Frequency domain estimation of covariate effects in multichannel brain evoked potential data

Abstract

An evoked potential is the recorded brain electrical response to a stimulus such as an auditory click. In most evoked potential experiments, the response is studied as a function of covariates such as stimulus characteristics and drug states. We propose a frequency domain model of multichannel evoked potential data that includes parameters representing the effects of covariates on the amplitude and latency (time from stimulus presentation) of the response. The variability of the response among scalp electrodes is modeled by the activity of one or more equivalent electrical dipoles. The frequency domain representation allows considerable data reduction, facilitates modeling of the noise, and leads to a simple approximate expression for the latency effects. We describe maximum likelihood estimation of the model parameters and construction of approximate confidence intervals for the covariate effects. We report the results of a simulation study in which we evaluated the bias of the estimators and the coverage rate of the confidence intervals. We also report the results of an application to auditory evoked potentials recorded from five subjects.