Multilevel hybrid principal components analysis for region-referenced functional electroencephalography data.

Area of Science:

• Neuroscience
• Biostatistics
• Data Science

Background:

• Electroencephalography (EEG) data are high-dimensional with multi-level structures.
• Current analysis methods often oversimplify EEG data by focusing on specific regions or features.
• This simplification can obscure comprehensive differences in brain signals.

Purpose of the Study:

• To propose a novel dimension reduction technique for complex EEG data.
• To develop a method that captures both between- and within-subject variability.
• To enhance the analysis of functional brain data across multiple conditions or visits.

Main Methods:

• Introduced Multilevel Hybrid Principal Components Analysis (M-HPCA).
• M-HPCA combines vector and functional principal components analysis.
• Employs a mixed-effects modeling framework with a minorization-maximization algorithm and bootstrap for estimation.

Main Results:

• M-HPCA effectively decomposes total variation into between- and within-subject variance.
• Demonstrated M-HPCA's applicability in analyzing event-related potentials (ERPs) and power spectral densities (PSDs) in autism studies.
• Simulations confirmed the finite sample properties of the proposed methodology.

Conclusions:

• M-HPCA provides a more comprehensive approach to analyzing high-dimensional EEG data.
• The method allows for detailed examination of brain signal differences across individuals and experimental conditions.
• M-HPCA offers a robust framework for neuroscientific research involving complex functional data.