Individual-Specific Areal-Level Parcellations Improve Functional Connectivity Prediction of Behavior

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Mapping

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) enables the creation of personalized brain parcellations.
• Previous work established a multi-session hierarchical Bayesian model (MS-HBM) for network-level parcellations.
• Areal-level parcellations are spatially localized, but their contiguity is debated.

Purpose of the Study:

• To extend the MS-HBM for estimating individual-specific areal-level brain parcellations.
• To evaluate different MS-HBM variants based on spatial contiguity assumptions.
• To assess the performance of these parcellations in predicting behavior and task-based fMRI.

Main Methods:

• Developed three variants of the MS-HBM for areal-level parcellations, considering different contiguity constraints.
• Estimated individual-specific parcellations using 10 minutes of rs-fMRI data.
• Validated parcellations against out-of-sample rs-fMRI and task-fMRI data, and assessed behavioral prediction accuracy.

Main Results:

• Individual-specific MS-HBM parcellations from 10 minutes of data outperformed other methods using 150 minutes.
• MS-HBM parcellations achieved superior behavioral prediction performance.
• The strictly contiguous MS-HBM showed the best resting-state homogeneity and task activation uniformity, with minor differences among variants in behavioral prediction.

Conclusions:

• Areal-level MS-HBMs effectively capture behaviorally relevant individual-specific brain features.
• These advanced parcellations surpass traditional group-level approaches.
• Publicly available models and parcellations facilitate further research in personalized brain mapping.