MS-HBM

MS-HBM estimates individual-specific areal-level cortical parcellations from resting-state functional magnetic resonance imaging (rs-fMRI) using a multi-session hierarchical Bayesian modeling framework to capture session and inter-subject variability.

Key Features:

• Hierarchical Bayesian Framework: Employs a multi-session hierarchical Bayesian approach for robust estimation of individual-specific cortical maps from rs-fMRI while modeling variability across sessions and individuals.
• Areal-Level Parcellation Estimation: Produces areal-level parcellations that focus on spatially localized cortical regions that do not span multiple lobes.
• Parcellation Variants: Provides three variants—Distributed MS-HBM (dMSHBM) allowing spatially distributed parcels, Contiguous MS-HBM (cMSHBM) enforcing strictly contiguous parcels, and Gradient-Infused MS-HBM (gMSHBM) incorporating gradient information to define parcel boundaries.
• Data Efficiency: Demonstrates the ability to estimate individual-specific parcellations from approximately 10 minutes of rs-fMRI data while outperforming approaches requiring up to 150 minutes for resting-state and task-related fMRI predictions.
• Behavioral Prediction Performance: Functional connectivity derived from MS-HBM parcellations achieves superior performance in predicting behavioral outcomes compared with alternative parcellation approaches.
• Homogeneity and Activation Uniformity: cMSHBM yields the highest resting-state homogeneity and most uniform task activation within parcels, while gMSHBM shows numerically superior behavioral prediction despite non-significant differences among variants.

Scientific Applications:

• Individual-specific brain mapping: Enables generation of subject-specific areal parcellations for studies requiring individualized cortical maps.
• Low-data cohort studies: Facilitates studies with limited scanning time or large cohorts by producing high-quality parcellations from short (≈10 min) rs-fMRI acquisitions.
• Functional connectivity and behavior linking: Supports extraction of functional connectivity features for prediction and analysis of behavioral outcomes.
• Comparison to group-level analyses: Provides individualized parcellations that can improve interpretation of resting-state and task-related fMRI results relative to group-level parcellations.

Methodology:

Implements a multi-session hierarchical Bayesian model applied to rs-fMRI to estimate individual-specific areal parcellations, with three explicit variants (dMSHBM, cMSHBM, gMSHBM) and gMSHBM incorporating gradient information to define parcel boundaries.