Related Experiment Video
Updated: Jan 16, 2026
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
GraSTI-ACL: Graph spatial-temporal infomax with adversarial contrastive learning for brain disorders diagnosis based
Biao He1, Erni Ji2, Xiaofen Zong3
1School of Biomedical Engineering, Medical School, Shenzhen University, Shenzhen, 518060, Guangdong, China; Guangdong Provincial Key Laboratory of Biomedical Measurements and Ultrasound Imaging, Shenzhen, 518060, Guangdong, China.
This study introduces Graph Spatial-Temporal Infomax (GraSTI) with adversarial contrastive learning for brain disorder diagnosis using resting-state fMRI. The method enhances diagnostic accuracy for Alzheimer's, MDD, and bipolar disorder by capturing spatial and temporal brain network dynamics.
Area of Science:
- Neuroimaging
- Machine Learning
- Computational Neuroscience
Background:
- Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for brain disorder research, offering valuable spatial and temporal data.
- Graph neural networks (GNNs) show promise for analyzing functional connectivity networks (FCNs) from rs-fMRI.
- Traditional static FCNs struggle to capture dynamic brain activity patterns, limiting diagnostic potential.
Purpose of the Study:
- To propose a novel graph augmentation strategy, GraSTI, for improved FCN construction from rs-fMRI.
- To integrate GraSTI into an adversarial contrastive learning (ACL) framework for enhanced diagnostic accuracy.
- To evaluate the method's effectiveness in diagnosing Alzheimer's disease (AD), major depressive disorder (MDD), and bipolar disorder (BD).
Main Methods:
- Developed Graph Spatial-Temporal Infomax (GraSTI) based on the information bottleneck principle to preserve global and local brain network information.
- Integrated GraSTI within an adversarial contrastive learning framework (GraSTI-ACL) to balance representation effectiveness and robustness.
- Applied the GraSTI-ACL method to rs-fMRI datasets from patients with AD, MDD, and BD.
Main Results:
- GraSTI-ACL achieved significant diagnostic accuracy improvements across AD (0.13%–23.56%), MDD (1.23%–13.81%), and BD (2.53%–24.53%) compared to existing methods.
- The method demonstrated strong interpretability, identifying key brain regions and connections relevant to specific disorders.
- The approach effectively captures both spatial and temporal dynamics in functional brain connectivity.
Conclusions:
- GraSTI-ACL offers a powerful and interpretable approach for brain disorder diagnosis using rs-fMRI.
- The method's ability to integrate spatial and temporal information advances the application of GNNs in neuroimaging.
- This work highlights the potential of adaptive graph augmentation and contrastive learning for clinical neuroscience.
