Abstract

Accurate segmentation of retinal structures from fundus images is essential for the early detection and monitoring of ocular diseases such as diabetic retinopathy, glaucoma, and age-related macular degeneration. Traditional manual and semi-automated methods are time-consuming, error-prone, and limited in scalability, while classical image processing techniques struggle with variations in illumination and anatomical diversity. To address these challenges, we present RETINASEG, a deep learning-based diagnostic framework that integrates Convolutional Neural Networks (CNNs) for disease classification and a U-Net architecture for retinal structure segmentation. The system was trained on a curated dataset of 1,200 training and 200 testing fundus images, annotated by expert ophthalmologists and categorized into four classes: cataract, diabetic retinopathy, glaucoma, and normal. Experimental evaluation demonstrates improved robustness and precision, with correlation coefficient (CC) of 0.208, normalized scanpath saliency (NSS) of 0.8172, Kullback–Leibler divergence (KLD) of 2.573, and structural similarity index (SSIM) of 0.169. These results highlight the model’s ability to capture complex retinal features despite data limitations. RETINASEG reduces reliance on manual annotation, enhances diagnostic accuracy, and supports large-scale screening, offering significant potential for clinical deployment and improved patient outcomes.