CAB-Net: Channel Attention Block Network for Pathological Image Cell Nucleus Segmentation

In histopathological image analysis, cell nucleus segmentation plays an important role in the clinical analysis and diagnosis of cancer. However, due to the different morphology of cells, uneven staining and the existence of a large number of dense nuclei, it is still challenging to accurately segment the nucleus. In order to learn more specific key feature information during the training process, this paper proposed a network model called the CAB-Net that uses channel attention to enhance the learning of feature information on each channel. The network uses the channel attention mechanism to extract the key features in each channel, generates weights to judge the importance of the features, and then weights them into the original image. This aims to strengthen the extraction of key information, thereby generating a more characteristic feature map, and enabling the model to make more accurate judgments. We added a boundary smoothness constraint, in order to better identify cell nuclei with unclear boundaries and achieve more accurate cell nucleus segmentation. The experimental results show that the method in this paper achieves good performance on the cell segmentation data set