Semantic Segmentation of Liver Magnetic Resonance Images Based on Hybrid Attention Mechanism
Chenghao Zhang, Jin Li, Lingfei Wang, Yu Xin Zhang, Peng Wang · 2023
Liver segmentation has been a hot research topic in medical image processing technology, and liver segmentation is the basis for liver vascular segmentation, liver tumor segmentation, and subsequent 3D reconstruction. From the viewpoint of data distribution, the background of liver images is complex and variable, and the boundary with surrounding tissues is blurred; from the viewpoint of the segmentation method, the U-Net network structure leads to the absence of important features due to the adoption of the traditional pooling operation. To address the above limitations, this paper proposes a network structure that mixes U-Net with the CBAM attention mechanism to enhance feature extraction in the encoding part, strengthen the acquisition of global contextual information in the bottleneck layer, and promote the fusion of deep and shallow features in the decoding part. After experiments, all segmentation indexes of the improved network are improved compared with the benchmark network U-Net, and it is proved that CBAMU-Net can be well used for the segmentation of MR liver images.