Segmentation and Detection of Vascular Bifurcations and Crossings in Retinal Images

Symptoms of many diseases, such as diabetes and hypertension, are characterised by changes in the structure of the retinal vascular tree. Identifying the location of blood vessels, vascular bifurcations and crossovers, in retinal images, may be helpful for predicting those diseases. However, a manual identification of blood vessels, vascular bifurcations and crossovers, in retinal fundus image, is a complex task that may take hours. In this paper, we present a three stages segmentation of blood vessels, consisting of: (i) enhancement of background and vessels, (ii) application of the Gaussian adaptive threshold, and finally, (iii) improvement of the resulting image by elimination of noise. In addition, skeletonisation and extraction of vessel geometrical features are used for detection of bifurcations and crossings. Experimental evaluation, using the DRIVE dataset, indicates that the proposed segmentation approach correctly identify 95.3% of the pixels belonging to blood vessels. Moreover, skeletonisation and geometrical features detect 95.0% of bifurcations and crossings.