An Automated Mechanism for Early Screening and Diagnosis of Diabetic Retinopathy in Human Retinal Images

Diabetic retinopathy is a disease that is a primary cause of blindness among diabetes patients. A regular screening of patients with diabetes is important to enable early intervention. The focus of this paper is on designing a robust and computationally efficient automated system to assist in the early screening and diagnosis of diabetic retinopathy for early treatment. Digital analysis and mathematical morphology operations are implemented appropriately to detect and locate different features and lesions in fundus retinal images. One of our goals here is to detect bright lesions and dark retinal feature. Bright lesions consist of exudates and optic disc, dark retinal features consist of hemorrhages, blood vessels, and microaneurysms in retinal images of the human retina. The system algorithm was extensively evaluated on a database of 89 images with the corresponding experts (ophthalmologists) manually highlighted ground truth images. An image by image visual evaluation yields 84% success rate for exudates detection with a sensitivity of 86% and specificity of 80%. For hemorrhages detection the success rate is 81% with a sensitivity of 83% and specificity of 71%. The success rate for microaneurysms detection is 89.1% with a sensitivity of 90% and specificity of 20%.