Automatic detection of exudates in color fundus retinopathy images

Abstract

Objective: The aim of this paper is to design a computationally intelligent method to determine exudates, the Non-Proliferative Diabetic Retinopathy (NPDR) symptom which is considered to be the initial stage of retinopathy disease. If NPDR is not identified at its earlier stage, it may lead to Proliferative Diabetic Retinopathy (DR), the complicated stage of retinal symptom that may leads to blindness. Methods: It is proposed to develop an automatic computer aided detection system that screen a large number of people to identify the DR in its earlier stage for proper treatments. In this work, images are taken from publicly available e-optha database. Analysis mainly considers three stages which include removal of optic disc and normalization done by histogram processing; texture information extracted using Gray Level Co-Occurrence Matrix (GLCM) and classification is done with the help of Support Vector Machines (SVM). Findings: Preprocessing method used in this work enhances the contrast of the low or poor quality images. Here, optic disc segmentation is performed, which helps in providing better result by avoiding the misclassification of optic disc as lesions. GLCM method provides different texture features to SVM o provide better result in identifying exudate lesions. Conclusion: The intelligent machine learning approaches aid the ophthalmologists with accurate and efficient detection of abnormalities in fundus images. Through this system, the abnormal retinal images can be identified in its initial stage and an accurate assessment of retinal disease is possible.