Group Surrogate Data Generating Models and similarity quantification of multivariate time-series: A resting-state...

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Non-invasive brain analysis is crucial for understanding brain function and diseases.
• Existing methods for analyzing multivariate time-series data have limitations when applied to group datasets.

Purpose of the Study:

• To develop novel computational techniques for analyzing group-level brain dynamics from resting-state functional magnetic resonance imaging (rs-fMRI) data.
• To introduce a new similarity measure for comparing multivariate time-series and assess its utility in subject identification and data analysis.

Main Methods:

• Extended vector auto-regressive surrogate techniques to group data using a Group Surrogate Data Generating Model (GSDGM).
• Defined and implemented a Multivariate Time-series Ensemble Similarity Score (MTESS) for comparing multivariate time-series.
• Applied GSDGM and MTESS to human and marmoset rs-fMRI datasets from the Human Connectome Project.

Main Results:

• GSDGM generated biologically plausible human brain dynamics.
• MTESS demonstrated high accuracy in subject identification and revealed similarity differences between cortical and subcortical regions.
• Group surrogate approaches were confirmed to generate plausible group centroid time-series.
• GSDGM and MTESS successfully performed fingerprint analysis, distinguishing normal and outlier rs-fMRI sessions.

Conclusions:

• The developed GSDGM and MTESS provide powerful tools for analyzing group brain dynamics.
• These methods enhance the potential for accurate in silico simulations and clinical applications in neuroscience.
• The techniques facilitate robust data fingerprinting and outlier detection in large-scale neuroimaging datasets.