Feature extraction from the fundus images for the diagnosis of Diabetic Retinopathy

Abstract

Diabetic retinopathy refers to an advanced eye screening technology by which eye related diseases can be detected at an early stage. Detection of lesions in fundus images can assist in early stage of a Diabetic Retinopathy. A robust, adaptive and conceptually efficient approach for localization of features and lesions in a fundus image is discussed in this paper. As certain features have common properties, features, correlations techniques proposed. This method utilizes the intersection of the abnormal thickness in the blood vessels to find out the approximate location of the optic disk. Then it is further localized using color images. Later the proposed method show that more features such as the hemorrhages, micro aneurysms, hard exudate and soft exudates can be detected. Evaluation of the algorithm on some images on database with different illumination, contrast and DR stages has given success rate as high as 90% for hemorrhage. Further the other features extracted has given out about 95% for microaneurysm, 95% of sensitivity and 94% of specificity for exudates identification and yields 97% of success rate for optic disk localization. Hence the investigations are suitable to be registered in the form of a brief write up in this paper.