Stable EEG Spatiospectral Patterns Estimated in Individuals by Group Information Guided NMF

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Engineering

Background:

• Electroencephalography (EEG) oscillations are crucial for understanding development, intelligence, cognitive states, and neural disorders.
• Nonnegative matrix factorization (NMF) is effective for group-level EEG spectral analysis, but lacks individual-level component optimization.
• Existing methods struggle to preserve individual EEG characteristics while ensuring cross-participant pattern correspondence.

Purpose of the Study:

• To introduce a novel framework, group-information guided NMF (GIGNMF), for subject-specific EEG component extraction.
• To preserve individual EEG characteristics and establish cross-participant pattern correspondence.
• To enhance the understanding of cognitive function and aid in clinical neuropathological diagnosis.

Main Methods:

• A three-stage framework utilizing group-level consensus patterns derived from standard NMF.
• An optimal procedure to determine the number of components.
• A multi-objective optimization strategy employing one-unit NMF with group-level patterns as references.

Main Results:

• Demonstrated the feasibility of GIGNMF in identifying EEG spatiotemporal patterns.
• Successfully extracted novel individual electrophysiological characteristics from synthetic and real EEG data.
• Validated performance using synthetic signals and electroencephalography recordings from Alzheimer's disease patients.

Conclusions:

• GIGNMF effectively extracts subject-specific EEG components while maintaining group-level information.
• The method provides novel insights into individual electrophysiological characteristics.
• GIGNMF shows promise for advancing the understanding of cognitive function and clinical diagnosis of neuropathological conditions.