Study On The Segmentation Of Blood Vessel And Other Related Technology In Retinal Image

As human healthy are closely associated with the development of medicalscience, digital image processing technology has caught the attention of biomedicalfield from the very first. Image processing and analysis techniques have been widelyused in basic medical research and clinical diagnosis. The studies have shown thatthere is a close relationship between retinal vascular abnormalities associated with theincidence and severity of hypertension, stroke, brain atrophy, coronary heart diseaseand other systemic disease. Because retinal vascular abnormalities usually occurbefore the hypertension and other related diseases, it may indicate the occurrence ofthese diseases. As a result of the limit on the formation of image and the complexityof the retinal structure, the digital fundus image shows various kinds of features. Andit is very difficult to analysis the digital retinal image.The segmentation of vessel bloods and the classification of arteries and veins arehuge problems in the analysis of retinal image. Segmentation based on morphology,center lines and Hessian matric are three parts in our research on extracting vesselbloods. The method based on morphology combined with morphological operations,the characteristic of the cross-section of vessels to enhance these fundus images. Thena hysteresis thresholding is using to get binary image of the vascular network. Thesegmentation based on the center lines is using the center lines of bloods along withthe bit-plane of digital retinal image to extract vessels. And the segmentation usingHessian matrix is based on multi-scale space theory, the Hessian matrix and Otsu’sthreshold method. The second problem is the classification of arteries and veins inretinal image. Supervised and unsupervised classifications are main ways ofdistinguishing between artery and vein. In the unsupervised classification, the fundusimage is divided into four quadrants and fuzzy clustering is used to classify artery andvein. The supervised method finds many features, and then uses classifier to markA/V.The implementation of the accurate extraction of the retinal vessels meets theneeds of medical research and clinical diagnosis and treatment. According to the studyof arteriovenous difference, the automation of A/V classifier has been implemented. All of methods have been tested on DRIVE. Finally, the validity and accuracy of ourmethods have been verified.