Discrimination of microaneurysm in color retinal images using texture descriptors

Abstract

The occurrence of diabetic retinopathy and diabetic mellitus has been increasing worldwide. Currently, ophthalmologists face a lot of challenges in identifying various stages in diabetic retinopathy. Among these stages, the early stage is microaneurysm. A new computer-aided microaneurysm detection system based on texture features is presented. The histogram of texture descriptor reveals the texture characteristics of each pixel which effectively increases the accuracy of MA detection than shape-based features. The extracted features using Local Binary Pattern contribute more while discriminating the lesions using support vector machine classifier. Validations based on free-response receiver operating characteristic score and area under curve are obtained for ROC, MESSIDOR and DIARETDB1 dataset. The ability to process images with different intensities and less computation time guarantees the robustness of this system.