This study enhances automated segmentation of blood vessels in retinal images, providing a valuable tool for early cardiovascular disease detection and improving the accuracy of non-invasive diagnostic methods. This chapter focuses on the development and application of a method to segment arteries and veins in infrared reflectance fundus images. A database of 23 retinal images was used, each with its corresponding blood vessel segmentation mask. The images, with a resolution of 768 × 768 pixels and in grayscale, underwent data augmentation, expanding the dataset by 25 to 50 times. Three deep learning approaches were employed: U-Net, ResNet50 + U-Net, and ResNet50 + U-Net with attention layers. These approaches aimed to facilitate the identification and segmentation of blood vessels for further analysis. The proposed method achieved an IoU of 0.5186 and a Dice coefficient of 0.705 for arteries and 0.64 for veins, demonstrating its potential for accurate blood vessel segmentation. These findings highlight the potential of deep learning-based segmentation to improve the early detection of cardiovascular diseases by enabling accurate analysis of vessel caliber changes in retinal images. This chapter contributes to advancing early detection methods for cardiovascular diseases by leveraging the capabilities of deep learning for blood vessel segmentation in retinal images.