Classification of diabetic retinopathy using textural features in retinal color fundus image

Early, diagnosis is essential for diabetic patients to avoid partial or complete blindness. This work presents a new analysis method of texture features for classification of Diabetic Retinopathy (DR). The proposed method masks the blood vessels and optic disk segmented and directly extracts the textural features from the remaining retinal region. The proposed method is much simpler with comparison of the other methods that detect the defective regions first and then extract the required features for classification. The Haralick texture measures calculated are used for classification of DR. The proposed method is evaluated through a classification of DR using both Support Vector Machine (SVM) and Artificial Neural Network (ANN). The results of SVM have a better accuracy (87.5%) over ANN (79%). The performance of the proposed method is presented also in terms of sensitivity and specificity.