Abstract: Aiming at the problem that convolutional operations in UNet structures are insensitive to global feature information and cannot get long-distance dependencies in images, this paper proposes a medical image segmentation model based on U-shaped structures, which adopts the method of transformer block and skip connection, that can extract global feature information and local feature information in the image and obtain long-distance dependencies in the image. The resolution of the spatial feature map is changed and restored by sampling up and down. Skip connection fuse multi-scale feature information to further integrate information in the data. The experiments on the ISIC2017 dataset demonstrate that the Jaccard similarity coefficient of this paper proposed TUnet model is improved by 0.755 8 compared to some existing advanced medical image segmentation models. It is verified that the addition of the transformer module in the model makes the long-distance dependencies in the image extracted, and the model has a good application prospect.