2.5D cascaded context-based network for liver and tumor segmentation from CT images

The existing 2D/3D strategies still have limitations in human liver and tumor segmentation efficiency. Therefore, this paper proposes a 2.5D network combing cascaded context module (CCM) and Ladder Atrous Spatial Pyramid Pooling (L-ASPP), named CCLNet, for automatic liver and tumor segmentation from CT. First, we utilize the 2.5D mode to improve the training efficiency; Second, we employ the ResNet-34 as the encoder to enhance the segmentation accuracy. Third, the L-ASPP module is used to enlarge the receptive field. Finally, the CCM captures more local and global feature information. We experimented on the LiTS17 and 3DIRCADb datasets. Experimental results prove that the method skillfully balances accuracy and cost, thus having good prospects in liver and liver segmentation in clinical assistance.