well as pathological structures which are red and bright lesions.  The proper segmentation of the retinal anatomical structures is essential for the effective diagnosis of diseases like glaucoma, diabetic retinopathy, cerebral diseases, etc.  Because glaucoma affects the optic disc and other diseases harm the blood vessels.   With the help of efficient segmentation of optic disc and vessels, we can easily diagnose these diseases.  For this reason, here we propose a technique of morphological-based thresholding which segments the optic disc and retinal blood vessels exactly.  With the help of a publicly available database namely the High-Resolution Fundus (HRF) database, this proposed method is validated.  This database consists of 45 images which are grouped into three sets.  For every set of images, we obtain the measures of sensitivity, specificity, and accuracy. For optic disc segmentation, we achieve the average of 0.8837, 0.9958, and 0.9932 in sensitivity, specificity, and accuracy respectively for the three sets of HRF databases.  And for the vessel segmentation, we achieve an average of 0.7181 sensitivity, 0.9693 specificity and 0.9573 accuracy. And these results show that the proposed technique yields the better performance than the existing one in the literature.