Ph.D. Tezi Görüntüleme | |||||||||||||||||||||
|
|
||||||||||||||||||||
Summary: Diabetic retinopathy (DR) is an eye disease caused by diabetes, and results blindness if it is not diagnosed at early stages. DR is the major disease that causes blindness in theworld. Patients with diabetes should be monitored using fundus imaging by an ophthalmologist for early diagnosis and treatment of DR. However, incidence of DR diseaserises considerably depending on increased number of diabetes patients. Adequate amount of time and number of experts are needed to monitor DR patients. Unfortunately, theseresources is unavailable at sufficient rates. Hence, we proposed automatic methods for diagnosing DR disease that will reduce diagnosis time of ophthalmologist by eliminatingdiagnosis of unnecessary images. In this thesis, a novel method for assessing adequacy of fundus image quality is proposed at first. After identification of suitable retinal images,fundamental anatomical structures of retina, which are blood vessels, optic disc and fovea, are detected. These anatomical structures are crucial for diagnosis of abnormalities in theretina. Finally, machine learning algorithms are used to identify hard and soft exudates, hemorrhages and microaneurysms.Key Words: Retinal image analysis, Image quality assessment, Diabetic retinopathy, Blood vessels, Optic disc, Fovea, Image classification, Image processing, Machinelearning. |