DASGC-Unet: An Attention Network for Accurate Segmentation of Liver CT Images

The precise segmentation of lesions can assist doctors to complete efficient disease diagnosis. Unet is widely used in the field of medical image segmentation due to its excellent feature fusion ability. However, the deep network based on Unet has poor ability to extract lesion features and insufficient segmentation accuracy. This is because the amount of medical image data is generally small, the lesion area is small, and Unet ignores the importance of different information. To overcome these shortcomings, we propose a Unet-based attention network for accurate segmentation of liver CT images. Specifically, we first creatively design an attention mechanism module (DASGC) that pays attention to both multi-scale spatial information and inter-channel information at the same time, which can give more weight to important feature information and perform the feature information screening task well. Secondly, based on the advantages of DASGC's efficient development of limited information, we masterly design an improved Unet network (DASGC-Unet) to solve the problem that the Unet network cannot effectively use less image information to complete accurate segmentation. Finally, on the LiTS2017 public dataset, our method achieves the best results on mIoU, IoU, and Dice coefficient compared to other advanced attention mechanism networks.