Multiregional integration in the brain during resting-state fMRI activity

Area of Science:

• Neuroscience
• Cognitive Neuroscience
• Computational Neuroscience

Background:

• Functional magnetic resonance imaging (fMRI) allows modeling dependencies between brain regions.
• Predicting brain activity from other regions is possible, but identifying unique multi-region integration remains challenging.
• Sparse models can improve the identification of key interregional dependencies by reducing false positives.

Purpose of the Study:

• To identify brain regions where activity uniquely depends on the integration of multiple predictor regions.
• To determine key regions contributing to observed activity through multiregional integration.
• To demonstrate the utility of whole-brain recursive feature elimination (RFE) for data-driven modeling.

Main Methods:

• Utilized resting-state fMRI data from 46 subjects.
• Applied whole-brain recursive feature elimination (RFE) to select minimal sets of ROIs predicting activity in a target ROI.
• Quantified multiregional integration by measuring the gain in prediction accuracy with multiple predictors versus single predictors.

Main Results:

• Identified regions exhibiting significant multiregional integration.
• Revealed fronto-parietal integration networks and limited integration in primary sensory areas.
• Observed redundancy between certain brain regions in predicting activity.

Conclusions:

• Whole-brain RFE effectively generates accurate, data-driven models with minimal ROIs.
• Cortical integration networks are identifiable during resting-state activity.
• The study highlights the complex interplay of brain regions during resting states.