Modelling variability in functional brain networks using embeddings

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Functional neuroimaging estimates cognitive networks but often at a group level.
• Existing methods overlook subpopulation structures within recording sessions.
• Individual session variability is crucial for understanding cognition.

Purpose of the Study:

• To develop a method for modeling individual functional neuroimaging sessions.
• To explicitly account for subpopulation structure in functional network estimation.
• To create session "fingerprints" for clustering similar functional networks.

Main Methods:

• Proposed HIVE (HMM with Integrated Variability Estimation).
• Utilized embedding vectors, inspired by Natural Language Processing.
• Applied a hierarchical Hidden Markov Model (HMM) to estimate dynamic functional networks.

Main Results:

• HIVE successfully uncovered true subpopulation structures in simulated data.
• Demonstrated improved performance compared to existing functional network estimation approaches.
• Identified meaningful sources of variation across a population using real magnetoencephalography data.

Conclusions:

• HIVE offers a powerful new approach for modeling individual sessions in functional neuroimaging.
• The method effectively leverages group-level information while preserving individual session characteristics.
• Session fingerprints derived from HIVE reflect significant population variability.