Research on point cloud guided segmentation method for liver,tumor and vessels
LI Yang
CHEN Chunxiao
BAI Shaokang
CHEN Bokai
WANG Liang
XIAO Yueyue
Abstract：To address the challenges of precise segmentation of the liver,tumors and vessels in preoperative planning for primary liver cancer ablation,we proposed a point cloud guided segmentation network(PCG-Net).Firstly,by using HU value thresholding and gradient computation,the three-dimensional CT image was transformed into a structured point cloud.Then,an improved RandLA-Net was utilized to encode the topological information of the point cloud,from which a feature map rich in cross-slice contextual infor-mation was generated through differentiable Gaussian splatting(GS).Finally,this feature map was subsequently fused with the original CT slices and fed into a U-Net integrated with a convolution block attention module(CBAM)to accomplish the segmentation task.On the 3D-IRCADb-01 dataset,PCG-Net achieved Dice similarity coefficient(DSC)of 0.934,0.618 and 0.725,and 95%Hausdorff distances(HD95)of 1.618,9.019 and 5.251 for the liver,tumors and vessels,respectively,superior overall performance against com-parative methods.The experimental results indicated that PCG-Net could effectively preserve the topological integrity of vascular seg-mentation and improve the accuracy and robustness of multi-target segmentation.The research can provide reliable technical support for precise surgical planning.
Keywords：Liver cancerMedical image segmentationPoint cloud guided segmentation networkIrreversible electroporationLiver segmentationTumor segmentationVessel segmentation
Publication Date：2026-04-30
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：7( 104-110 )