SwinDCA-Net: Leveraging Swin Transformer and Dynamic Connection Attention for Medical Image Segmentation

This paper introduces SwinDCA-Net, a novel architecture tailored for medical image segmentation, with a specific emphasis on vessel segmentation. The model incorporates Swin Transformer Modules into the U-Net framework, blending local feature extraction with global context modeling to enhance the segmentation of complex structures. Additionally, a Dynamic Connection Attention Module is implemented within the skip connections, allowing for adaptive reweighting of features across different scales, thereby improving feature fusion. Through extensive experiments, SwinDCA-Net consistently demonstrates superior performance, surpassing existing models in metrics such as F1-Score, sensitivity, and accuracy. These results confirm the model's ability to accurately segment thin and complex structures.