Detection of exudates from clinical fundus images using machine learning algorithms in diabetic maculopathy.

Purpose of the study: As per the estimation of the World Health Organization, around 400,000,000 people may be affected by diabetes in the year 2014. It is rapidly increasing double by the upcoming year 2030. Diabetic mellitus, also called diabetes, can cause critical issues of retinal disease such as diabetic retinopathy (DR) which causes blindness. Exudates are considered as the major sign of diabetic maculopathy (DM) which may lead to visual disturbances. Methods: The main purpose is to develop a computer-aided technique for the detection of hard exudates to assist in the diagnosis of DR. It mainly focuses on the identification on the yellow lipids that include hard exudates. The diseased retinal image is given as an input to the classifier; it produces the output as exudates or non-exudates. In the proposed work, the classifier such as support vector machine and multilayer perceptron are used to find out the accurate prediction of exudates and non-exudates. First, the fundus images are allowed to preprocessing technique to get the filtered, contrast enhanced image. Second, the features such as blood vessel segmentation implements morphological operation for the detection of hard exudates by measuring the size of the lesion, and optic disc (OD) is measured and eliminated by comparing the parameters with the size of the lesions. Third, the segmented images are given as an input to the classifier such as SVM and MLP; it classifies and given as an output about the presence or absence of the exudates. In the proposed work, the experimental tests are carried out, and the results are verified on a different set of images. Results: The average accuracy values obtained using SVM and MLP classifiers for 140 images from real-time databases collected from Aravind Eye Hospital, Coimbatore are 88% and 95% respectively. [ABSTRACT FROM AUTHOR]