Let's Not Waste Time: Using Temporal Information in Clustered Activity Estimation with Spatial Adjacency Restrictions...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Modeling

Background:

• Functional magnetic resonance imaging (fMRI) measures brain activity.
• Accurate brain parcellation is crucial for understanding neural networks.
• Existing methods may lack spatial contiguity or efficient temporal modeling.

Purpose of the Study:

• To extend the CAESAR Bayesian approach for whole-brain fMRI functional parcellation.
• To incorporate Gaussian process (GP) priors for modeling temporally smooth hemodynamic signals.
• To address computational challenges of GP inference in long fMRI time series.

Main Methods:

• Utilized distance-dependent Chinese restaurant processes (dd-CRPs) for flexible, spatially constrained partitioning.
• Implemented efficient GP inference to model hemodynamic signals, mitigating cubic time-point scaling.
• Employed a population Monte-Carlo algorithm to accelerate convergence.

Main Results:

• Demonstrated the benefits of CAESAR with GP priors using simulated data.
• Successfully parcellated resting-state fMRI data from the Human Connectome Project.
• CAESAR achieved robust and scalable whole-brain clustering of fMRI timecourses.

Conclusions:

• The extended CAESAR method provides a robust and scalable Bayesian framework for fMRI brain parcellation.
• Efficient GP inference and population Monte-Carlo methods overcome computational limitations.
• This approach yields physiologically meaningful and spatially contiguous brain parcellations.