Tien Dung Nguyen: Extension of a decomposition method for bispectra and application to EEG data

BCCN Berlin / Technische Universität Berlin

Abstract

Cross-frequency coupling (CFC), which generally expresses interactions between frequency, phase, and/or amplitude components of a signal, has been linked to various cognitive processes. The isolation of individual source contributions that together approximate the measured interaction can provide important insights into the generative principles of the observed phenomenon. One class of promising CFC estimators is based on the cross-bispectrum, a third-order polyspectrum that evaluates nonlinear interactions between three channels at up to three frequencies. A recently developed method decomposes cross-bispectra applied to sensor-level electroencephalographic (EEG) data into components that can be interpreted as source-level cross-bispectra and a spatial topography. These components together approximate the original sensor cross-bispectrum.

Using simulations, we test this method's ability to distinguish phase-amplitude coupling (PAC) generated by a single self-interacting source (univariate PAC), from PAC generated by two distinct interacting sources (bivariate PAC). More generally, we assess whether this method successfully distinguishes within-site from between-site bispectral source interactions. For this analysis pipeline, we also introduce an algorithm to statistically test identified source interactions for significance. By comparing the decompositions of normal and antisymmetrized cross-bispectra, we find that univariate source interactions can be unambiguously distinguished from signals that exhibit bivariate source interactions. However, while bivariate source interactions can also be statistically identified, we observe that the presence of univariate interactions within the same signal cannot be conclusively excluded.

Finally, we analyze ten subjects of the LEMON dataset that exhibit strong harmonics of the alpha wave under the hypothesis that these harmonics are generated by distinct, nonlinearly interacting sources. We observe that, while the results generally vary across subjects, common topographies and sources can be identified, suggesting the presence of common generators of higher harmonics.

 

 

Additional information:

Master thesis defense

 

Organized by:

Prof. Dr. Stefan Haufe & Dr. Guido Nolte