Related Experiment Video
Updated: Jul 27, 2026
Application of Granger Causality Analysis of the Directed Functional Connection in Alzheimer's Disease and Mild Cognitive Impairment
Published on: August 7, 2017
CI-GNN: A Granger causality-inspired graph neural network for interpretable brain network-based psychiatric diagnosis
Kaizhong Zheng1, Shujian Yu2, Badong Chen1
1National Key Laboratory of Human-Machine Hybrid Augmented Intelligence, National Engineering Research Center for Visual Information and Applications, and Institute of Artificial Intelligence and Robotics, Xi'an Jiaotong University, Xi'an, China.
We introduce a novel Granger causality-inspired graph neural network (CI-GNN) for brain network analysis. This interpretable model identifies causal subgraphs for psychiatric diagnosis, improving decision transparency and reliability.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Computational Psychiatry
Background:
- Graph Neural Networks (GNNs) are increasingly used for brain network-based psychiatric diagnosis.
- Existing GNN explainers often lack causal inference and can suffer from spurious correlations, limiting their clinical utility.
- There is a need for interpretable GNNs that provide faithful and causally relevant explanations.
Purpose of the Study:
- To develop a built-in interpretable GNN (CI-GNN) that identifies causally influential subgraphs for psychiatric diagnosis.
- To ensure explanations are faithful and directly related to diagnostic decisions, avoiding spurious correlations.
- To evaluate CI-GNN's performance against existing methods on synthetic and real-world brain disease datasets.
Main Methods:
- Proposed a Granger causality-inspired GNN (CI-GNN) within a graph variational autoencoder framework.
- Utilized a conditional mutual information (CMI) constraint to disentangle causal (α) and non-causal (β) subgraph representations.
- Theoretically justified the CMI constraint for capturing causal relationships.
Main Results:
- CI-GNN achieved superior performance compared to three baseline GNNs and four state-of-the-art explainers.
- Demonstrated improved reliability and conciseness of explanations, supported by clinical evidence.
- Successfully identified influential subgraphs causally linked to psychiatric conditions.
Conclusions:
- CI-GNN offers a robust and interpretable approach for brain network analysis in psychiatric diagnosis.
- The model provides causally-informed explanations, enhancing trust and clinical applicability of GNNs.
- CI-GNN represents a significant advancement in explainable AI for neuroscience and mental health.
