Hybrid principal components analysis for region-referenced longitudinal functional EEG data.

Area of Science:

• Neuroscience
• Biostatistics
• Developmental Psychology

Background:

• Electroencephalography (EEG) data have complex regional, functional, and longitudinal structures.
• Standard EEG analyses often reduce data dimensionality by averaging, losing valuable information.
• Autism Spectrum Disorder (ASD) research requires sophisticated methods to analyze neural data.

Purpose of the Study:

• To develop a novel hybrid principal components analysis (PCA) for region-referenced longitudinal functional EEG data.
• To avoid collapsing information across the regional, functional, and longitudinal dimensions of EEG.
• To analyze EEG data from a word segmentation paradigm in typically developing (TD) children and children with ASD.

Main Methods:

• Proposed a hybrid PCA combining vector and functional PCA for high-dimensional EEG data.
• Utilized a product of one-dimensional eigenvectors and eigenfunctions from marginal covariances.
• Employed a mixed-effects framework with bootstrap testing for group-level inference in sparse data.

Main Results:

• The hybrid PCA successfully decomposes region-referenced longitudinal functional EEG data without information loss.
• Analysis revealed significant group-region differences in brain activity between TD children and children with ASD.
• The method provides valuable insights into neural processing differences in ASD during speech perception.

Conclusions:

• The proposed hybrid PCA is a computationally feasible, non-parametric approach for complex EEG data.
• This method offers enhanced understanding of neural mechanisms in neurodevelopmental disorders like ASD.
• The framework is suitable for sparse data applications and group-level inference.