MGCL-DA

Multi-view Graph Contrastive Learning with Dynamic Self-aware and Cross-sample Topology Augmentation for Brain Disorder Diagnosis

Hao Zhang¹, Xiaoyun Liu², Shuo Huang¹, Yonggui Yuan², Daoqiang Zhang³, Li Zhang¹

¹ Nanjing Forestry University
² Southeast University
³ Nanjing University of Aeronautics and Astronautics

Overview

MGCL-DA is a multi-view graph contrastive learning framework for rs-fMRI-based brain disorder diagnosis. It dynamically constructs self-aware and cross-sample topology augmentations and applies semantic-aware contrastive constraints to learn discriminative brain-network representations. The work was accepted by MICCAI 2025 as an Early Accept paper.

Method Highlights

• Complementary topology augmentation: models individual-specific patterns and inter-subject functional heterogeneity.
• Dynamic view updating: progressively refines the augmented brain-network representations during training.
• Multi-view contrastive learning: uses min-max constraints to preserve both shared and complementary semantics.

Repository Structure

Network module	Description
net/model.py	Main MGCL-DA architecture with three ST-GCN branches
net/SelfAwareAugmented.py	Self-aware topology augmentation
net/CrossSampleAugmented.py	Cross-sample topology augmentation
net/DynamicUpdate.py	Dynamic augmentation update mechanism
net/tgcn.py	Temporal graph convolution layer

News

• Jun. 2025 — MGCL-DA was accepted by MICCAI 2025 as an Early Accept paper.

Citation

If you find this work useful, please cite:

@InProceedings{ZhaHao_Multiview_MICCAI2025,
  author    = {Zhang, Hao and Liu, Xiaoyun and Huang, Shuo and Yuan, Yonggui and Zhang, Daoqiang and Zhang, Li},
  title     = {Multi-view Graph Contrastive Learning with Dynamic Self-aware and Cross-sample Topology Augmentation for Brain Disorder Diagnosis},
  booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
  year      = {2025},
  publisher = {Springer Nature Switzerland},
  volume    = {LNCS 15971},
  pages     = {532--542}
}

License

This project is released under the MIT License.