Vessel Segmentation Based on a Channel-Attention U-Net Algorithm

Abstract

Vessel segmentation is a fundamental task in medical image analysis and plays an important role in disease diagnosis and treatment assessment. However, existing segmentation methods often show limited adaptability to feature extraction from single-channel X-ray coronary angiograms, which restricts their performance in segmenting small vessels and low-contrast vascular regions. To address these limitations, this study proposes a Channel-Attention U-Net, termed CA-UNet, which integrates residual connections and a channel attention mechanism. Based on the conventional encoder–decoder architecture of U-Net, the proposed method introduces a residual-enhanced double-convolution block to alleviate gradient vanishing in deeper networks. In addition, a dual-pooling channel attention module is incorporated to enhance the selection of discriminative vascular features. Furthermore, the data loading, normalization, and augmentation strategies are optimized to improve the adaptability of the network to single-channel PGM grayscale images. Under three-fold out-of-fold evaluation, CA-UNet achieved the highest mean Dice coefficient (0.7450) and IoU (0.5972) among the evaluated models, while maintaining real-time-rate inference at 41.7 frames per second. These results indicate that CA-UNet provides an effective balance of segmentation accuracy, stability, and computational efficiency for vessel segmentation.