Issues relating to the subtraction phase in EOG artefact correction of the EEG

International Journal of Psychophysiology 44 (2002) 187–195

Issues relating to the subtraction phase in EOG artefact correction of the EEG Rodney J. Crofta,b,c,*, Robert J. Barryb,c a

Department of Cognitive Neuroscience and Behaviour, Imperial College Medical School, St. Dunstans Road, London W68RF, UK b Brain and Behaviour Research Institute, University of Wollongong, Northfields Ave., Wollongong 2522, Australia c Department of Psychology, University of Wollongong, Northfields Ave., Wollongong 2522, Australia Received 5 December 2000; received in revised form 7 November 2001; accepted 14 November 2001

Abstract An important method for removing the effect of ocular artefact from the EEG is ‘EOG correction’. This method estimates the proportion of ocular artefact that is in the EEG, and removes it by subtraction. To date, EOG correction research has focused on problems relating to the estimation of the correction coefficients. Using both mathematical rationale and empirical data, this paper addresses issues relating to the subtraction phase, such as the magnitude of error that can be expected due to EOG correction. Using ERP methodology, principal component and regression analyses, it is shown that the N1P2 complex propagates forward to the horizontal and radial (but not vertical) electrooculograms (EOG), and it is shown mathematically that this will result in EOG-correction error. Assuming an accurate estimate of ocular contamination of the EEG, maximal subtraction-phase error of the N1P2 complex was found to be a prefrontal attenuation of 15–22%, decreasing to central and occipital enhancements of 3–4% and 13– 14%, respectively. The magnitude of this subtraction-phase error is compared to between-subject ERP variability and to error associated with EOG rejection (omitting data contaminated by ocular artefact). It is argued that such EOG correction error is small relative to both artefact rejection procedures and to normal variability found in ERP studies, and that it is less pernicious than artefact rejection procedures. 䊚 2002 Elsevier Science B.V. All rights reserved. Keywords: EEG; EOG; Correction; Ocular artefact; Subtraction phase; Forward propagation

1. Introduction Voltage changes generated by eye movements contaminate the EEG substantially and need to be accounted for. One method of doing this is EOG correction, where the amount of ocular artefact *Corresponding author. Brain and Behaviour Research Institute, University of Wollongong, Wollongong 2522, Australia. Tel.: q61-2-4221-3732; fax: q61-2-4221-4163. E-mail address: [email protected] (R.J. Croft).

contaminating the EEG is estimated and subtracted from the EEG. This has substantial advantages over the commonly used method of controlling eye movement through fixation instructions and rejecting data contaminated by eye movements (for a review see Croft and Barry, 2000a). However, there remains a difficulty with EOG correction that has not been dealt with adequately in the literature.

0167-8760/02/$ - see front matter 䊚 2002 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 7 - 8 7 6 0 Ž 0 1 . 0 0 2 0 1 - X

188

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

In order to understand this difficulty, it is useful to consider ‘EOG correction’ as consisting of two distinct parts. The first part is concerned with the calculation of the fraction (B) of ocular voltage that reaches the electrode site to be corrected. We shall refer to this first part as the ‘calculation phase’. The second part of the procedure is to remove this fraction of the EOG from the EEG; we shall refer to this as the ‘subtraction phase’. Much of the EOG correction literature is concerned with improving the accuracy of the calculation phase (e.g. Whitton et al., 1978; Verleger et al., 1982; Gratton et al., 1983; Semlitsch et al., 1986; Croft and Barry, 2000b), but little quantitative research has been performed concerning EOG subtraction-phase error. A number of researchers have suggested that EOG subtraction-phase error may occur as a result of neural potentials contaminating the EOG (e.g. Verleger et al., 1982; Gasser et al., 1985; Berg and Scherg, 1994; Jung et al., 2000a). However, although the presence of neural potentials in the EOG has been reported (Kennedy et al., 1948; Gasser et al., 1985), to the authors’ knowledge this ‘forward propagation’ has only been demonstrated once (Iacono and Lykken, 1981). Detailed information regarding forward propagation or its affect on EOG-correction is therefore not available. Methods have been proposed to account for subtraction-phase error (e.g. Berg and Scherg, 1994; Vigario, 1997; Jung et al., 2000a), but until Iacono and Lykken’s (1981) demonstration of forward propagation is replicated and the degree of resultant subtraction-phase error is shown to be important, it is difficult to determine whether such options are necessary. 1.1. How does the error occur in the subtraction phase? Neural potentials are thought to propagate forward to the EOG (Iacono and Lykken, 1981), and so the measured EOG is thought to reflect (at least) two components, eye movement and neural potentials. This is indicated in Eq. (1), where MEOG is the voltage measured about the eyes, TEOG is the true portion of this voltage due to eye movements, and MEEGFP is the portion of

EEG measured in the EOG due to forward propagation of neural potentials. As demonstrated in Eq. (2), the goal of EOG correction is to subtract a portion (B) of TEOG from the measured EEG (MEEG) at a particular scalp site, yielding an estimate of EEG uncontaminated by ocular artefact (TEEG). But as we only have access to MEOG and not TEOG wEq. (1)x, Eq. (2) is not possible and we must use Eq. (3) instead (where estTEEG is the EOG-corrected or estimated true EEG). Thus when we EOG-correct, we are really doing it according to Eq. (4), and we not only remove a portion of ocular artefact (B*TEOG), we also remove a portion of EEG from the MEEG (B*MEEGFP). This is the subtraction-phase error. MEOGsTEOGqMEEGFP

(1)

TEEGsMEEGyBUTEOG

(2)

estTEEGsMEEGyBUMEOG

(3)

estTEEGsMEEGyBUTEOG yBUMEEGFP

(4)

1.2. What is the magnitude of this error? If we consider Eqs. (2) and (4), we see that the subtraction-phase error (the difference between true and estimated EEG) is equal to B*MEEGFP. As B may be readily obtained using standard linear regression, in order to calculate subtraction-phase error all that remains is to determine MEEGFP. If we define this error as a proportion of the neural potential estimated at a site j, then the error (E) can be calculated using Eq. (5), where B is the EOG to EEG propagation rate, BFP is the proportion of a neural potential present at site j that is also present in the EOG, and the summation takes place over k EOG channels. The problem is that the calculation of BFP is not trivial. The difficulty is that in order to Ejs8ŽBjkUBFPjk.

(5)

determine the proportion of the neural potential that reaches an EOG channel, the EEG signal at the EOG would need to be much larger than the background EOG signal (Croft and Barry, 1998),

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

and because ocular voltages are very large, this is rarely the case. Some studies have carefully separated ocular from neural potentials in channels near the eyes, but it is difficult to determine how the results of these studies apply to the current discussion as results were not reported in terms of EOG channels. For instance, Berg and Scherg (1994) and Jung et al. (2000b) reported waveforms in monopolar channels above and below the eyes, but as vertical EOG (VEOG) is typically recorded as the difference between these channels, such potentials may cancel out and thus not result in EOG subtraction-phase error. This possibility is likely because the line of least resistance from brain to eye electrodes may run through the orbit, which is roughly equidistant from electrodes above and below the eye. Horizontal and radial EOG channels may be more prone to forward propagation because they are computed differently, suggesting that subtraction-phase error may differ depending on the EOG channels employed. This study was thus designed to determine the magnitude of EEG that reaches the EOG, and consequently the magnitude of error, if any, involved in the EOG subtraction phase. To determine the magnitude of BFP, this study utilised a number of techniques to increase the relative EEG magnitude in the EOG. It averaged EEG signals using standard ERP techniques and further enhanced the EEG signal by averaging subjects’ ERP waveforms together. Principal component analysis reduced these grand average ERPs into orthogonal waveforms which allowed ERP and ocular components to be dissociated within both the EEG and EOG channels, and within these refined signals, regression analyses then determined the proportion of ERP components present in the EEG that were also present in the EOG, or BFP. This paper assumes that the calculation phase produces accurate correction coefficients, and hence considers only error related to the subtraction phase. 2. Materials and methods 2.1. Subjects Eight male and eight female subjects participated in the study and were paid for their time. Each

189

gave written informed consent and was free to withdraw from the study at any time without penalty. The study was approved by the medical school’s human ethics committee. 2.2. Data acquisition EEG was recorded from 28 scalp sites (FP1, FP2, F7, F8, F3, Fz, F4, FTC1, FTC2, C3, Cz, C4, CP1, CP2, TCP1, TCP2, T3, T4, P3, Pz, P4, T5, T6, PO1, PO2, O1, Oz, O2) and referred to the left mastoid using tin electrodes. Subjects were grounded midway between Fz and FPz. EOG was recorded above (E1) and below (E3) the left eye and from the outer canthi of the left (E5) and right (E6) eyes. Providing measures of vertical, horizontal and radial eye movement, VEOG was computed as E1yE3, HEOG as E5yE6, and REOG as (E1qE3)y2, respectively. REOG, although not typically employed in EOG correction techniques, is thought by some to be necessary to account for eye movements in the plane perpendicular to both the vertical and horizontal planes (Elbert et al., 1985), and by others to account for the eyelid component of blinks (Croft, 2000). A gain of 2500 was used for each channel with a bandpass of 0.05–100 Hz. Impedances were kept below 5 kV and data were digitised at 500 Hz. 2.3. Stimuli Stimuli consisted of 100 binaural 1000-Hz tones (30-ms duration plus 10-ms rise and fall time), at each of 60, 70, 80, 90 and 100 dB SPL intensities. These were presented through headphones in a pseudo-random fashion with a variable ISI (1800– 2200 ms). 2.4. Procedure Upon arrival at the laboratory, subjects completed consent forms and had EEG recording apparatus attached. They were seated in an armchair in an electrically shielded sound attenuated booth where their hearing thresholds were checked. Subjects then completed a series of personality and drug use questionnaires while the auditory stimulus

190

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

Fig. 1. Factors 4 and 5, corresponding to HEOG and VEOG deflections, respectively, are shown with ‘F1 to F3’, the summation of factors 1 to 3. ‘F1 to F3’ shows the characteristic N1P2 complex beginning at 100 ms.

sequence was presented through headphones (duration 17 min). 2.5. Data analysis 2.5.1. Backward propagation The proportion of ocular potential present in the EEG was estimated using the RAAA procedure (Croft and Barry, 2000b). This procedure involves a two-stage iterative process whereby VEOG and HEOG B values are calculated from averaged horizontal and vertical saccades using simultaneous multiple regression (Croft and Barry, 2000c), and REOG B values are calculated from averaged blink data (following the removal of VEOG and HEOG potentials from REOG and EEG channels). 2.5.2. Forward propagation To estimate forward propagation, epochs were defined as 0–300 ms post stimulus. No artefact rejection or correction procedures were employed on these data. One ERP per site was created for each subject by averaging 80-, 90- and 100-dB tone epochs together. For each site, the 16 resultant ERPs were averaged together to form grand average ERPs. The grand average ERPs for each of the 28 scalp sites plus three EOG channels were entered

into a principal component analysis, resulting in five factors (31 site=150 time point data matrix; Varimax rotation; covariance matrix; 99.8% variance explained; rescaled eigenvalues)0.6). Factors 1, 2 and 3 did not correlate with VEOG or HEOG traces, whereas factor 4 corresponded to the HEOG trace (rSpearmans0.87) and factor 5 to the VEOG trace (rSpearmans0.81). To minimise the effect of non-neural potentials on the EOG and EEG, factors 4 and 5 were removed from each EOG and EEG channel’s ERP, giving ‘F resid’ (derived as the unstandardised residual after predicting each channel from factors 4 and 5 using simultaneous multiple regression; Fig. 1). Simple linear regression analyses separately estimated the proportion of F resid from each scalp site present in each of the VEOG, HEOG and REOG channels. 3. Results The averaging procedure succeeded in minimising the EOG, with a VEOG range over the grand average epoch of only 6.55 mV. The grand average ERP waveform at Cz, along with F resid at Cz, VEOG, HEOG and REOG can be seen in Fig. 2. Estimates of the proportion of F resid that propagated from each site to each of the EOG channels are given in Table 1. These forward propagation values were entered into Eq. (5) to give estimates

Fig. 2. The grand average ERP waveform at Cz, along with F resid at Cz, VEOG, HEOG and REOG (waveforms remaining after the removal of factors 4 wHEOGx and 5 wVEOGx) are shown.

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

191

Table 1 Estimates of backward (EOG to EEG) and forward (EEG to EOG) propagation are given for each EOG and scalp channel Site

FP1 FP2 F7 F8 F3 Fz F4 FTC1 FTC2 C3 Cz C4 CP1 CP2 TCP1 TCP2 T3 T4 P3 Pz P4 T5 T6 PO1 PO2 O1 Oz O2

Backward propagation

Forward propagation

Error (% of estTEEG)

VE

HE

RE

VE

HE

RE

VE

HE

RE

E

0.23 0.24 0.11 0.10 0.13 0.14 0.13 0.09 0.09 0.09 0.11 0.10 0.10 0.10 0.08 0.07 0.06 0.06 0.08 0.09 0.08 0.06 0.06 0.08 0.08 0.06 0.06 0.06

0.02 0.08 y0.09 0.18 0.01 0.05 0.09 y0.01 0.12 0.03 0.06 0.08 0.05 0.06 0.02 0.09 y0.01 0.12 0.04 0.06 0.07 0.03 0.08 0.05 0.06 0.05 0.05 0.06

0.33 0.24 0.11 0.07 0.00 y0.04 y0.02 y0.03 y0.04 y0.12 y0.15 y0.12 y0.15 y0.16 y0.12 y0.11 y0.07 y0.06 y0.16 y0.18 y0.16 y0.12 y0.11 y0.19 y0.20 y0.16 y0.17 y0.16

0.02 0.01 0.01 0.00 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.02 0.01 0.01 0.00 y0.01 y0.02 0.03 0.03 0.03 0.01 0.00 0.08 0.08 0.30 0.31 0.30

y0.21 y0.18 y0.20 y0.18 y0.12 y0.10 y0.11 y0.13 y0.12 y0.10 y0.08 y0.10 y0.11 y0.11 y0.15 y0.16 y0.23 y0.17 y0.19 y0.16 y0.20 y0.48 y0.43 y0.29 y0.28 y0.52 y0.47 y0.47

0.68 0.65 0.49 0.43 0.29 0.26 0.27 0.29 0.26 0.24 0.19 0.23 0.25 0.26 0.32 0.31 0.42 0.31 0.40 0.36 0.40 0.76 0.55 0.59 0.57 0.83 0.76 0.72

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 2 2 2

0 y1 2 y3 0 y1 y1 0 y1 0 0 y1 y1 y1 0 y1 0 y2 y1 y1 y1 y1 y3 y1 y2 y3 y2 y3

22 16 5 3 0 y1 y1 y1 y1 y3 y3 y3 y4 y4 y4 y3 y3 y2 y6 y6 y6 y9 y6 y11 y11 y13 y13 y12

22 15 7 0 0 y1 y1 y1 y2 y3 y3 y4 y4 y5 y4 y5 y3 y4 y7 y7 y8 y11 y9 y12 y12 y14 y13 y13

Error is also shown as percentage errors of the N1P2 deflection at the scalp sites in the left-hand column (i.e. forward=backward propagation) and ‘E’ is the summation of all these EOG errors.

of the error caused by EOG correction. The degree of error, as a percentage of neural potential, ranged from an attenuation of 15–22% prefrontally, to an enhancement of 3–4% centrally and 12–14% occipitally (Table 1). 4. Discussion The results demonstrate that the N1P2 complex propagates forward to the EOG, with the degree of propagation different for the three eye channels. Essentially no N1P2 potential was recorded in VEOG, presumably because equal amounts of the complex were recorded above and below the eye and were removed through subtraction (i.e.

VEOGsE1yE3). Thus the present study has failed to replicate Iacono and Lykken (1981) in that neural potentials were not observed in VEOG. Slightly more N1P2 was recorded at HEOG, and because REOG is calculated as the average of potentials above and below the eyes, substantial N1P2 propagated to REOG. It follows that there was very little subtraction-phase error involved with vertical and horizontal EOG correction (2% and -3% of the EEG, respectively), but that correction of the radial channel incurred subtraction-phase error ranging from 0 to 22% of the EEG (Table 1). That subtraction-phase error was primarily restricted to the radial channel means that EOG-

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

192 Table 2 N1 amplitudes Site

N1P2 magnitude (mV) TEEG N1

MEEG S.D. N1

FP1 y4.5 8.1 FP2 y3.6 7.9 F7 y4.5 4.2 F8 y5.1 3.8 F3 y8.1 4.6 Fz y9.8 3.7 F4 y8.7 4.4 FTC1 y7.2 3.3 FTC2 y7.7 2.9 C3 y9.8 1.9 Cz y10.8 2.4 C4 y9.1 2.2 CP1 y8.7 1.7 CP2 y8.6 2.0 TCP1 y5.3 1.6 TCP2 y6.7 1.8 T3 y2.3 1.7 T4 y3.4 1.5 P3 y4.7 1.5 Pz y5.5 1.7 P4 y4.9 1.9 T5 y0.7 1.3 T6 y0.7 2.3 PO1 y3.4 1.7 PO2 y3.8 1.8 O1 y1.9 1.9 Oz y1.4 2.0 O2 y1.7 2.2

estTEEG d

N1

d

dyS.D.

9.3 y13.8 y3.5 y1.0 y0.1 10.8 y14.4 y3.1 y0.5 y0.1 2.1 y6.6 y4.2 y0.3 y0.1 0.9 y6.0 y5.1 0.0 0.0 y0.3 y7.8 y8.1 0.0 0.0 y1.4 y8.4 y10.0 0.1 0.0 y0.9 y7.8 y8.8 0.1 0.0 y1.8 y5.4 y7.2 0.0 0.0 y2.3 y5.4 y7.9 0.2 0.1 y4.4 y5.4 y10.1 0.3 0.2 y4.2 y6.6 y11.1 0.3 0.1 y3.1 y6.0 y9.5 0.3 0.1 y2.7 y6.0 y9.1 0.4 0.2 y2.6 y6.0 y9.0 0.4 0.2 y0.5 y4.8 y5.5 0.2 0.1 y2.5 y4.2 y7.0 0.3 0.2 1.3 y3.6 y2.4 0.1 0.0 0.2 y3.6 y3.5 0.1 0.1 0.1 y4.8 y5.0 0.3 0.2 y0.1 y5.4 y5.8 0.4 0.2 y0.1 y4.8 y5.3 0.4 0.2 2.9 y3.6 y0.7 0.1 0.1 2.9 y3.6 y0.7 0.1 0.0 1.4 y4.8 y3.8 0.4 0.2 1.0 y4.8 y4.3 0.5 0.3 1.7 y3.6 y2.2 0.3 0.1 2.2 y3.6 y1.6 0.2 0.1 1.9 y3.6 y1.9 0.2 0.1

Fig. 3. The true N1P2 complex is shown at FP1 (TEEG), along with the same wave contaminated by a 60-mV vertical eye movement (MEEG) and the EOG corrected waveform (estTEEG).

error, that resulting from the EOG-correction of a 60 mV VEOG deflection. In the above data, baseline-to-peak N1 amplitudes at FP1, Cz and Oz were y4.5, y10.8 and y1.4 mV, respectively. Using the backward propagation values from Table 1 it can be seen that a 60-mV vertical eye movement would alter these values to q9.3, y4.2 and q2.2 mV, respectively. After EOG correcting the data, these peaks would be estimated as y3.5, y 11.1 and y1.6 mV, respectively, deviations from

‘TEEG’ denotes the magnitude of the true N1 deflection, ‘MEEG’ the N1 magnitude after contamination of a 60 mV vertical eye movement and ‘estTEEG’ the MEEG magnitude after EOG correction using VEOG, HEOG and REOG. ‘S.D.’ denotes the standard deviation of the N1 deflection over the 16 subjects, ‘d’ denotes error, or the difference between TEEG and each of MEEG and estTEEG, and ‘dyS.D.’ represents error as a function of ‘S.D.’

correction procedures that do not use the radial channel would have essentially avoided subtraction-phase error in the present data set. However, as the relative accuracy of procedures that do and do not employ a radial channel has not been determined, we shall consider the worst-case scenario, that of the error produced by employing all three EOG channels. If we take the N1 component of the above N1P2 complex as an example, Table 2 provides an example of this subtraction-phase

Fig. 4. The true N1P2 complex is shown at Oz (TEEG), along with the same wave contaminated by a 60-mV vertical eye movement (MEEG) and the EOG corrected waveform (estTEEG). The corrected waveform is barely distinguishable from the uncontaminated waveform.

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

193

Fig. 5. Scalp distributions of residual data (i.e. after removing eye movement potentials) are shown at the N1 (left; 100 ms) and P2 (right; 180 ms) latency ranges. Both exhibit maxima over fronto-central sites characteristic of the N1P2 complex.

true EEG of q1.0, q0.3 and y0.2 mV, respectively. Thus, even the 22% prefrontal error was small — it merely appeared large because it was described relative to the N1 component where it was small, at prefrontal sites. As examples of maximal error (relative to true EEG), the above true, contaminated and EOG-corrected FP1 and Oz waveforms are given in Figs. 3 and 4. The above errors were determined by both forward and backward propagation rates wEq. (5)x, and although backward propagation at a site may be constant within-subject, forward propagation and thus error will depend on the neural source’s location and orientation. That is, where a greater (lesser) portion of a source propagates to the EOG than does the N1 deflection, error will be larger (smaller) than that described in Tables 1 and 2. As a guide, we may expect sources more posterior than that of the N1P2 deflection (primary auditory cortex) to be affected less by subtraction-phase error, and those more anterior to be affected more. However, due to such complexities as the orientation of the source, these patterns need to be determined empirically, and correspondingly it must be remembered that the present results are based on the analysis of one ERP complex only. The present results are discrepant with those of Iacono and Lykken (1981). Why this may be the case is not clear. One possibility is that eye electrodes in that study were not placed equidistant

from the centre of the orbit. However, as error in that case would be proportional to the deviation from ‘equidistance’, this is unlikely to explain the visually-discernible EEG in their VEOG traces. A more likely explanation is that the degree to which neural potentials propagate through the centre of the orbits is proportional to the distance from them. Thus, the prefrontal ‘kappa’ of Iacono and Lykken may have differentially affected E1 and E3, whereas the more posterior ERP sources of Verleger et al. (1982) and the present study may have passed through the orbits more centrally. Having established that subtraction-phase error does occur, we need to consider to what extent this error should concern the researcher. We shall address this by comparing this error to that resulting from a common alternative to EOG-correction, and to data variability in general. First, let us consider the most common alternative to EOG correction-instructing subjects to fixate and refrain from blinking, and then eliminating epochs contaminated by the remaining ocular artefact. If we employ the backward propagation values of Table 1, it can be seen that in order to restrict error to the same magnitude as the correction-phase error, the rejection criterion would have to be set impractically low. For example, continuing with the above examples of FP1 and Oz, in the present data set it would take uncorrected VEOG deflections of only 2.6 and 3.3 mV, respectively to cause

194

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

the same magnitude of error as the subtraction phase. As discussed above, subtraction-phase error will differ for different neural sources, but even if it were magnified 10-fold to account for such variations, a VEOG rejection criterion would still need to be restricted to an impractical level (33 mV) so as not to exceed subtraction-phase error. Furthermore, valid implementation of the rejection method has been shown to be problematic (Verleger, 1993), and the instructions to ‘fixate and not blink’ impose secondary tasks that have been shown to reduce the N1, P3 and CNV waveforms (Weerts and Lang, 1973; Verleger, 1991; Ochoa and Polich, 2000). So, as well as making interpretation of cognitive function problematic, this introduces bias in the form of subjects’ differential abilities to cope with the secondary task, and the bias is not understood well enough to allow for compensation. Another means of assessing the importance of the above subtraction-phase error is to compare it to the magnitude of error typically encountered in EEG studies. As an example, the variability of the N1 deflection across the 16 subjects is given in Table 2. The mean absolute error was 0.11 N1 standard deviations and the largest error was at PO2 where subtraction-phase error was 0.26 standard deviations. Error was smallest frontally, where REOG has little effect on the EEG. So, in the present data set the subtraction-phase error was small relative to EOG rejection and fixationyblinkabstinence instruction, and relative also to between-subject variability in N1 amplitude. It is important to note that that error introduced in the subtraction-phase is less pernicious than that introduced by the combination of instructing subjects not to move their eyes and EOG rejection. This is because the error related to failing to reject data contaminated by ocular artefact is dependent on the eye movements which do occur, and that related to the fixation and blink-abstinence instruction is related to cognition. Conversely, subtraction-phase error is dependent only on the physical parameters of the head and the location of the neural source — it is thus independent of eye movement and dependent on cognition only in so far as the location of the source varies. This means that correction-phase error will be consistent with-

in-subject (if there was a 10% reduction of the N1 deflection at one site in a subject, then this 10% reduction would occur in all conditions unless the dipole moved), and will only differ between subjects in so far as the subjects’ source locations and head parameters differ. The advantage of removing the bias due to eye movement can be seen in Figs. 3 and 4, where the EOG-corrected waveforms retained their shape, whereas the uncorrected EEG did not. It should be noted that there is no guarantee that the PCA used in this study has validly separated the neural from ocular voltages, and thus the conclusions drawn from the present study should best be viewed as preliminary. It may be argued, for instance, that the EOG factors (factors 4 and 5) may have contained a substantial portion of the N1P2 complex, and so the calculation of EEG residual waveforms (the N1P2 waveform minus the effect of factors 4 and 5) may have removed this portion of the N1P2 complex from both the EEG and EOG waveforms, making it merely a mathematical artefact that very little N1P2 was measured in the EOG waveforms. However, any N1P2 present in factors 4 and 5 also would have been removed from the F resid scalp values and thus would not have affected the forward propagation values — the forward propagation values of Table 1 will represent the contamination of the F resid waveforms on the EOG. Furthermore, the F resid waveforms exhibit a topography characteristic of N1P2 (see Fig. 5) and correlate strongly with the waveforms prior to the removal of factors 4 and 5 (e.g. Cz and Cz Resid; rSpearmans0.91), suggesting that factors 4 and 5 did not contain a substantial portion of the N1P2 complex. It is concluded that the N1P2 complex does propagate forward to the EOG, and correspondingly that, in EOG correction, the subtraction of a portion of the measured EOG introduces error into the corrected EEG. This is dependent on the type of EOG channels employed, with REOG the principle source of error. This error is independent of both cognition and eye movement, and was found to be small relative to the main alternative to EOG correction, rejecting data contaminated by ocular artefact, and also relative to between-subject variation in the N1 deflection. It is concluded that

R.J. Croft, R.J. Barry / International Journal of Psychophysiology 44 (2002) 187–195

subtraction-phase error in EOG correction of the N1P2 component (at least) is not large enough to be of concern to the researcher, and that the commonly held belief that subtraction-phase error may make EOG correction overly problematic has not been supported. References Berg, P., Scherg, M., 1994. A multiple source approach to the correction of eye artifacts. Electroencephalogr. Clin. Neurophysiol. 90, 229–241. Croft, R.J., 2000. The differential correction of eyelid-movement and globe rotation artefact from the EEG (Reply to Verleger). J. Psychophysiol. 14, 207–209. Croft, R.J., Barry, R.J., 1998. EOG correction: A new perspective. Electroencephalogr. Clin. Neurophysiol. 107, 387–394. Croft, R.J., Barry, R.J., 2000a. Removal of ocular artifact from the EEG: A review. Neurophysiol. Clin.yClin. Neurophysiol. 30, 5–19. Croft, R.J., Barry, R.J., 2000b. EOG correction of blinks with saccade coefficients: A test and revision of the alignedartifact average solution. Clin. Neurophysiol. 3, 444–455. Croft, R.J., Barry, R.J., 2000c. EOG correction: which regression should we use? Psychophysiology 37, 123–125. Elbert, T., Lutzenberger, W., Rockstroh, B., Birbaumer, N., 1985. Removal of ocular artifacts from the EEG: A biophysical approach to the EOG. Electroencephalogr. Clin. Neurophysiol. 60, 455–463. ¨ Gasser, T., Sroka, L., Mocks, J., 1985. The transfer of EOG activity into the EEG for eyes open and closed. Electroencephalogr. Clin. Neurophysiol. 61, 181–193. Gratton, G., Coles, M.G.H., Donchin, E., 1983. A new method for the off-line removal of ocular artifact. Electroencephalogr. Clin. Neurophysiol. 55, 468–484.

195

Iacono, W.G., Lykken, D.T., 1981. Two-year retest stability of eye-tracking performance and a comparison of electrooculographic and infrared recording techniques: Evidence of EEG in the electro-oculogram. Psychophysiology 18, 49–55. Jung, T., Makeig, S., Humphries, C., et al., 2000a. Removing electroencephalographic artifacts by blind source separation. Psychophysiology 37, 163–187. Jung, T., Makeig, S., Westerfield, M., Townsend, J., Courchesne, E., Sejnowski, T.J., 2000b. Removal of eye activity artifacts from visual event-related potentials in normal and clinical subjects. Clin. Neurophysiol. 111, 1745–1758. Kennedy, J.L., Gottsdanker, R.M., Armington, J.C., Gray, F.E., 1948. A new electroencephalogram associated with thinking. Science 108, 527–529. Ochoa, C.J., Polich, J., 2000. P300 and blink instructions. Clin. Neurophysiol. 111, 93–98. Semlitsch, H.V., Anderer, P., Schuster, P., Presslich, O., 1986. A solution for reliable and valid reduction of ocular artifacts, applied to the P300. Psychophysiology 23, 695–703. ¨ Verleger, R., Gasser, T., Mocks, J., 1982. Correction of EOG artifacts in event-related potentials of the EEG: aspects of reliability and validity. Psychophysiology 19, 472–480. Verleger, R., 1991. The instruction to refrain from blinking affects auditory P3 and N1 amplitudes. Electroencephalogr. Clin. Neurophysiol. 78, 240–251. Verleger, R., 1993. Valid identification of blink artefacts: are they larger than 50 mV in EEG records? Electroencephalogr. Clin. Neurophysiol. 87, 354–363. Vigario, R.N., 1997. Extraction of ocular artefacts from EEG using independent component analysis. Electroencephalogr. Clin. Neurophysiol. 103, 395–404. Weerts, T.C., Lang, P.J., 1973. The effects of eye fixation and stimulus response location on the contingent negative variation (CNV). Biol. Psychol. 1, 1–19. Whitton, J.L., Lue, F., Moldofsky, H., 1978. A spectral method for removing eye movement artifacts from the EEG. Electroencephalogr. Clin. Neurophysiol. 44, 735–741.