Retinal vessel segmentation is essential for the early diagnosis of diseases such as diabetic retinopathy, hypertensive retinopathy, and age-related macular degeneration. Manual segmentation of fundus images is time-consuming and prone to variability, limiting large-scale screening. This paper presents RETINASEG, a deep learning-based system for automated pixel-level segmentation of retinal vessels from fundus images. The proposed framework combines image enhancement techniques such as contrast normalization, CLAHE, and noise reduction with an encoder–decoder architecture based on U-Net and transformer-enhanced models. To address challenges including thin vessel detection and class imbalance, data augmentation and class-balanced loss functions are employed during training. Experimental results on DRIVE and STARE datasets demonstrate strong performance, achieving high accuracy and robustness across datasets. A web-based interface with real-time visualization and explainable AI support further enhances clinical usability. RETINASEG enables scalable, reliable, and automated retinal analysis for early disease detection and tele-ophthalmology applications.