Coronary artery segmentation is vital for accurate vessel delineation, serving as a critical step in the diagnostic workflow and therapeutic planning for coronary artery disease. However, challenges such as the complex geometry of coronary vessels and low contrast in angiographic images hinder segmentation accuracy. To address these challenges, RDMAUNet is proposed, a deep learning model that combines residual deformable convolutions and multi-attention mechanisms for precise coronary artery segmentation. The residual deformable convolution block (RDBlock) dynamically adjusts the receptive field to more effectively model complex vascular morphology. The SCFusion module enhances feature representation by integrating spatial and channel-wise attention mechanisms, whereas the coordinate attention module, applied during upsampling, reinforces the perception of spatial structural information. The evaluation shows that RDMAUNet surpasses traditional methods across multiple metrics. The model shows strong potential for clinical applications, providing a reliable and efficient solution for automated coronary artery analysis and treatment planning.