【摘要】 提出了一种基于U型分割网络融合了空间注意力机制与通道注意力机制以及共享感知机（shared multi-layer perceptron,Shared-MLP）的SCAMU-net(spatial-channel attention mechanism U-shaped segmentation network)网络。为了解决训练样本有限和类别不平衡的问题，引入了数据增强技术，对传统的U-net进行改进，并与注意力机制结合。通过引入SCAM模块，注意力机制能够自动学习特征之间的相关性，并根据注意力权重对特征进行加权聚合，从而更好地捕捉到图像中不同位置的重要信息。这样可以使得模型能够更准确地定位病灶边界并识别出更细微的病变。所提出的方法在公开数据集Liver进行验证表明，该方法的Dice系数、IOU系数、精确率和召回率分别为90.13%、83.67%、89.97%、90.48%；与多种分割方法进行比较，该方法对CT图像中肝肿瘤的分割具有优势。更多还原

【Abstract】 This paper proposes a SCAMU-net(Spatial-Channel Attention Mechanism U-shaped segmentation network) that integrates spatial and channel attention mechanisms as well as a Shared Multi-layer Perceptron(Shared-MLP) into the U-shaped segmentation network. To address the challenges of limited training samples and class imbalance, data augmentation techniques are introduced.The traditional U-net is improved and combined with attention mechanisms. The SCAM module is introduced to enable the attention mechanism to automatically learn the correlation between features and aggregate them with weighted attention, thus better capturing important information from different positions in the image. This allows the model to more accurately locate lesion boundaries and identify finer lesions. Validation on the Liver dataset demonstrates that the proposed method achieves Dice coefficient,Intersection over Union coefficient, precision, and recall of 90.13%, 83.67%, 89.97%, and 90.48% respectively. Compared with various segmentation methods, this approach demonstrates advantages in segmenting liver tumors in CT images.更多还原