Edge and Density-Aware Attention Network for Automatic Liver Segmentation

Aiming at solving the inefficiency and subjectivity of doctors' manual liver segmentation in liver disease management, and considering the importance of liver segmentation in disease prevention, diagnosis, treatment and prognosis evaluation, this paper proposes an Edge and Density-Aware Attention Network (EDAA-Net). We improved EDAA-Net on the basis of U-Net network. In particular, we adopted the deep residual structure instead of the traditional convolution block to solve the problem of gradient vanishing. In order to enhance the sensitivity of the network to edge and density information and improve prediction accuracy, we designed an Edge and Density-Aware Attention module (EDAAM). To capture contextual information at different scales, we introduced the Atrous Spatial Pyramid Pooling (ASPP) module. Further, we used a Squeeze-and-Excitation (SE) module to adaptively adjust the channel feature response to highlight key features of a particular segmentation task. To verify the effectiveness and performance of the model, we conducted tests on two publicly available datasets. The experimental results have shown that compared with the current state-of-the-art methods, EDAA-Net have performed better, and have also achieved satisfactory results when dealing with small liver areas.