Unsupervised and supervised retinal vessels extraction methods from the fundus images

Abstract

Retinal blood vessels are acknowledged as an indispensable newlineelement in both ophthalmological and cardiovascular disease diagnosis. newlineComputer-supported design of the medical investigative systems facilitates newlinehealthcare professionals to diagnose pathologies faster and more precisely. newlineAccurate segmentation of the retinal blood vessel has hence become the newlineessential requirement for automatic or computer-aided diagnostic systems. They newlinehelp to generate useful information to diagnose and monitor eye diseases like newlinediabetic retinopathy, hypertension, Macular degeneration, and glaucoma. This newlinethesis presents a set of solutions for the segmentation of retinal vessels. newlineInitially, an Unsupervised Morphological Approach (U-MAR) newlineis proposed to extract retinal blood vessels from fundus images. Mathematical newlineMorphology with a modified Top-hat transform is used in this method for newlinepreprocessing, and hysteresis thresholding is used for the extraction of blood newlinevessels. U-MAR method is evaluated on DRIVE-dataset, and the performance newlineis compared with the state-of-the-art methods. The proposed method achieved newlinean average accuracy of 95.95%, which shows that the approach is efficient for newlinecomputer-based retinal vessel segmentation. newlineSecondly, an un-supervised automated retinal vessel extraction newlineframework is presented using an enhanced filtering and hessian-based newlineapproach with hysteresis thresholding (U-EHT). Segmentation as a whole can newlinelose certain thin vessels that are very important for the diagnosis of eyerelated newlinediseases. Hence U-EHT method mainly focused on segmenting the newlinethin vessels and wide vessels individually. Wide vessels and thin vessels are newlineextracted separately by applying two different scales using a modified newlineHessian matrix and eigenvalues approach. newline