Related Experiment Video
Updated: May 2, 2026
11:50
A Standardized Pipeline for Examining Human Cerebellar Grey Matter Morphometry using Structural Magnetic Resonance Imaging
Published on: February 4, 2022
3.9K
Semantic-spatial feature-fused cortical surface parcellation: a scale-unified spatial learning network with boundary
Hailiang Ye1, Siqi Liu1, Ming Li2
1Department of Applied Mathematics, College of Sciences, China Jiliang University, Hangzhou, China.
Medical & Biological Engineering & Computing
|November 16, 2024
Summary
This study introduces a novel Scale-Unified Spatial Learning Network (SSLNet) for cortical surface parcellation, improving accuracy in mapping brain regions. SSLNet effectively addresses node distribution challenges and enhances boundary node labeling for better cognitive and mental disorder research.
Area of Science:
- Neuroscience
- Computer Science
- Medical Imaging
Background:
- Cortical surface parcellation is crucial for understanding brain function and mental disorders.
- Existing Graph Neural Networks (GNNs) face challenges with complex spatial structures and uneven node distribution.
- Accurate labeling of boundary nodes remains a significant issue in cortical parcellation.
Purpose of the Study:
- To develop an advanced network for cortical surface parcellation that overcomes limitations of previous GNNs.
- To improve the exploitation of spatial information and address uneven node distribution.
- To enhance the accuracy of boundary node identification in cortical mapping.
Main Methods:
- Development of a Scale-Unified Spatial Learning Network (SSLNet) incorporating a boundary contrastive loss.
- Implementation of a scale-unified spatial learning module with integrated spatial coordinates and semantic structure.
- Design of neighbor feature extraction and aggregation strategies with spatial scale unification.
- Construction of a universal boundary contrastive loss to improve feature discriminability of boundary nodes.
Main Results:
- SSLNet effectively learns spatial features by integrating spatial coordinates and semantic structure.
- Spatial scale unification mitigates learning issues caused by varying node distributions.
- The boundary contrastive loss enhances feature distinctiveness for boundary nodes without adding complexity.
- Experiments on the Mindboggle dataset show SSLNet achieving superior dice scores and accuracy compared to existing methods.
Conclusions:
- SSLNet offers a robust solution for cortical surface parcellation, improving spatial learning and boundary node accuracy.
- The proposed methods effectively address key challenges in GNN-based cortical mapping.
- SSLNet demonstrates significant potential for advancing research in human cognition and mental disorders.
