The Influence of EEG References on the Analysis of Spatio-Temporal Interrelation Patterns

Abstract

The characterization of the functional network of the brain dynamics has become a prominent tool to illuminate novel aspects of brain functioning. Due to its excellent time resolution, such research is oftentimes based on electroencephalographic recordings (EEG). However, a particular EEG-reference might cause crucial distortions of the spatiotemporal interrelation pattern and may induce spurious correlations as well as diminish genuine interrelations originally present in the dataset. Here we investigate in which manner correlation patterns are affected by a chosen EEG reference. To this end we evaluate the influence of 7 popular reference schemes on artificial recordings derived from well controlled numerical test frameworks. In this respect we are not only interested in the deformation of spatial interrelations, but we test additionally in which way the time evolution of the functional network, estimated via some bi-variate interrelation measures, gets distorted. It turns out that the median reference as well as the global average show the best performance in most situations considered in the present study. However, if a collective brain dynamics is present, where most of the signals get correlated, these schemes may also cause crucial deformations of the functional network, such that the parallel use of different reference schemes seems advisable.

Keywords: EEG reference; electroencephalography; functional network; multivariate analysis; time series analysis.

Figure 1

Average correlation matrices derived from model 7 for different simulated EEG references. The diagonal elements of all matrices are drawn in gray in order to improve visibility.

Figure 2

Average Mean Phase Coherence matrices derived for model 7 by using different simulated EEG references. The diagonal elements of all matrices are drawn in gray in order to improve visibility.

Figure 3

Average deviation D (Equation 15) from the original, viz. untransformed interrelation pattern caused by different reference schemes estimated for synthetic data derived from each of the models listed in Table 1. Left columns, indicated by a gray shadow, show the results obtained for the correlation matrix, right columns those for the mean phase coherence matrix.

Figure 4

(A) Correlation coefficient between the earlobe A1 and active electrodes T3, Cz and T4 estimated with a running window of length 3,840 data points (corresponding to 30 s) over the whole night recording of a clinically healthy, 26-year-old male subject. Sleep stages (W = awake, S1–S4 denotes sleep stages 1–4 and REM denotes Rapid Eye Movement sleep periods) are indicated by colored shadows in each panel. (B) Shows a heat map of the average correlation between the active electrodes and A1 and (C) the respective standard deviation. Both quantities are estimated over the whole night recording.

Figure 5

Correlation matrices estimated for median-referenced data previously transformed to different popular EEG-references. (A) A1A2, (B) global average, (C) F3F4, and (D) Hjorth. Data have been derived from Model 7 of Table 1, like in Figures 1, 2.

Figure 6

Power spectra averaged over all active electrode signals transformed to different reference schemes estimated for synthetic data derived from Model 1 (A), Model 2 (B), Model 7 (C), and Model 8 (D). The insets show the power spectra of some of the corresponding reference signals.

Figure 7

In the lower panel the time evolution of the average absolute value of the non-diagonal elements of the correlation matrices 〈|C|〉, estimated for an extracranial recording of a focal onset seizure, is shown. Vertical black solid lines indicate seizure onset and offset. Dashed vertical lines mark three instances of time for which the cross-correlation matrices, obtained for different reference montages, are shown in the upper part of the figure. The first two columns of six matrices correspond to the moment just after seizure onset (around second 610), the next two columns to the instant marked in the central part of the seizure (about second 635) and the matrices of the last two columns are estimated for the marked instant of time during the post-seizure period (about second 690). In all cases the diagonal elements are artificially set to gray color in order to improve visibility.

Figure 8

Distribution of non-diagonal elements of correlation matrices estimated from reference transformed data derived from model 1 of Table 1. Reference signals are extracted from real EEG-recordings of a healthy subject during sleep stage 4 (A,C) and REM sleep (B,D), for the δ−band (A,B) and the fast β−band (C,D) as described in the main text.

Figure 9

Median and 95% confidence interval of the Pearson coefficient comparing signals recorded by the same electrode before and after the application of Independent Component Analysis for the elimination of eyelid-movement. EEG recordings stem from data set three described in the method section.