3D liver segmentation algorithm based on reconstructed convolutional network

To address the challenge of accurately distinguishing the liver from adjacent organs in CT images, which often have similar intensities, we introduce a novel liver CT image segmentation network that leverages reconstruction convolution and attention mechanisms. Building upon the 3D UNet framework, we enhance the network's feature processing and learning capabilities through attentional reconstruction convolution layers. Additionally, we propose a spatial gate attention module that combines the benefits of attention gates and spatial attention, effectively enhancing feature representation in key areas while suppressing irrelevant information. Rigorous evaluation on publicly available datasets, FLARE21 and LiTS17, yields Dice coefficients of 97.0% and 94.2%, respectively, confirming the superior accuracy of our approach in automatic liver segmentation.