Deep Learning-Based Retinal Blood Vessel Segmentation Using U-Net Architecture

Retinal blood vessel segmentation is an essential component in the diagnosis and monitoring of various ophthalmological conditions, such as diabetic retinopathy, glaucoma or macular degeneration, providing valuable information about the state of blood vessels and ocular blood circulation. Traditional methods for segmenting blood vessels from retinal images present certain limitations in terms of accuracy and robustness, especially under variable lighting conditions, the presence of artifacts or low-quality images. However, current techniques based on convolutional neural networks have demonstrated the ability to achieve precise and robust segmentations in a variety of conditions, outperforming traditional methods. Nevertheless, neural networks face challenges regarding the limited availability of retinal image datasets necessary for training and evaluating these solutions. In this paper we describe a solution for retinal blood vessel segmentation using the U-Net architecture. This neural network achieves impressive results even with very small datasets, demonstrating the efficiency and robustness of the U-Net architecture for retinal blood vessel segmentation.