Effects of electrode density and electrolyte spreading in dense array electroencephalographic recording

Abstract

Objective: High-density EEG recording offers increased spatial resolution but requires careful consideration of how the density of electrodes affects the potentials being measured. Power differences as a function of electrode density and electrolyte spreading were examined and a method for correcting these differences was tested.

Methods: Separate EEG recordings from 8 participants were made using a high-density electrode net, first with 6 of 128 electrodes active followed by recordings with all electrodes active. For a subset of 4 participants measurements were counterbalanced with recordings made in the reversed order by drying the hair after the high-density recordings and using a fresh dry electrode net of the same size for the low-density recordings. Mean power values over 6 resting eyes open/closed EEG recordings at the 6 active electrodes common to both recording conditions were compared. Evidence for possible electrolyte spreading or bridging between electrodes was acquired by computing Hjorth electrical distances. Spherical spline interpolation was tested for correcting power values at electrodes affected by electrolyte spreading for these participants and for a subset of participants from a larger previous study.

Results: For both the complete set and the counterbalanced subset, significant decreases in power at the 6 common electrodes for the high-density recordings were observed across the range of the standard EEG bands (1–44 Hz). The number of bridges or amount of electrolyte spreading towards the reference electrode as evidenced by small Hjorth electrical distances served as a predictor of this power decrease. Spherical spline interpolation increased the power values at electrodes affected by electrolyte spreading and by a significant amount for the larger number of participants in the second group.

Conclusions: Understanding signal effects caused by closely spaced electrodes, detecting electrolyte spreading and correcting its effects are important considerations for high-density EEG recordings. A combination of scalp maps of power density and plots of small Hjorth electrical distances can be used to identify electrodes affected by electrolyte spreading. Interpolation using spherical splines offers a method for correcting the potentials measured at these electrodes.

Introduction

Adequate spatial sampling of scalp electrical potentials is necessary for optimal resolution of the sources within the brain (Tucker, 1993, Srinivasan et al., 1998, Lantz et al., 2003). In recent years new solutions for high-density EEG recording have been implemented but have been slow to come into general use due to the difficulty of applying large numbers of electrodes particularly when scalp abrasion is required, the increased possibility for electrolyte spreading or bridging of the shorter distances between electrodes, and the greatly increased data processing requirements often involving artifact scoring by humans. One solution, the128-channel geodetic sensor net, (GSN128; Electrical Geodesics Inc., Eugene, OR) offers relatively quick and easy application without the need for scalp abrasion (Ferree and Tucker, 1999; Ferree et al., 2001). However, the risk of electrolyte spreading and increased processing requirements remain.

A GSN128 electrode consists of an Ag/AgCl pellet in contact with a saline (KCl) soaked sponge encased in a flared plastic pedestal designed to keep the saline electrolyte in contact with the scalp and shield it from being wicked away by strands of hair. These electrode pedestals are arranged in a geodetic structure (see Tucker, 1993 for details) using slightly elastic monofilament line so that the tension tends to push them vertically down onto the scalp. Opportunities for electrolyte spreading occur during initial positioning of electrodes on the scalp and subsequent wiggling of individual electrode pedestals in order to obtain good scalp contact with sufficiently low electrode impedances. Electrolyte drying or additional spreading due to wicking by hair could potentially change electrolyte spreading effects during EEG recordings.

Comparisons of low- and high-density EEG recordings have been primarily concerned with the spatial resolution of electrical sources in the brain (Spitzer et al., 1989, Tucker, 1993, Srinivasan et al., 1998) while the effects that large numbers of highly conductive electrodes and associated electrolyte have on potentials measured on the scalp surface boundary have been largely ignored. However, one direct comparison of ERPs recorded at high density using the GSN128 and a conventional 30-channel system (Electro-Cap International Inc.) was made by Kayser et al. (2000). Their results for ERPs averaged over 17 subjects with both 30- and 128-channel data showed smaller ERP amplitudes and lower signal-to-noise ratios for the 128-channel data which could be related to electrode density and electrolyte spreading. Since our EEG work is largely concerned with estimating parameters of brain electrical asymmetry (Davidson, 1988, Tomarken et al., 1992) we set out to perform a similar comparison of low- and high-density EEG recordings in order to assess the effects on electrode power estimates due to increased electrode density and possible effects (especially asymmetrical ones) of electrolyte spreading. Computation of asymmetrical brain electrical measures involves power estimates from a participant during a single recording session so that possible asymmetrical effects of electrolyte spreading are not removed by averaging. Initial analyses of frontal asymmetry scores from a study which used EEG data collected with the high-density GSN128 indicated that overall test-retest stability (alpha 8–13 Hz F3/4, average mastoid reference, r=0.22, n=150, P=0.007) was significantly lower than that of a comparable low-density study (alpha 8–13 Hz F3/4, average ears reference, r=0.66, n=85, P=0.001; Tomarken et al., 1992) (Fisher's test for independent correlations: Z=−4.13, P<0.0001; Fisher, 1921). In addition, male participant's test-retest stability (alpha 8–13 Hz F3/4, average mastoid reference r=0.56, n=18, P=0.017) was higher than for females (alpha 8–13 Hz F3/4, average mastoid reference r=0.22, n=132, P=0.013) (Fisher's test for independent correlations: Z=1.50, P<0.14; Fisher, 1921), suggesting that hair length might be a factor.

A simple experiment was run to test directly for power differences between low- and high-density EEG recordings. The GSN128 was first applied with only 6 electrodes soaked in electrolyte and active, resting recordings were made, and the net was removed and reapplied with all electrodes soaked in electrolyte and active. A second set of resting recordings was made, and power densities were compared between conditions for the original 6 active electrodes. For a subset of the participants recordings were made in reversed order (128 followed by 6 active electrodes) to verify that results were not affected by the order in which the recordings were made. For the high electrode density portion of the reversed order recordings the recently released on-line bridge detection software (Electrical Geodesics Inc., 2003) was tested for comparison with the off-line method described in this paper.

In order to detect electrolyte spreading or bridging Hjorth electrical distances were computed off-line using the method suggested in Tenke and Kayser (2001). The Hjorth electrical distance between two electrodes is simply the temporal variance of their difference potential, which would be decreased by electrolyte spreading or bridging between them. Thus anomalously small Hjorth electrical distances could indicate electrolyte spreading or bridging and a series of them between an electrode and the reference electrode could be the source of a power decrease at that electrode. Such an effect would seriously compromise an asymmetry score based on homologous electrodes where the power at one site was affected more than its homologous site by electrolyte spreading for the duration of all resting EEG recordings.

A practical solution to power changes associated with electrolyte spreading was examined and a second experiment was performed to extend the method to a different data set with more participants. A subset of 17 participants from a large study of EEG asymmetry was selected based on a moderate number of small Hjorth electrical distances in the left frontal area where power decrements could potentially affect frontal asymmetry scores. It was predicted that interpolation of voltages at the affected electrodes using values from the remaining unaffected electrodes would significantly change their powers. Given the importance of variations in alpha power for the assessment of regional activations associated with cognitive (e.g. Davidson et al., 1990) and affective (see e.g. Davidson et al., 2000a, for review) processes, log-transformed power density in the alpha band was examined.