Multivariate Bi-Extremal Cross-Frequency Interactions in Brain Connectivity

Oregon Convention Center 

Spectral association plays a vital role in the study of functional brain connectivity, but traditional measures focus on linear spectral associations found in the bulk of the distribution. In certain studies, such as risk analysis, the interest shifts to connectivity in the tails of the distribution, as this reveals crucial information pertaining to extreme events, e.g., seizures. This motivates us to extend the notion of spectral association into the tail of the periodogram (given a specific frequency band) to study electroencephalogram (EEG) signals of seizure-prone neonates. Existing models are limited to tail of univariate periodogram or the tail associations of filtered series. In this study, we develop a non-stationary extremal dependence model for multivariate time series, that permits different dependence behaviour during different brain phases, i.e., high and low activity. This allows us to identify key tail-frequency connectivity at specific frequency bands that could trigger an outburst of energy, and we discuss these novel scientific findings alongside a comparison of the extremal behaviour of brain signals for ictal and non-ictal patients.

extreme value theory

spectral analysis

electroencephalogram (EEG)

spectral clustering

conditional extremes

periodogram 

Section on Statistics in Imaging