Canonical Hidden Markov Model Networks for studying M/EEG

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Network Analysis

Background:

• Magneto/encephalography (M/EEG) reveals dynamic brain networks crucial for understanding brain function.
• Hidden Markov Models (HMMs) are effective for inferring reproducible brain networks from M/EEG data.
• Current HMM studies often use small, isolated datasets, limiting generalizability and increasing computational burden.

Purpose of the Study:

• To develop and validate a 'canonical HMM' as a standardized reference for analyzing M/EEG brain networks.
• To enable efficient comparison of brain network dynamics across diverse M/EEG datasets.
• To provide an open-access resource of canonical brain networks for the research community.

Main Methods:

• Trained Hidden Markov Models (HMMs) with varying states (4-16) on a large dataset of 1849 MEG recordings (N=621).
• Developed canonical HMMs applicable in both parcellated source space and sensor space.
• Demonstrated the application of the canonical HMM using independent M/EEG datasets.

Main Results:

• The canonical HMM successfully describes brain activity across diverse M/EEG datasets using a shared set of brain networks.
• The approach was validated on boutique MEG and EEG datasets, showing its utility in comparing individuals and studies.
• Canonical HMMs were made publicly available, facilitating broader research applications.

Conclusions:

• A canonical HMM provides a computationally efficient and reproducible framework for analyzing dynamic brain networks across M/EEG studies.
• This standardized approach allows for robust comparison of brain network states within and between datasets.
• The open-access canonical HMM resource promotes wider adoption and advancement in M/EEG-based brain network research.