الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 152 |  |  |
| | | from | 152 | | |
Abstract
Diabetes mellitus (DM) over a long period gives rise to deterioration of small retinal blood vessels. These retinal blood vessels leak fluids and blood, which cause retinal tissues swelling. As a result, it may cause diabetic retinopathy (DR), which is a significant complication of DM. DR is one of the leading causes of blindness in the working-age population worldwide. It is caused by high blood sugar levels, which damages retinal blood vessels and leads to vision loss. Therefore, the vasculature of the retina can be considered as a good indication of many retinal diseases. However, segmenting these vessels and analyzing them can be useful in detecting and diagnosing many diseases, including DR. On the other hand, the clinical features, such as neovascularization, microaneurysms, and hemorrhages, are presented on the people suffering from DR. Neovascularization is the unusual appearance of new blood vessels in many eye parts containing, of course, the retina. The walls of these new vessels are weak and may break and bleed. One of the primary effects of neovascularization and bleeding is the appearance of new vascular crossover and bifurcation points in the retinal vasculature network. Therefore, accurate early detection of these signs is essential to prevent blindness and avoid DR complications. The main challenge in DR diagnosis is that it requires specialized training. Also, its clinical grading is, to some extent, subjective. The diagnosis of DR requires manual measurements and visual assessment of the changes that happen in the retina, which is a highly complex task. Therefore, manual diagnosis and analysis of the retinal images will be a time-consuming and tedious process. So, automatic detection and diagnosis will minimize time and effort, which will help in the early detection of the disease. Thus, there is an unmet clinical need for a non-invasive and objective diagnostic system that can improve the accuracy of early diagnosis of DR as well as detecting different DR grades. In this thesis, we propose a comprehensive novel computer-aided diagnosis (CAD) system for detecting early signs of DR as well as grading different stages of DR mainly based on analyzing optical coherence tomography angiography (OCTA) scans. First, we develop a segmentation technique for retinal blood vasculature network that segments both superficial and deep retinal plexuses for both normal subjects and DR patients. To segment these vessels, the system integrates three various models, which are appearance, spatial, and prior probability models. To overcome the low contrast between the background tissues of the retina and the blood vessels, we integrated the 3D MarkovGibbs random field (MGRF) with the 1st-order intensity model. In the end, the segmentation was refined by extracting connected regions using a 2D connectivity filter. Following the segmentation step, four new local features are estimated from the segmented retinal blood vessels and the foveal avascular zone (FAZ): i) vessels density, ii) blood vessel caliber, iii) distance map of the FAZ, and iv) bifurcation and crossover points. These features are then fed to a two-stage classifier to diagnose the OCTA images into normal (no-DR) or mild DR case.