<p>Based on the U-Net architecture and its Transformer-integrated variant, excellent performance has been demonstrated in medical image segmentation tasks. However, there still exists a semantic gap between the feature maps passed through the decoder and skip connections, while the network cannot effectively utilize inherent spatial information within the images. To address the above issues, we propose a medical image segmentation network that integrates residual axial attention and edge prediction. Firstly, for the challenges of blurred target boundaries and complex variations in medical images, a channel-spatial transformer based on residual axial attention is designed. This module enhances the feature maps transmitted through skip connections by introducing positional information and integrating features from multiple stages in the encoder. Secondly, a semantic alignment module based on residual axial attention is developed to mitigate the semantic gap issue in the encoder-decoder architecture. Additionally, by incorporating the edge prediction loss function, region information is used as a geometric constraint for the image segmentation task to encourage the model to generate continuous and clear target boundaries. Finally, the proposed method is evaluated on five different datasets, including binary and multi-class segmentation tasks. Experimental results demonstrate that compared to state-of-the-art segmentation methods, this approach achieves higher evaluation scores and finer segmentation results on various public datasets while reducing standard deviation.</p>