Development of diagnostic methods of retina fundus components with image processi̇ng and artificial learning

Fundus images are the process of visualizing the retinal structure of the eye. These images contain information about many diseases that impact the eye and other organs. To obtain this information, retinal fundus images are analyzed thoroughly. These analyzes are made by specialist physicians using computer-aided diagnosis systems. The success of this analysis process depends on the knowledge and experience of the specialist physicians. In addition, it is a time-consuming and complicated process. With the advance of technology, better analysis systems are needed. In this thesis, retinal fundus images were examined in detail and computer-aided diagnosis systems are developed over these images. The first of these systems focused on localization detection of the optic disc, which has an important position on the retina. In the proposed system, fundus images in RGB color space are moved to new color space and optic disc localization was detected in this color space. The second proposed system is a method that allows the extraction of retinal blood vessels, which is used in the detection and diagnosis of many diseases starting with diabetic retina. In this proposed method, retinal blood vessels are segmented with pixel-based features obtained from fundus images. The third modeled system is a system that is capable of distinguishing between the arteries and veins of the retina blood vessels. In this modeled system, image patches extracted from retina blood vessels are given as inputs to a designed convolutional neural network. The output of the convolutional neural network enables the distinction between the arteries and veins of the retina blood vessels. The last system designed is a system that can distinguish fundus images as glaucoma or healthy. Here, the EfficientNet model, which is one of the current methods, is used. All of the systems proposed are tested on the fundus datasets which are publicly available. Performance results of all of the proposed systems are satisfactory in the sense that they are competitive with those of state-of-the-art methods used in the literature. As a consequence, developed systems can be used as decision support systems to help physicians.