MPA-Net: Multi-scale Pyramid Attention network for Liver Tumor Segmentation

Liver tumor is one of the most deadly cancers around the world. Since the traditional liver tumor segmentation method is not only time-consuming but also low-precision, a precise and automatic liver tumor segmentation model is urgently needed in clinical practice. To address these above, we present a novel Multi-scale Pyramid Attention network for liver tumor segmentation. This proposed method replaces the traditional convolution network into pyramid convolution, which simultaneously can process the input data and capture different levels of detail features. Moreover, the Squeeze-and-Excitation (SE) block is introduced to obtain the channel relevance between feature maps. We conduct extensive experiments to validate MPA-Net on the MICCAI 2017 Liver Tumor Segmentation Challenge dataset. A large number of experiments demonstrate that the MPA-Net achieves better performance than other state-of-the-art methods. The dice per case score and dice global score of liver tumor segmentation are 0.771 and 0.826.