Automated Segmentation of Hard Exudates Using Dynamic Thresholding to Detect Diabetic Retinopathy in Retinal Photographs

Retinal images are in use by ophthalmologists for the clinical analysis and diagnosis of different retinal diseases. The ocular disease known as Diabetic Retinopathy (DR) is a retinal disease that causes microvascular changes in the eye retina. Hard Exudates (HE) a retinal lesion can be seen as bright yellowish spots in colored fundus photograph and as bright white blobs in red free fundus image. The aim of this paper is to propose an automated technique for the identification of HE to help in the diagnosis of DR. The proposed method use dynamic thresholding for the segmentation of HE after calculating intensity based parameters of the input retinal image. Before the segmentation of exudates the Optic Disc (OD) is removed from the image using averaging and morphological operations. A sensitivity of 98.73%, specificity of 98.25% and accuracy 97.62% for HE segmentation is achieved respectively on 1200 images from publically available database MESSIDOR. Compared with the state of art, the proposed automated technique has a reasonable accuracy and can be used as a trustworthy clinical diagnostic tool.