Automated classification of exudates from digital fundus images

Abstract

Diabetic Retinopathy and Diabetic Macular Edema are diseases that affect vision and eventually may lead to blindness. Early detection is a must to prevent the progression of the disease imploring the need for effective computer-aided diagnostic techniques. In the following research paper, a robust method has been proposed to segment hard exudates from digital, color fundus images using anisotropic diffusion and adaptive thresholding followed by a support vector machine for classification. The geometrical, shape and orientation features have been used to correctly classify the segmented objects as exudates or false pixels. The proposed technique has a high specificity and eliminates false positives correctly when applied across a wide range of images. The exudates segmented have a high degree of accuracy and no false positives are generated in case of non-diseased images. The proposed method has been tested on a total 189 images of the DIARETDB1 and MESSIDOR database and achieves an accuracy of 92.13% and 90% respectively. The proposed method can be used in the development for some computer aided technology for ocular diseases detection from fundus images.