Artery and Vein Classification in Retinal Images Using Graph Based Approach

Digital image analysis of eye fundus images has several benefits than current observer based techniques. The characteristic symptoms of different systemic diseases like hypertension, glaucoma, diabetes and cardiovascular disorder etc. affects retinal vessels. Diseases like diabetes show abnormalities and diameter changes in retinal blood vessels. In hypertension retinal blood vessels show dilatation and elongation of main arteries and veins. Arteriolar to venular diameter ratio (AVR) express high blood pressure levels, diabetic retinopathy and retinopathy of prematurity. Among other image processing operations the estimation of AVR requires vessel segmentation, accurate vessel width measurement and artery or vein classification [1]. Hence the identification of arteries is essential to detect eye diseases. The work has been done on automated classification of retinal vessels and hence it is a challenging task. INTRODUCTION: Nowadays for image analysis graph based methods have been used which are useful for retinal vessel segmentation, retinal image registration and retinal vessel classification[2]. The segmented vessels are analyzed using type of intersection and then assigned artery or vein labels to each vessel segment. So the combination of labels and intensity features decides final artery or vein class. Most of the methods uses intensity features to differentiate between arteries and veins. Due to the acquisition process, very often the retinal images are nonuniformly illuminated and exhibit local luminosity and contrast variability, which can affect the performance of intensity-based A/V classification methods. For this reason, proposed method which uses additional structural information extracted from a graph representation of the vascular network. The results of the proposed method will show improvements in overcoming the common variations in contrast inherent to retinal images. LITERATURE REVIEW: With respect to said work an extensive literature survey is conducted accordingly which is presented as below, 1. Several features like visual and geometrical have explored the methods for artery or vein classification. Vessel diameter is not a reliable feature for artery or vein classification since it can be affected by diseases [3]. 2. MartinezPerez et al. (2002) In semi automatic method [4] geometrical and topological features of single vessel segments and sub trees are calculated. Significant points are detected through the skeleton extracted from the segmentation result. For labelling purpose root segment of the tree is tracked and then algorithm will search for its unique terminal points and decide if the segment is artery or vein. 3. Grisan et al.(2003) In optic disc zone arteries rarely cross arteries and veins rarely cross veins[5] hence by using vessel structure represented by tracking the classification is propagated outside this zone where little information is available to discriminate between arteries and veins. By applying ‘a divide at impera’ approach which partitioned a concentric zone around the optic disc into quadrants performs more robust local classification analysis. 4. S.Vazquez et al.(2009) In color based clustering algorithm with a vessel tracking method[6] NOVATEUR PUBLICATIONS International Journal Of Research Publications In Engineering And Technology [IJRPET] ISSN: 2454-7875 VOLUME 2, ISSUE 3, March -2016 2 | P a g e retinal images are divided into four quadrants and then it combines the result. Then by using tracking strategy based on minimal path vessel segments are joined to support the classification by voting. 5. C. Kondermann, D. Kondermann et al.(2007) Two feature extraction methods and two classification methods[7], based on support vector machine and neural network to classify retinal vessels. One of the feature extraction methods is based on ROI (Region Of Interest) around each centerline point while the other is profile based. To reduce dimensionality of feature vectors principal component analysis is used. 6. M. Niemeijer, B. van Ginneken et al.(2009) Image feature and classifier is an automatic method for classifying retinal vessels into arteries and veins[8]. A set of centreline features is extracted and a soft label is assigned to each centerline, indicating it’s being a vein pixel. 7. R.Estrada, C.Tomasi et al.(2012) present a methodology[9] for vessel structure in human retina using Dijkstra’s shortest-path algorithm. The method requires no manual intervention, preserves vessel thickness and follows vessel branching naturally and efficiently. 8. M. Niemeijer, X. Xu, A. Dumitrescu, P. Gupta et al.(2011) In the classification method[10] is considered as a step in calculating AVR value. The estimation of AVR requires vessel segmentation, accurate vessel width measurement and artery vein classification hence slight error can produce large influence on the final value.