Methodological considerations when measuring and analyzing auditory steady-state responses with multi-channel EEG

Abstract

The auditory steady-state response (ASSR) has been traditionally recorded with few electrodes and is often measured as the voltage difference between mastoid and vertex electrodes (vertical montage). As high-density EEG recording systems have gained popularity, multi-channel analysis methods have been developed to integrate the ASSR signal across channels. The phases of ASSR across electrodes can be affected by factors including the stimulus modulation rate and re-referencing strategy, which will in turn affect the estimated ASSR strength. To explore the relationship between the classical vertical-montage ASSR and whole-scalp ASSR, we applied these two techniques to the same data to estimate the strength of ASSRs evoked by tones with sinusoidal amplitude modulation rates of around 40, 100, and 200 Hz. The whole-scalp methods evaluated in our study, with either linked-mastoid or common-average reference, included ones that assume equal phase across all channels, as well as ones that allow for different phase relationships. The performance of simple averaging was compared to that of more complex methods involving principal component analysis. Overall, the root-mean-square of the phase locking values (PLVs) across all channels provided the most efficient method to detect ASSR across the range of modulation rates tested here.

Keywords: Auditory steady-state response (ASSR); Electroencephalography (EEG); Envelope following response (EFR); Multi-channel EEG; Phase locking value (PLV).

Graphical abstract

Fig. 1

Schematic representation of the AM stimuli used for measuring ASSR.

Fig. 2

The local (single-channel at Cz) spectral magnitudes (left panel) and PLVs (right panel) computed with different reference strategies, averaged across participants. The solid bars are calculated with the EEG re-referenced to CAR and the striped bars are calculated with the EEG re-referenced to LMR. The striped bars therefore represent the classical vertical montage. Error bars represent the standard error of the mean across participants. The grey bars represent the noise floor, which is the average spectral magnitude or PLV at unmodulated frequencies surrounding the tagged frequency, as defined in methods section.

Fig. 3

The group-average PLVs at different locations, re-referenced with the CAR. The bars are clustered by AM frequency. Within a frequency cluster, the dark blue bars show the PLVs measured at mastoids channels (M1+M2); the lighter blue bars show the PLVs measured at vertex channels (Cz + FCz); and the green bars show the PLV of the horizontal difference ASSR (M1-M2). Error bars represent the standard error of the mean across participants. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4

Circular average of the phase delay between the ASSR phase recorded at the mastoid (M1+M2) and vertex (Cz + FCz) channels with CAR. The horizontal dashed line represents the phase difference equaling to π, i.e., the ASSR recorded at mastoid and vertical channels were in opposite phase. The error bars represent the circular standard deviation of phase differences.

Fig. 5

The topomap of ASSR phase with LMR and CAR. The color represents the circular mean of estimated phase of all participants. With CAR, the values were mostly clustered around one of two opposite phases, as the voltage pattern of dipole under a neutral reference. With LMR, the phases were approximately uniform except for the 34-Hz ASSR. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 6

Average z-scores of the ASSR measured with the various methods. The top row represents the ASSR calculated on the EEG re-referenced to CAR, and the bottom row represents the ASSR calculated on the EEG re-referenced to LMR. The three columns of plots represent the ASSR at 34 Hz, 91 Hz, and 217 Hz, from left to right. The horizontal solid lines represent the z-score where the ASSR is the same as the mean of noise floor. The horizontal dashed lines represent the z-score with 99% confidence rejecting the null hypothesis that there was no ASSR above noise floor. Error bars represent the standard error of the mean z-score across participants.

Fig. 7

The hypothesized interpretation of the enhanced ASSR when re-referenced to mastoid channels. Panel A shows the spatial relationship between the vertical-montage active electrodes (Fz & M1) and the brainstem vertical dipole. The brainstem dipole is approximately vertical. Panel B shows the ASSR signal recorded at Fz and M1 with CAR and LMR. The top two rows of panel B are recorded ASSR with CAR, and the bottom two rows are ASSR with LMR. It was assumed that LM was identical to M1. When ASSR was referenced to CAR, the phase of recorded ASSR was opposite as the CAR was approximately neutral. When ASSR was referenced to LMR, there was no signal at mastoid but the strength at Fz was enhanced. From another point of view, Fz – M1 = (Fz – CAR) – (M1 – CAR). Figure created with BioRender.com.

Fig. S1

The local (single-channel at Cz) spectral magnitudes (left panel) and PLVs (right panel) computed with different reference strategies, averaged across participants. The main difference between Fig. S1 and Fig. 2 is that all 2000 trials for the 34-Hz condition were used in Fig. S1 and Fig. 1 only used the 1000 trials from session 1. The solid bars are calculated with the EEG re-referenced to CAR and the striped bars are calculated with the EEG re-referenced to LMR. The striped bars therefore represent the classical vertical montage. Error bars represent the standard error of the mean across participants. The grey bars represent the noise floor, which is the average spectral magnitude or PLV at unmodulated frequencies surrounding the tagged frequency, as defined in methods section.

Fig. S2

The group-average PLVs at different locations, re-referenced with the CAR. The bars are clustered by AM frequency. Fig. S2 and Fig. 2 is that all 2000 trials for the 34-Hz condition were used in Fig. S2 and Fig. 3 only used the 1000 trials from session 1. Within a frequency cluster, the dark blue bars show the PLVs measured at mastoids channels (M1+M2); the lighter blue bars show the PLVs measured at vertex channels (Cz + FCz); and the green bars show the PLV of the horizontal difference ASSR (M1-M2). Error bars represent the standard error of the mean across participants.

Fig. S3

Circular average of the phase delay between the ASSR phase recorded at the mastoid (M1+M2) and vertex (Cz + FCz) channels with CAR. Fig. S3 and Fig. 3 is that all 2000 trials for the 34-Hz condition were used in Fig. S3 and Fig. 4 only used the 1000 trials from session 1. The horizontal dashed line represents the phase difference equaling to π, i.e., the ASSR recorded at mastoid and vertical channels were in opposite phase. The error bars represent the circular standard deviation of phase differences.

Fig. S4

The topomap of ASSR phase with LMR and CAR. Fig. S4 and Fig. 4 is that all 2000 trials for the 34-Hz condition were used in Fig. S4 and Fig. 5 only used the 1000 trials from session 1. The color represents the circular mean of estimated phase of all participants. With CAR, the values were mostly clustered around one of two opposite phases, as the voltage pattern of dipole under a neutral reference. With LMR, the phases were approximately uniform except for the 34-Hz ASSR.

Fig. S5

Average z-scores of the ASSR measured with the various methods. Fig. S5 and Fig. 5 is that all 2000 trials for the 34-Hz condition were used in Fig. S5 and Fig. 6 only used the 1000 trials from session 1. The top row represents the ASSR calculated on the EEG re-referenced to CAR, and the bottom row represents the ASSR calculated on the EEG re-referenced to LMR. The three columns of plots represent the ASSR at 34 Hz, 91 Hz, and 217 Hz, from left to right. The horizontal solid lines represent the z-score where the ASSR is the same as the mean of noise floor. The horizontal dashed lines represent the z-score with 99% confidence rejecting the null hypothesis that there was no ASSR above noise floor. Error bars represent the standard error of the mean z-score across participants.

Fig. S6

The raw output from the various methods with CAR. Each row represents a method to quantify ASSR and each column represents the AM frequency (34 Hz, 91 Hz, and 217 Hz). The solid line represents the estimated strength at the AM frequency averaged across participants. The dashed line represents the mean of estimated strength at the untagged frequencies averaged across participants, and the shadow around dashed line represents the range where the measured ASSR is not significantly different from those at untagged frequencies at p = 0.05 level.

Fig. S7

The raw output from the various methods with LMR. Each row represents a method to quantify ASSR and each column represents the AM frequency (34 Hz, 91 Hz, and 217 Hz). The solid line represents the estimated strength at the AM frequency averaged across participants. The dashed line represents the mean of estimated strength at the untagged frequencies averaged across participants, and the shadow around dashed line represents the range where the measured ASSR is not significantly different from those at untagged frequencies at p = 0.05 level.