A Tensor Decomposition-Based Approach for Detecting Dynamic Network States From EEG.

Area of Science:

• Neuroscience
• Cognitive Science
• Data Science

Background:

• Functional connectivity (FC) is crucial for understanding brain processes.
• Existing FC research often assumes static networks, but empirical evidence suggests dynamic changes.
• Understanding the dynamics of FC is key to comprehending brain network formation and dissolution.

Purpose of the Study:

• To develop and validate a method for characterizing the dynamics of functional connectivity networks (FCNs).
• To identify critical time points (change points) where FCNs exhibit significant alterations.
• To summarize FCN states between identified change points.

Main Methods:

• A two-step approach using tensor representation of FCNs across time and subjects.
• Identification of change points via subspace distance on low-rank tensor approximations.
• Network summarization through tensor-matrix projections.

Main Results:

• Application to electroencephalogram (EEG) data during a cognitive control task.
• Detected change points align with known error-related negativity (ERN) intervals.
• Significant connectivities identified in medial-frontal regions, consistent with ERN amplitude measures.

Conclusions:

• The proposed tensor-based method offers superior change-point detection and state summarization compared to matrix-based methods like singular value decomposition.
• This method effectively captures the topological structure of FCNs.
• The findings highlight the importance of dynamic network analysis in neuroscience.