Cross-atlas Identification of Narrative Hubs via Multi-embedding Graph Models in fMRI Data

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Graph Theory

Background:

• Investigating brain processes in narrative comprehension is a key goal in cognitive neuroscience.
• Previous functional magnetic resonance imaging (fMRI) studies often overlook dynamic network properties, relying on correlation-based analyses.

Purpose of the Study:

• To present a new graph-based framework for identifying influential brain regions in narrative comprehension.
• To combine composite node importance scoring with node embedding algorithms for network analysis.

Main Methods:

• Developed and validated a graph-based framework using controlled simulations with stochastic block models (SBM).
• Assessed seven embedding algorithms for node influence, link prediction, and community detection.
• Applied the framework to fMRI data using Harvard-Oxford and Schaefer atlases for narrative comprehension analysis.

Main Results:

• The framework successfully identified influential cortical regions involved in narrative processing.
• Consistent engagement of default mode, salience, and limbic networks was observed across different stories and atlases.
• The findings highlight the central role of these networks in narrative comprehension.

Conclusions:

• The proposed graph-based framework offers a reliable and interpretable method for identifying key brain regions in cognitive tasks.
• This work bridges graph representation learning and cognitive neuroscience, providing a scalable basis for future research.
• The study advances our understanding of naturalistic cognition, dynamic brain connectivity, and linguistic features.