Esophageal Image Segmentation with Dual Attention Based on TransUNet

Currently, the diagnosis of esophageal cancer tumors is mainly made by chest based CT scans. Because of the small area of the esophagus in chest-based CT images, it is difficult to determine the tumor region of esophageal cancer using traditional image segmentation networks. So we came up with the idea of combining the traditional U-Net architecture with Transformer. Although integrated variants perform well in many medical segmentation tasks, they lack the ability to take advantage of location and channel features. In order to solve the above problems, this paper proposes a new model for medical image segmentation, called CCA-TransUNet. And we combine the Coordinate Channel Attention(CCA) module and the Efficient Channel Attention (ECA) module into a new module, which we call the Dual Efficient Attention (DEA) module. CCA-TransUNet to Transformer and DEA module integrated into the traditional u-shaped structure, enhance the ability to extract the image position and channel features. By adding CCA modules in each skip connection layer, the network feature extraction capability is greatly enhanced. We have verified the superior performance of our network through numerous experiments on our collected esophageal dataset. We used CAA-TransUNet to perform experiments on our collected dataset, and the average DSC(%) reached 76.44% and the average HD95(mm) reached 3.45.