Abstract—Diabetic retinopathy ranks among the most prevalent preventable causes of blindness worldwide. Failure to detect this condition early on can lead to permanent vision loss. With the increasing number of diabetic patients, there’s a pressing challenge due to the scarcity of ophthalmologists. To assist these professionals in decision-making, the development of an automated Diabetic Retinopathy (DR) screening system is recommended. A key aspect of diagnosing diabetic eye disease involves identifying retinal lesions, particularly hard exudates. Proposed research suggests employing an advanced Genetic Algorithm, leveraging Egyptian Vulture Optimization (EVO), and a ResNet-based U-Net model to detect hard exudates in retinal images for early-stage retinopathy diagnosis. The study utilizes the IDRiD retinal image database for training and validation, with the e-Ophtha dataset serving as the test images for the proposed algorithms. Results from the IDRiD set demonstrate an Intersection over Union (IoU) score of 89.5 and a loss of 0.149, while the e-Ophtha dataset yields an IoU score of 77.2 and a loss of 0.29. This innovative approach, trained with a limited number of images, outperforms existing learning algorithms, showcasing promising potential for enhanced diabetic retinopathy diagnosis.
 
Keywords—fundus image, diabetic retinopathy, machine learning, optimization algorithm


Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.