Back to all papers

VCon-Net: A Cerebral Artery Segmentation Network with Enhanced Topological Connectivity.

November 17, 2026pubmed logopapers

Authors

Fan L,Li B,Yang Y,Wen C,Sun H,Wang Z,Huang S,Wang T,Zhu S,Yang B,Zhang L,Gong Y,Liu J,Liu Y

Affiliations (4)

  • Department of Biomedical Engineering, College of Chemistry and Life Science, Beijing University of Technology, Beijing, China.
  • Cardiology, Peking University First Hospital, Beijing, China.
  • Center for Medical Metrology, National Institute of Metrology, Beijing, China.
  • Department of Biomedical Engineering, College of Chemistry and Life Science, Beijing University of Technology, Beijing, China. [email protected].

Abstract

Precise segmentation of cerebral artery from computed tomography angiography serves as a fundamental prerequisite for the clinical evaluation and morphological analysis of cerebrovascular diseases, particularly stroke. However, oversegmentation and undersegmentation routinely compromise topological connectivity and accuracy. This study aims to develop a novel deep learning network to mitigate these challenges, enhancing topological connectivity and accuracy. To enforce topological connectivity, we propose a novel skeleton points distance loss that combines skeleton points with a 3D Euclidean distance field to penalize oversegmentation and undersegmentation via interskeleton point distance constraints. Built upon the U-Net architecture, VCon-Net integrates an axialdynamic convolution module and a multiattention mechanism to enhance the perception of slender arteries and capture spatial-channel-temporal features. A total of 202 cases were included in this study, with 141 cases utilized for model development, 35 for independent testing and performance evaluation, and the remaining 26 analyzed as special clinical cases. On the independent test set, VCon-Net achieved a Dice of 0.8552, a Jaccard of 0.7513, a precision of 0.8568, a sensitivity of 0.8645, with HD<sub>95</sub> reduced to 3.16 mm, and β<sub>0</sub> error reduced to 2.85. Compared with existing segmentation methods, the proposed framework yields favorable performance. VCon-Net achieves cerebral artery segmentation with enhanced topological connectivity, providing a reliable morphological foundation, which offers an essential morphology basis for downstream clinical evaluations.

Topics

Journal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.