A Bayesian Model for EEG Frequency-Domain Functional Connectivity

• Activity Number: 15
• Type: Topic Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Wesley Thompson*+
• Companies: University of California, San Diego
• Keywords: EEG ; Functional Connectivity ; Frequency Domain ; Bayesian ; MCMC

Abstract

In this paper we develop a Bayesian model for frequency-domain analysis of multichannel EEG time series. We use a complex normal approximation (Whittle likelihood) of the distribution of the Fourier-transformed data. The concentration matrix of this likelihood is the inverse cross-spectrum. Positive-definiteness constraints are removed by performing a modified Cholesky decomposition. The resulting parameters are smoothed over frequencies by modeling them as a linear combination of spline basis functions. This formulation allows for fixed effects from covariates and random effects due to repetition within or across subjects. Thus functional connectivity maps can be summarized on the mean or group levels or presented at the subject-level. Markov Chain Monte Carlo methods are used to estimate parameters and perform inference. This method is validated on real and simulated data.