Brain network dynamics fingerprints are resilient to data heterogeneity.

Area of Science:

• Neuroscience
• Data Science
• Biomarker Discovery

Background:

• Large multi-site neuroimaging datasets are crucial for understanding brain-behavior links and identifying biomarkers for neurological and psychiatric conditions.
• Variability across samples in these datasets can lead to biased or erroneous conclusions.

Purpose of the Study:

• To validate the estimation of individual brain network dynamics fingerprints using data-driven dynamical models.
• To assess sources of variability in large resting-state functional magnetic resonance imaging (rs-fMRI) datasets.

Main Methods:

• Utilized Hidden Markov Models (HMM) to analyze brain network dynamics in rs-fMRI data.
• Trained a stable HMM on a homogeneous dataset (Human Connectome Project) and applied it to a heterogeneous dataset with diverse scanning conditions.
• Examined the impact of scanning factors on inferring brain activity patterns.

Main Results:

• Replicated findings on non-random sequences of brain states.
• Demonstrated that time-varying brain activity patterns serve as robust, subject-specific fingerprints.
• Identified scanning factors that induce high variability in the data.

Conclusions:

• Large datasets can train models for interrogating subject-specific brain activity.
• Unique trajectories of brain activity changes can be recovered for each individual.
• Caution is advised as data acquisition methods influence the reliability of inferred brain patterns.