Automatic grading of macular degeneration from color fundus images

Abstract

Diabetic retinopathy causes blindness due to the physiological changes in the retina of human eye, which occurs due to the progression of diabetes. Diabetic retinopathy images differ from normal fundus images by lesions such as: microaneurysm, hemorrhages, exudates, cotton wool spots and variations in blood vessels etc. Appearance of these features on the retina leads to vision loss. The sharp vision is affected severely when the features appear on the macula as it contains higher concentration of cones. In this paper macula and fovea (macula center) are detected based on the localization. The detection of these feature is essential for automatic grading of macular edema or degeneration. Depending on the number of lesions on the macula the severity of the macular degeneration can be predicted. The method is tested on DRIVE, Aria and DIARETDB1 databases. The method successfully detects the macula and fovea for all the images. The accuracy of the proposed method of macula detection is found to be 100% in normal images. The method is also applied on images with lesions. Here the overlapped region of the macula and lesions are detected to find the severity of macular degeneration.