A novel graph neural network framework for resting-state functional MRI spatiotemporal dynamics analysis

Background:

Graph neural networks (GNNs) are essential for studying brain functional connectivity and neural network activity, but the high dimensionality and complexity of resting-state functional magnetic resonance imaging (rs-fMRI), coupled with its spatiotemporal dynamic characteristics, pose challenges for traditional GNNs in spatiotemporal dynamic analysis.

Methods:

This paper proposes a novel GNN framework aimed at the spatiotemporal dynamic analysis of rs-fMRI data. The framework constructs static spatial graphs and dynamic temporal graphs from rs-fMRI data. Then, a GNN is employed to extract spatial features, and a long short-term memory network (LSTM) is used to capture dynamics temporal features. Additionally, a diffusion connection strategy is implemented to facilitate the interaction between static spatial information and dynamic temporal information, enhancing the model’s stability and generalization capability. To achieve effective fusion of spatiotemporal information, an adaptive fusion method based on a self-attention mechanism is introduced, which improves the model’s ability to represent complex spatiotemporal patterns.

Results:

Experimental results on two public datasets demonstrate that this framework performs excellently in brain region connectivity classification task. Compared to traditional methods and existing GNN models, this framework significantly improves classification accuracy and model robustness, proving its superiority and practicality in the analysis of rs-fMRI data.

Conclusion:

The proposed GNN framework provides an effective tool for spatiotemporal dynamic analysis of rs-fMRI data. This method not only enhances the understanding of complex brain network structures but also shows great potential in practical applications, promising to advance further research in brain science.