Fig. 1
From: U-Net combined with multi-scale attention mechanism for liver segmentation in CT images
Overview of MSA-UNet architecture. In the encoder, we replaced the sequence of the two convolutional layers in the U-Net structure with the proposed MRSB. AASPP was a module used to capture contextual feature information, which was inserted between the encoder and the decoder. In addition, we did not use the skip connection in UNet, but used the proposed RASM. At the output of the model, we used a 1 × 1 convolution and a sigmoid activation function to obtain the final prediction result