High performance automated diagnosis system for hypertensive retinopathy

High blood pressure (hypertension), defined as the rise of blood pressure, can cause serious damage to the human body, which can lead to death. One of the devastating effects of hypertension is on retina of the eye. This vascular problem which is causing retinal deterioration is called "Hypertensive Retinopathy". Hypertensive retinopathy, which is characterized by thickening or narrowing of the retinal vessels, may result in vision loss and even blindness in later stages of the disease if it is not diagnosed and treated early. To detect presence of hypertensive retinopathy, retinal images must be examined in detail by eye specialists but making these examinations manually may take a lot of time and lead to false diagnosis. For this reason, intelligent computer systems that produce automatic and fast results for these diagnoses procedures are required. The aim of this thesis is to be able to diagnose hypertensive retinopathy with high accuracy and high performance. Firstly, blood vessels in the retinal images that are taken from publicly available DRIVE and STARE databases are segmented for making examination in detail by applying a set of image processing methods. The Arteriovenous Ratio (AVR) is used to detect and calculate changes in the segmented blood vessels due to hypertensive retinopathy. It is decided whether there is hypertensive retinopathy according to the calculated AVR value for each retinal image. These operations that are applied on the retinal images for the diagnosis of hypertensive retinopathy are time consuming procedures. Therefore, improvements that are done on the Graphical Processing Unit (GPU) make it possible to perform diagnosis operations in a much shorter time and without compromising accuracy. Keywords: hypertensive retinopathy, AVR, high performance system, image processing, image segmentation, image enhancement, GPU.