ETC-Net: an efficient collaborative transformer and convolutional network combining edge constraints for medical image segmentation

Medical image segmentation is a fundamental task in auxiliary diagnostic medical endeavors. In the domain of medical image segmentation, the U-Net network, combined with the Transformer architecture, has emerged as the dominant model. Its strength lies in its ability to effectively integrate both local and global contextual information. This integration significantly improves the overall efficacy of medical image segmentation. However, there are still some challenges in comprehensive modeling of global and local features with the combination of transformer and U-Net: 1) due to the discrete nature of the combination, it is difficult to balance the importance of global and local features; 2) deeper feature encoding leads to the neglect of target edge details, resulting in blurry segmentation boundaries. To solve the above problems, we propose a feature collaborative medical image segmentation network called ETC-Net (Efficient Transformer with Convolutional Network that combines edge constraints). Firstly, the Convolutional Neural Network and Transformer branches are added in parallel to the full convolutional attention-based U-Net model to extract global and local features, respectively. This is beneficial for exploring different features and retain the important information. The significance module is then designed to additionally supervise the prediction of the target edges to compensate for the information lost at the target edges, thus improving the model's ability to learn detailed features. Experiments conducted on cardiac images, pathological images, and H&E stained tissue image datasets demonstrate the model's effectiveness, with Dice scores of 93.48, 91.44, and 79.29%, respectively, which are superior to compared models. The source code will be made available at https://github.com/shilong1202/ETC-Net.