MSCANet: A Multi-Branch Space Cross Attention Network for Medical Image Segmentation

In recent years, with the wide application of U-Net network architecture combined with channel or spatial attention mechanism in the field of medical image segmentation, the segmentation accuracy has been greatly improved. Among them, the channel or spatial attention mechanism can be more effective in extracting feature representations. However, for cross-channel network relations, if only channel dimension reduction operation is carried out, there is a high probability that some features of depth vision will be lost. Therefore, in this paper, we propose MSCANet, a multi-branch cross-spatial U-shaped attention network that preserves features on each channel while maintaining low computation. We have conducted extensive experiments on CVC-ClinicDB, 2018 Data Science Bowl and ISI-2018 datasets to verify the effectiveness and robustness of MSCANet in medical image segmentation tasks.