Robust and smooth Couinaud segmentation via anatomical structure-guided point-voxel network

Xukun Zhang, Sharib Ali, Tao Liu, Xiao Zhao, Zhiming Cui, Minghao Han, Shuwei Ma, Jingyi Zhu, Yanlan Kang, Le Wang, Xiaoying Wang, Lihua Zhang

Abstract: Highlights•Introduced a novel automatic Couinaud liver segmentation framework, leveraging a dual-branch point-voxel fusion for enhanced spatial and semantic modeling.•Developed a local attention module and a novel feature-level distance loss, ensuring smooth and cohesive segmentation boundaries.•Demonstrated superior segmentation performance and robustness on public liver datasets (3Dircadb, LiTS, MSD8). Specifically, in out-of-distribution (OOD) testing of LiTS dataset, our method exceeded the voxel-based 3D UNet by approximately 20% in Dice score, and outperformed the point-based PointNet2Plus by approximately 8% in Dice score.