Single-trial log transformation is optimal in frequency analysis of resting EEG alpha

Abstract

The appropriate definition and scaling of the magnitude of EEG oscillations is an underdeveloped area. The aim of this study was to optimize the analysis of resting EEG alpha magnitude, focusing on alpha peak frequency and non-linear transformation of alpha power. A family of non-linear transforms, Box-Cox transforms, were applied to find the transform that (a) maximized a non-disputed effect: the increase in alpha magnitude when the eyes are closed (Berger effect), and (b) made the distribution of alpha magnitude closest to normal across epochs-within each participant, or across participants. The transformations were done either at the single epoch level, or at the epoch-average level. Alpha peak frequency showed large individual differences, yet good correspondence between various ways to estimate it in two minutes of eyes-closed and two minutes of eyes-open resting EEG data. Both alpha magnitude and the Berger effect were larger for individual alpha than for a generic (8-12 Hz) alpha band. The log transform on single epochs (a) maximized the t-value of the contrast between the eyes-open and eyes-closed condition when tested within each participant, and (b) rendered near-normally distributed alpha power across epochs and participants, thereby making further transformation of epoch-averages superfluous. The results suggest that the log-normal distribution is a fundamental property of variations in alpha power across time in the order of seconds. Moreover, effects on alpha power appear to be multiplicative rather than additive. These findings support the use of the log transform on single epochs to achieve appropriate scaling of alpha magnitude.

This article is protected by copyright. All rights reserved.

http://ift.tt/2nu6t8W