Publication

Abstract : Diabetic Retinopathy DR and Glaucoma are the ocular diseases that are main causes of vision loss. DR and glaucoma are asymptomatic in nature. Early detection can help to reduce disease-related vision loss. Manual method of detecting these diseases is time consuming, expensive and requires skilled supervision. Factors such as poor contrast of fundus images, disease-specific segmentation approach and anatomical factors may all lead to inaccuracies in segmentation-based approach for localizing landmarks. Automated diagnosis of DR and glaucoma can overcome the shortcomings of manual method and reduce the burden on ophthalmologists during mass screening. This work proposes an automated framework for detection of DR and glaucoma using non-linear features that include Higher Order Spectra HOS, various types of entropies Kapur entropy, Renyi entropy, Yager entropy and Shannon entropy and Fractal Dimension FD. These non-linear features are capable of capturing minute pixel variations in fundus images for classification of normal, DR or glaucoma. For classification, Support Vector Machine SVM classifier with different kernels was used. Results shows that SVM-Radial Basis Function RBF kernel combination resulted in maximum accuracy of 85%, sensitivity of 84% and specificity of 94.32%. Based on the results it shows that the three-class classification of normal, DR and glaucoma using non-linear feature analysis and SVM can be useful for mass screening.