<p>Accurate segmentation of retinal blood vessels is essential for the early diagnosis and effective treatment of various ophthalmic diseases. However, the complex structure of blood vessels, variations in vessel width, and the presence of low-contrast regions in retinal images pose significant challenges. To address these challenges, this paper presents a collaborative transformer attention U-Net (CTAUNet) framework for blood vessel segmentation. A U-Net architecture built with convolutional neural network (CNN) is employed by CTAUNet to achieve accurate blood vessel localization, while a swin transformer module is integrated to capture both fine details and broad image context within the retinal image. Furthermore, a collaborative feature fusion mechanism effectively combines multi-scale features from both CNN and transformer branches, improving the segmentation of blood vessels across different scales. Extensive experiments conducted on multiple benchmark datasets (STARE, CHASE DB1, and DRIVE) show improvement over existing methods, achieving F1-scores of 0.872, 0.888, and 0.879 and accuracy of 0.965, 0.989, and 0.981 on DRIVE, STARE, and CHASE DB1 datasets, respectively.</p>