An automatic system for the detection of optic disc and pathologies in retinal images

There are an incredibly large number of undiagnosed Diabetic patients who are unaware of their disease, placing them at a greater risk of Diabetic Retinopathy. As there is no cure or method to prevent it, early detection of symptoms benefits in giving proper treatment. So there is a need for developing a medical system for automatic screening of Diabetic Retinopathy. This paper demonstrates a complete framework for the detection of bright as well as dark lesions in Retinopathy images. The Optic Disc is localized and segmented out from the color fundus images after some pre-processing steps such as filtering and local contrast enhancement. The resulting images are segmented using Fuzzy C-means clustering for the detection of exudates regions. For the detection of hemorrhages, the vascular structure is segmented out from the images following illumination correction. Then the hemorrhages are detected based on background estimation technique. The algorithm was tested and evaluated on two datasets namely DIARETDB0 and MESSIDOR. The proposed methodology illustrated an accuracy percentage of 99.6 for the detection of Optic Disc, 96 for Exudates and 87 for Hemorrhages.