A Deep Learning-Based Multi-Branch Network for Accurate Vascular Segmentation

Vascular interventional robots enhance the precision and safety of minimally invasive procedures, particularly in complex scenarios such as cerebral and coronary interventions. Accurate vessel segmentation from medical images is crucial for robotic navigation, path planning, and intraoperative risk assessment. However, vessel segmentation remains challenging due to low contrast, fine-scale structures, and complex vascular morphology. To address these challenges, a novel dual-branch convolutional module (DBCM) is proposed. The DBCM combines standard convolutions and depthwise separable convolutions in parallel, followed by a Coordinate Attention mechanism and a residual connection. The upper branch captures semantic features, while the lower branch enhances computational efficiency and local feature representation. This hybrid design allows the model to effectively capture both global and local features, improving segmentation accuracy. The residual connection ensures stable gradient flow, while the low computational cost makes the approach suitable for real-time robotic-assisted vascular interventions. The method was evaluated on the DRIVE and CHASE_DB1 datasets, with experimental results demonstrating superior performance in Acc (0.9653), F1-score (0.8156), and IoU (0.6886), compared to state-of-the-art methods, while maintaining a lightweight architecture suitable for real-time clinical applications.