Semantic-spatial feature-fused cortical surface parcellation: a scale-unified spatial learning network with boundary contrastive loss
Medical & Biological Engineering & Computing · 2025
DOI: 10.1007/s11517-024-03242-5 ↗Kurz gefasst
Fachartikel in der Zeitschrift Medical & Biological Engineering & Computing, 2025. Verfasst von Hailiang Ye und 4 weiteren Autor:innen. Beteiligte Einrichtung: Department of Applied Mathematics, College of Sciences, China Jiliang University, Hangzhou, China. Hauptthemen laut MEDLINE: Neural Networks, Computer, Cerebral Cortex, Spatial Learning, Image Processing, Computer-Assisted. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 47 Literaturangaben. Gefördert von: Open Research Fund of Key Laboratory of Intelligent Education Technology and Application of Zhejiang Province, Zhejiang Normal University, National Natural Science Foundation of China.
Abstract (Originalsprache)
The cortical surface parcellation provides prior guidance for studying mental disorders and human cognition. Graph neural networks (GNNs) have gained popularity in this task to preserve its spatial structure. However, previous GNNs struggled to effectively exploit the information contained in the complex spatial structure of the cortical surface and generally encountered an uneven node distribution issue. Meanwhile, labeling boundary nodes was also identified as a widespread problem in this task. Accordingly, this paper develops a scale-unified spatial learning network with a boundary contrastive loss (SSLNet) for cortical surface parcellation. Its core is the scale-unified spatial learning module. It devises neighbor feature extraction and aggregation strategies by fully integrating spatial coordinates and semantic structure to learn effective spatial features of local neighborhoods. More importantly, spatial scale unification is incorporated into this module to mitigate the negative effect on spatial learning caused by node distribution differences among local areas. Additionally, a universal boundary contrastive loss is constructed, enhancing the feature discriminability of boundary nodes by constraining them to be close to the same class nodes and apart from different class nodes in the feature space. It considerably improves boundary performance without increasing parameters or changing the network structure. Experiments regarding public Mindboggle demonstrate that the dice score and accuracy of SSLNet achieve 89.8 % and 90.89 % , respectively, surpassing existing methods.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Applied Mathematics, College of Sciences, China Jiliang University, Hangzhou, China.
- Key Laboratory of Intelligent Education Technology and Application of Zhejiang Province, Zhejiang Normal University, Jinhua, China.
- School of Mathematical and Physical Sciences, Macquarie University, Sydney, NSW, Australia.
- Department of Applied Mathematics, College of Sciences, China Jiliang University, Hangzhou, China. icteam@163.com.
Förderung
- Open Research Fund of Key Laboratory of Intelligent Education Technology and Application of Zhejiang Province, Zhejiang Normal University
- National Natural Science Foundation of China
Zitieren
APA
Ye, H., Liu, S., Li, M., Zhu, H., & Cao, F. (2025). Semantic-spatial feature-fused cortical surface parcellation: a scale-unified spatial learning network with boundary contrastive loss. Medical & Biological Engineering & Computing. https://doi.org/10.1007/s11517-024-03242-5
Vancouver
Ye H, Liu S, Li M, Zhu H, Cao F. Semantic-spatial feature-fused cortical surface parcellation: a scale-unified spatial learning network with boundary contrastive loss. Medical & Biological Engineering & Computing. 2025. doi:10.1007/s11517-024-03242-5
BibTeX
@article{Ye2025,
author = {Ye, Hailiang and Liu, Siqi and Li, Ming and Zhu, Houying and Cao, Feilong},
title = {Semantic-spatial feature-fused cortical surface parcellation: a scale-unified spatial learning network with boundary contrastive loss},
journal = {Medical & Biological Engineering & Computing},
year = {2025},
doi = {10.1007/s11517-024-03242-5}
}
EUVIMED