基于空间通道自适应特征的肝脏病理图像分割网络OA

In view of the problems such as high similarity between the lesion area and the surrounding tissues,low contrast,and blurred boundaries in liver pathological images,this study proposes a liver pathological segmentation network based on spatially and channel-adaptive features.Through the hybrid calibration attention mechanism,the network can adaptively select the feature information calibrated in both spatial and channel dimensions,which is bene-ficial for the encoder to capture the important features related to liver lesions.Additionally,the atrous spatial pyramid pooling module is introduced at the deepest layer of the encoder to compensate for the missing multi-scale information in high-level features,thereby improving the segmentation accuracy of the model.Comparative experiments and abla-tion experiments are conducted on a private liver dataset,a public liver dataset,and two other public pathological data-sets for the proposed network.The experimental results show that,compared with other methods,the segmentation re-sults of the proposed network are better,and it effectively solves the problem of hepatocellular carcinoma segmentation.