Metody segmentace retinálních cév: konvenční x inteligentní metody

Abstract

The aim of this thesis is the creation of a conventional and intelligent algorithms for the segmentation of blood vessels on images of the retina obtained from the fundus camera system RetCam3, and the subsequent objective evaluation and comparison of the ability of these algorithms to correctly segment blood vessels. The main part of this thesis describes the implementation of individual algorithms - The Conventional method Maximum Principal Curvature, and two U-Net intelligent networks. The results of these methods are then objectively evaluated using their SSIM, MSE, Accuracy, Sensitivity, and Specificity in comparison to the real mask. These values are obtained for several configurations of the conventional algorithm, and for different epoch counts of the intelligent networks. As the last step of this work these values are compared between eachother, with the data showing much greater segmentation capabilities in intelligent methods.