Transformer-based functional time series modeling to unveil dynamic brain state transitions
Summary
This study used a novel transformer model to analyze brain activity, revealing dynamic changes in the default mode network during brain state transitions. These findings highlight crucial shifts in brain connectivity.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Cognitive Science
Background:
- Brain states are dynamic and understanding their transitions is key to cognitive function.
- Identifying these transitions requires advanced analytical methods for time-series brain data.
Purpose of the Study:
- To investigate dynamic transitions between brain states using a transformer-based model.
- To identify brain regions and connectivity patterns associated with these transitions.
Main Methods:
- Utilized a blood-oxygen-level-dependent transformer model (BolT) for time-series prediction.
- Extracted importance weights to identify highly and lowly contributed time points.
- Constructed connectivity matrices and calculated degree centrality for selected time points.
Main Results:
- Significant differences in degree centrality were found in the default mode network.
- Decreased degree centrality was observed at highly contributed time points.
- The default mode network shows reorganization during brain state transitions.
Conclusions:
- The study demonstrates the utility of the BolT model for analyzing dynamic brain states.
- Connectome reorganization within the default mode network is a key feature of brain state transitions.
- These findings advance our understanding of the neural basis of cognitive function.
