M.Sc. Tezi Görüntüleme

Student:	Uğur ŞEVİK
Supervisor:	Yrd. Doç. Dr. Cemal KÖSE
Department:	Computer Engineering
Institution:	Graduate School of Natural and Applied Sciences
University:	Karadeniz Technical University, Turkey

Title of the Thesis:	Automatic Segmentation of AgeRelated Macula Degeneration on Retina Images
Level:	M.Sc.
Acceptance Date:	1/1/1980
Number of Pages:	89
Registration Number:	i1832

Summary:

Age-Related Macula Degeneration (ARMD) is one of the most common eye diseases causing the vision lost over 65 years old. In this study, a method is proposed to determine the drusens or ARMDs which occur as yellow-white small accumulation on the macula in the beginning of the disease. In the application, the optic disc is first detected in order to localize the macula region. For the finding of optic disc, we used the vertical edge detection filter to benefit from the knowledge of the edges of the disc which results from vessels. Hence, the maximum value is obtained from the histogram of the filtered image by calculating the vertical total intensity value. Then, the maximum value is used to determine the optic disc. Then, macular region including drusen is localized from geometrical relation between the optic disk and macula. Finally, the macula is located, and then the optic disc is eliminated to prevent the mis-segmentation because of the similarity between the intensity distribution of drusen and optic disk before the segmentation.

In addition to the pathological lesions, the texture of an eye consists of vessels, optic disc and macula. Since our aim is to determine the pathological lesions, the normal retinal textures should be eliminated. In order to do this, a statistical and region growing methods are employed to segment the healthy areas of the macula. Then, a simple vessel elimination method is also used to segment the vessels in the macular area. Hence, the healthy texture is first segmented, and then the segmented image is inverted to determine the degenerated area with drusen.

Manual segmentation of ARMD is quite difficult and takes long time to segment. Thus, the user may easily make mistakes during the segmentation of the degenerated area. Therefore, it may not be quite suitable in determination and examination of the changes of the drusens. Hence, the proposed methods are employed to segment the images automatically. Here, consecutive images from the same patient are also compared with each other to follow up the changes of the diseases.

Keywords: Age-Related Macula Degeneration, ARMD, Statistical Image Processing, Histogram, Region Growing, Medical Image Processing