.png){kind=icon}
.png){kind=icon}
Return
Topo-UNet: A topology-aware multi-task network for pulmonary vessel segmentation
.png){kind=icon}
PRE
AI
L
Y
Y
W
J
Z
R
W
DOI：10.1016/j.artmed.2026.103439.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Introduce Topo-UNet network for precise pulmonary vessel segmentation and early disease diagnosis. • Combine Bidirectional Slice-wise ConvLSTM module with topology-aware auxiliary task to enhance vessel feature capture. • Experiments on CT and CTA of the ISICDM 2020 dataset show Dice over 90%, IoU over 83%, and model size only 68.51 MB. • Experiments on the CARVE14 and HiPaS 2025 dataset demonstrate model robustness. • Propose a grouping evaluation strategy for performance analysis of vessels of different sizes. • Introduce vessel refinement method, significantly improving the recognition rate of fine vessels.
Keywords:
Topo-UNet
pulmonary vessel segmentation
ConvLSTM
topology-aware
multi-task network
Journal
IF:
6.2
Papers:
2.5K
Citations:
7.8K
Organization
Cited Papers