Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Brain Network Classification Based on Graph Contrastive Learning and Graph Transformer

ZhiTeng Zhu, Lan Yao

TL;DR

This work tackles brain network classification under data scarcity by leveraging rs-fMRI-derived connectomes. It introduces PHGCL-DDGformer, which combines adaptive graph augmentation with edge perturbation and attribute masking, a dual-domain graph transformer that fuses local GCN with Gaussian-attentive global attention, and a graph contrastive learning objective that integrates feature-level InfoNCE $\mathcal{L}_G$ and topology-level $\mathcal{L}_{Topo}$ losses alongside the supervised cross-entropy $\mathcal{L}_{CE}$. On ABIDE and ADHD-200 datasets, the method achieves state-of-the-art performance by effectively modeling local, global, and topological information in brain networks. The approach provides robust graph representations for brain disease classification with limited data and points to future work on higher-order graph structures and efficiency improvements across broader neurological disorders.

Abstract

The dynamic characterization of functional brain networks is of great significance for elucidating the mechanisms of human brain function. Although graph neural networks have achieved remarkable progress in functional network analysis, challenges such as data scarcity and insufficient supervision persist. To address the limitations of limited training data and inadequate supervision, this paper proposes a novel model named PHGCL-DDGformer that integrates graph contrastive learning with graph transformers, effectively enhancing the representation learning capability for brain network classification tasks. To overcome the constraints of existing graph contrastive learning methods in brain network feature extraction, an adaptive graph augmentation strategy combining attribute masking and edge perturbation is implemented for data enhancement. Subsequently, a dual-domain graph transformer (DDGformer) module is constructed to integrate local and global information, where graph convolutional networks aggregate neighborhood features to capture local patterns while attention mechanisms extract global dependencies. Finally, a graph contrastive learning framework is established to maximize the consistency between positive and negative pairs, thereby obtaining high-quality graph representations. Experimental results on real-world datasets demonstrate that the PHGCL-DDGformer model outperforms existing state-of-the-art approaches in brain network classification tasks.

Brain Network Classification Based on Graph Contrastive Learning and Graph Transformer

TL;DR

This work tackles brain network classification under data scarcity by leveraging rs-fMRI-derived connectomes. It introduces PHGCL-DDGformer, which combines adaptive graph augmentation with edge perturbation and attribute masking, a dual-domain graph transformer that fuses local GCN with Gaussian-attentive global attention, and a graph contrastive learning objective that integrates feature-level InfoNCE and topology-level losses alongside the supervised cross-entropy . On ABIDE and ADHD-200 datasets, the method achieves state-of-the-art performance by effectively modeling local, global, and topological information in brain networks. The approach provides robust graph representations for brain disease classification with limited data and points to future work on higher-order graph structures and efficiency improvements across broader neurological disorders.

Abstract

The dynamic characterization of functional brain networks is of great significance for elucidating the mechanisms of human brain function. Although graph neural networks have achieved remarkable progress in functional network analysis, challenges such as data scarcity and insufficient supervision persist. To address the limitations of limited training data and inadequate supervision, this paper proposes a novel model named PHGCL-DDGformer that integrates graph contrastive learning with graph transformers, effectively enhancing the representation learning capability for brain network classification tasks. To overcome the constraints of existing graph contrastive learning methods in brain network feature extraction, an adaptive graph augmentation strategy combining attribute masking and edge perturbation is implemented for data enhancement. Subsequently, a dual-domain graph transformer (DDGformer) module is constructed to integrate local and global information, where graph convolutional networks aggregate neighborhood features to capture local patterns while attention mechanisms extract global dependencies. Finally, a graph contrastive learning framework is established to maximize the consistency between positive and negative pairs, thereby obtaining high-quality graph representations. Experimental results on real-world datasets demonstrate that the PHGCL-DDGformer model outperforms existing state-of-the-art approaches in brain network classification tasks.

Paper Structure

This paper contains 16 sections, 12 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. The PHGCL-DDGformer Model
  4. Graph Augmentation via Importance-Driven Perturbations
  5. Dual-Domain Graph Transformer Encoder
  6. Graph Contrastive Learning
  7. Experiments
  8. Experimental Settings
  9. Performance Comparison
  10. Parameter Analysis
  11. Impact of Sparsity Ratio on Model Performance
  12. Impact of Encoding Layers $L$
  13. Impact of Contrastive Constraint Hyperparameters $\lambda_{1}$ and $\lambda_{2}$
  14. Ablation Study
  15. Interpretability Analysis
  16. ...and 1 more sections

Figures (5)

  • Figure 1: Framework of the PHGCL-DDGformer model
  • Figure 2: Architecture of the proposed DDGformer encoder
  • Figure 3: Impact of correlation matrix sparsity ratio on PHGCL-DDGformer model performance
  • Figure 4: Impact of contrastive constraint hyperparameters $\lambda_{1}$ and $\lambda_{2}$ on PHGCL-DDGformer model performance
  • Figure 5: Visualization of brain region attention coefficients under CC200 template: Color intensity reveals autism spectrum disorder (ASD) relevance