[EEG] Article on eye movement artifacts & website for passive electrode products

Kelly hurtstotouchfire at gmail.com
Wed Feb 3 15:40:15 PST 2010

Also, forgot to add, there's a bunch of great arguments that follow
this.  It's all signal processing and artifact nerdery, but I found it
fairly hilarious.  I love a good methodology fight.

Deouell's lab has a few links:
http://pissaro.soc.huji.ac.il/~leon/Lab/ (also here's another link to
the text of the original article:

Here's the letter from Melloni et al 2009:

Synchronized oscillatory activity in the gamma frequency range has
been proposed as a neuronal mechanism for various cognitive processes,
ranging from perceptual binding (Singer, 1999) to motor control
(Schoffelen et al., 2005). In a recent paper in Neuron,
Yuval-Greenberg et al. (2008) claim that induced gamma oscillations
recorded by scalp electroencephalography (EEG) reflect miniature
saccades instead of cognitive or neuronal processes. Combining
high-precision eye tracking with EEG recordings, Yuval-Greenberg et
al. found that (1) the induced gamma band response (iGBR) follows
similar temporal dynamics as miniature saccade rate after display
change, (2) only trials containing miniature saccades and its
electrophysiological counterpart, namely the spike potential (SP),
contribute to the iGBR, (3) with a nose reference montage, the time
frequency decomposition of the SP shows a maximum over centro-parietal
electrodes, (4) the iGBR amplitude correlates with the rate of
miniature saccades, and (5) the difference between conditions in terms
of iGBR can be attributed to differences in miniature saccade rate and
amplitude. Based on these findings, Yuval-Greenberg et al. conclude
that the “iGBR is ocular rather than neuronal.” The generality of this
conclusion is questionable based on several points:

1. It has previously been shown that recording EEG signals against a
nose reference is highly prone to ocular artifacts (Trujillo et al.,
2005 L.T. Trujillo, M.A. Peterson, A.W. Kaszniak and J.J. Allen, EEG
phase synchrony differences across visual perception conditions may
depend on recording and analysis methods, Clin. Neurophysiol. 116
(2005), pp. 172–189. Article | PDF (655 K) | View Record in Scopus |
Cited By in Scopus (25)Trujillo et al., 2005). However,
Yuval-Greenberg et al. deliberately rereferenced their data to the
nose. As can be seen in their Figure 1, rereferencing the data to an
average reference results in a distribution of the iGBR with a maximum
around the orbits of the eyes that can hardly be mistaken for cortical
activity. Thus, by using an average reference instead of a nose
reference, artifactual activity can be identified based on its
topography. Furthermore, methods such as current source density (CSD)
analysis that utilize current density instead of voltage minimize the
effect of distant sources. Employing CSD instead of a nose reference,
Trujillo et al. (2005) were able to replicate the results of one of
the first EEG studies of the iGBR (Rodriguez et al., 1999) and
concluded that “although eye movement can contaminate synchrony
measures computed based on a nose reference, they do not appear to
account for all of the between condition differences.” Thus, the
problem of artifactual influences of miniature saccades on the iGBR
has been identified and successfully addressed before.

2. The time-frequency decomposition of a brief impulse such as the SP
shows a broad-band response. Although some studies do report such
broad-band results, other studies have shown either specific effects
for subbands of the gamma band or increases in one frequency band with
parallel decreases in another frequency band ([Ball et al., 2008] and
[Lutzenberger et al., 1995]). Such patterns are incompatible with the
time-frequency representation of an impulse. Furthermore,
Yuval-Greenberg et al. did not provide a time-frequency analysis
aligned to SP onset. Only an instantaneous iGBR can serve as evidence
for an ocular source of the parietal effects, while any time shift
would speak for a central origin caused for example by a neuronal
process that accounts for both the iGBR and the execution of a
miniature saccade.

3. The finding that two dipolar sources placed in the orbits of the
eyes explain most of the signal phase-locked to the saccadic event
could have been predicted. By taking the event-related potential (ERP)
locked to the saccadic event, all other activity not strictly
phase-locked to this event is suppressed (i.e., any cortical iGBR).
Hence, the finding that dipoles placed in this region explain most of
the variance is to be expected, irrespective of the existence or
nonexistence of cortical sources of iGBRs. If source analysis is to be
employed to decide upon this question, methods equally weighting
phase-locked and non-phase-locked activity like frequency domain
variable resolution tomography (FD-VARETA, Fernandez-Bouzas et al.,
1999) or beamformer approaches (Brookes et al., 2008) are most
informative. Additionally, the direct comparison between the iGBR
amplitude topography in Yuval-Greenberg et al.'s Figure 1 and the SP
topography in Yuval-Greenberg et al.'s Figure 6 may be misleading.
This is because the choice of reference has a different influence on
maps of signed raw signals or dipole topographies and on
time-frequency amplitude or power maps, where the absolute value of a
signal is taken either directly or by squaring. This is, for example,
reflected in the well-known fact that source locations cannot be
fitted from scalp topographies reflecting signal power. Apart from
this, the origin of the spike potential itself has not yet been
resolved. Some authors have attributed it to peripheral sources
(Thickbroom and Mastaglia, 1985), while others consider a cortical
origin (Balaban and Weinstein, 1985). At the core of the question at
hand is that the iGBR can only be said to be artifactual if a
peripheral generator is assumed. However, even if peripheral
generators contribute to the scalp-recorded iGBR, their influence can
be dissociated from cortical sources based on the topography of the
SP: based on the reference site, the SP can be maximal at anterior or
parietal electrodes, but it is always lateralized depending on the
horizontal direction of the saccade (Moster and Goldberg, 1990). Only
if the iGBR can be shown to be lateralized in accordance with the
saccade direction, a strong point for the artifactual origin of the
observed iGBR could be made. On the contrary, a lack of lateralization
would render a cortical origin of the iGBR likely. However, even
though Yuval-Greenberg et al. acquired saccade direction information
with a high-precision eye tracker, they do not provide a
saccade-direction-dependent analysis of their results. Thus, direct
evidence for Yuval-Greenberg et al.'s conclusions is still missing.
Instead, they sort out all trials containing miniature saccades,
possibly not only deleting artifactual activity from the dataset, but
also removing the cortical iGBR. This is especially relevant given
recent findings indicating that miniature saccades play an important
role in perception and attention ([Engbert, 2006] and [Martinez-Conde
et al., 2004]).

4. Yuval-Greenberg et al. failed to replicate numerous previous
reports on iGBR in response to familiar objects and faces (e.g.,
Zion-Golumbic et al., 2008). Interestingly, such effects have been
found to not depend on the selection of an average or nose reference
(Supp et al., 2007). Stimulus parameters such as contrast, size,
spatial frequency, and eccentricity as well as attention have been
shown to affect the amplitude and frequency of the evoked gamma band
response (Herrmann et al., 2004). The same holds true for the iGBR
recorded in the primary visual cortex of awake behaving monkeys
(Neuenschwander, personal communication). Thus, it is conceivable that
the stimulus parameters chosen by Yuval-Greenberg et al. were
ineffective in driving iGBRs while being particularly effective in
generating miniature saccades. For example, it is known that high
spatial frequency stimuli induce fewer and smaller miniature saccades
than low spatial frequency stimuli (Armington and Bloom, 1974).

The general statement that the “iGBR is ocular rather than neuronal”
made by Yuval-Greenberg et al. has to be further put into perspective
given that iGBRs have been observed before the time window during
which critical changes in miniature saccade rate occur (Widmann et
al., 2007), and that good correspondence of the iGBR in EEG and
intracranial recordings (which are not affected by muscular artifacts)
has been established in time, frequency, and topography (Ball et al.,
2008). Additional important points have been addressed in online
commentaries on the Neuron homepage.

In face of the missing analyses and the available contradictory
evidence, it is at least premature to conclude that all the iGBRs
measured by EEG can be explained by miniature saccades. However, the
paper by Yuval-Greenberg et al. certainly has its merits in reminding
us that EEG measurements can be contaminated by artifacts. Given that
EEG systems are comparatively cheap and widely available, their usage
in the field of cognitive neuroscience has greatly increased over the
past few years and has led to interesting results. At the same time,
the development of integrated software packages now allows for quick
analysis of the data. However, this might come at the price of losing
a sense of which artifacts may affect the signal and how much the
signal itself is transformed by different processing steps. Systematic
studies of the artifacts that can affect iGBRs are still missing. For
years, researchers have implicitly assumed that the artifacts that
contaminate ERPs, such as blinks and large eye movements, have a
similar influence on the iGBR. However, those artifacts may in fact
have less of an impact on the iGBR, whereas miniature saccades or
muscle artifacts seem to constitute a bigger problem. The challenge
for future research will be to dissociate true from artifactual
sources of the iGBR and to determine stimulus parameters that allow
for or preclude the detection of iGBRs. In conclusion, the results of
Yuval-Greenberg et al. make us aware of the nose, but they should not
discourage the scientific community from utilizing EEG in the study of
cognitive brain processes as indexed by iGBRs.

And the response from Yuval-Greenberg et al:

Empirical science is about seriously considering (and possibly ruling
out) alternative explanations for a given phenomenon. It is within
this framework that this discussion should be addressed. Seminal
intracortical work by Singer and colleagues suggested that neurons
responding to stimuli which are bound, e.g., by Gestalt laws, not only
display a persistent oscillation (i.e., periodic activity) in the
gamma range, but also synchronize the phase of these fluctuations with
each other (Gray et al., 1989). These findings suggest that phase
synchronization could serve for “binding” at the neural level (Singer,
1999). Since phase synchronized activity sums up, it stands to reason
that this “bound” activity could be measured from larger distance, and
it is natural to seek equivalents of these oscillations in the EEG.
(Note that there could be other types of high frequency non
oscillatory activity.) In our study, we pointed out that one of the
most prominent candidates for such an EEG correlate of neural
oscillation, namely the transient-broadband iGBR (iGBRtb), is likely
the wrong candidate. The iGBRtb was hypothesized to be an equivalent
of neural gamma oscillations related to binding or object
representation because of circumstantial evidence: it resembled the
animal findings in having roughly the same frequency, and it was
sensitive to apparently similar manipulations. However, our study
(Yuval-Greenberg et al., 2008) provided instead clear support for an
alternative explanation of the EEG iGBRtb, which went far beyond mere
correlation by showing a straightforward causal chain leading to the
observed potentials. As we explicitly stated, it is the iGBRtb, rather
than all induced gamma band activity, which was the target of our

The iGBRtb is a well defined response characterized by several
distinctive features: trial-to-trial latency jitter (hence “induced”
rather than “evoked”), broad frequency range (not, vert, similar30–80
Hz), relatively short duration (not, vert, similar100–150 ms), a
characteristic poststimulus latency (not, vert, similar200–300 ms),
and a posterior, parieto-occipital peak. We systematically explained
how the combination of two well-documented phenomena—the stereotypical
poststimulus spontaneous saccade-rate modulation (SRM; Rolfs et al.,
2008) and the unavoidable spike potential (SP) that accompanies the
onset of each saccade (Thickbroom and Mastaglia, 1985)—elicit such a
poststimulus average iGBRtb. Melloni et al. do not contest this core
model, which predicts an iGBRtb in most visual paradigms. In our view,
this alone should make any gamma activity resembling the above pattern
(see Melloni et al. [2007] [Figure 2A, not, vert, similar200 ms post
test-word] and Schadow et al., 2009) suspect of being a result of
saccadic SPs, unless direct evidence to the contrary is presented in
each case.

Melloni et al. note, as we did, that potential contamination of iGBRs
by eye movements was noted before our study ([Reva and Aftanas, 2004]
and [Trujillo et al., 2005]). However, these important reports did not
fully realize that the ocular potentials are not a source of random
noise (like blinks) but rather a natural, systematic source of signal
with a typical time course, which ubiquitously affects time-frequency
representations of scalp-measured potentials in visual experiments.
Consequently, despite these previous observations, and despite Melloni
et al.'s conclusion that the problem was “identified and successfully
addressed,” studies reporting the iGBRtb did (and still do) little to
rule out or remove this ocular signal. For example, Melloni et al.
suggest to use Laplacian transforms (CSDs) to attenuate the effect of
distant sources and of the reference (e.g., [Lutzenberger et al.,
1995], [Pulvermuller et al., 1997] and [Trujillo et al., 2005]). While
this may be useful, studies which showed the iGBRtb did not use CSD
(see [Melloni et al., 2007], [Martinovic et al., 2008] and [Schadow et
al., 2009], for recent examples). Moreover, supporting our suggestion
that the iGBRtb has an ocular source, when CSDs were used, the
spectrotemporal morphology of gamma findings in posterior channels was
very different from that of the potentials-derived iGBRtb
([Lutzenberger et al., 1995], [Trujillo et al., 2005] and
Zion-Golumbic et al., 2009 E. Zion-Golumbic, M. Kutas and S. Bentin,
Neural dynamics associated with semantic and episodic memory for
faces: Evidence from multiple frequency bands, J. Cogn. Neurosci.
(2009) in press.[Zion-Golumbic et al., 2009]).

CSDs are also not a magic bullet. Because of the near exponential
decay of the spike potential from front to back, the CSDs will be less
effective in filtering out ocular sources from anterior electrodes,
closer to the source. Therefore, if iGBR is measured as the average of
all CSD channels, as was done by Trujillo et al. (2005) for example, a
contribution from saccades cannot be ruled out.

Melloni et al. fault us for “deliberately” using a nose reference.
This is however one of the most common practices in this field (e.g.,
[Tallon et al., 1995] and [Schadow et al., 2009]). While a nose
reference is indeed especially sensitive to ocular artifacts, no
reference can cancel them out completely. Referencing to the mean of
all electrodes (“average reference”) strongly attenuates the SP
response in electrodes close to the mean (usually around Cz), but
inverts the sign of the SP at posterior electrodes. When the power of
this response is computed, two power peaks emerge, one posterior and
one anterior, which is strongest around the eyes (see our Figure S1;
note that since Melloni et al. found the comparison between the
topographies of the iGBR and the SP which we presented in our paper
confusing, we present here the SP in absolute values to allow direct
comparison of the iGBR and SP's distributions). Melloni et al. only
echo our message when they state that this distribution strongly
supports an ocular rather than a cortical source. Unfortunately,
periocular electrodes are omitted almost unanimously from published
topographic maps of iGBRtb, artifactually highlighting the weaker
posterior peak (Figure S1).

As noted by Melloni et al., the saccadic SPs are lateralized based on
saccade directions. As shown in the supplemental figure, the iGBRtb
indeed lateralizes similar to the SP, so that the data pass the
important test suggested by Melloni et al. Note however that with
miniature saccades, SP lateralization is weak and may not always be
clear, especially when temporal jitter is involved, as in the
computation of the iGBR.

Melloni et al. doubt the orbital source of the spike potential itself
and thus suggest that it is not an artifact. In fact, the last two
decades of literature show a wide consensus for an orbital source
(reviewed in our paper). As a counter-argument, Melloni et al. cite
Balaban and Weinstein (1985), who suggested a cortical source for the
SP, but these authors based their claim on the false premise that the
SP is not elicited by spontaneous saccades, which is clearly not the
case (as we and others have shown). Our dipole model of the SP served
to validate the orbital hypothesis (and was not presented as a
localization of the iGBR as was erroneously interpreted by Melloni et
al.). Moreover, the SP starts with the onset of saccade and peaks not,
vert, similar4 ms after saccade onset. Figure 4 in our paper
demonstrated the perfect temporal synchrony between eye movement onset
and spike potentials. Accordingly, and answering Melloni et al.'s
query, the saccade-aligned iGBR is indeed instantaneous with the
saccade onset (with the limitation of the temporal smearing of wavelet
transformation; See figure in Yuval-Greenberg and Deouell, 2009). This
simultaneity makes a cortical source (either motor or sensory) less
likely. Parenthetically, even if the spike had a nonocular source, it
certainly isn't the oscillatory response we set out to find.

Melloni et al. suggest that by excluding trials in which saccades
occurred within 150–350 ms (Figures 5 and 8 of our paper) we not only
eliminated the spectral signature of the SPs but also excluded the
very trials in which a genuine neural iGBR was produced. That is, they
suggest that a neural iGBR occurs only when saccades occur. Such
extreme correlation between two co-occurring phenomena is
theoretically possible. However, since we provided a simpler account
based on the SPs, this violation of Ockham's Razor rule (or law of
parsimony) puts the burden of the proof on those who want to argue for
the existence of a perfect correlation, and perfect simultaneity,
between two phenomena—cortical synchronization and eye movements.
Whereas Martinez-Condes and colleagues recently suggested that
microsaccades have a function in perception, especially in extrafoveal
regions, and Engbert and colleagues suggested that microsaccades are
affected by attention, these proposals do not predict in any way the
correlation suggested by Melloni et al. In any case, we believe that
direct, data-driven evidence is required rather than indirect
inductive reasoning as suggested by Melloni et al.

Although the iGBRtbs we reported are identical to those previously
reported, Melloni et al. suggest that they may be due to some unique
stimulus parameters that were “ineffective in driving iGBRs while
being particularly effective in generating miniature saccades.” We
believe this is highly implausible. First, numerous studies prior to
ours (reviewed in our paper) established that changes of the visual
display of almost any sort elicit the characteristic SRM. Second,
across the three experiments of our study we used very different
stimuli, with varying spatial frequencies. Third, we recently
replicated the findings connecting SP with iGBRtb in a group of 14
subjects using stimuli similar to those used in previous studies
(Busch et al., 2006). As these studies reported, familiar objects
induced more iGBRtb than unfamiliar objects. Crucially, consistent
with the saccadic origin of the iGBRtb, familiar objects elicited also
significantly more and larger saccades (see
Regarding our experiment 3, we indeed found larger iGBRtb for objects
than faces in apparent contradiction to Zion-Golumbic et al.'s study.
However our “objects” condition was very different from their
“watches” condition in being highly heterogeneous compared to the
homogenous categories of faces and watches. Melloni et al. accurately
note that different stimulus parameters may induce a different rate
and size of saccades—but this is exactly why different stimuli may
elicit different iGBRtbs. This is not to say that different categories
may not elicit different gamma band responses in the brain, which can
be recorded on the scalp (cf. Zion-Golumbic et al., 2009), only that
it is probably not the iGBRtb which reflects these processes.

Melloni et al. point to the existence of iGBR findings with latencies
that do not match the SRM or with narrower frequency band than the
typical iGBRtb. As we explained (in the paper and the ensuing online
correspondence on the Neuron website), such findings may indeed be
induced by other processes, including brain oscillations, and have no
bearing on the issue of the origin of the iGBRtb. Instead, we argue
that findings which do match the iGBRtb morphology should be carefully
evaluated. Moreover, because the critical parameters shaping the SRM
are not fully known (e.g., the effect of different visual or nonvisual
stimuli), even findings that deviate somewhat from the typical SRM
should be treated cautiously. This may also be the case regarding
bandwidth. The SP is a transient but has some width and is therefore
band limited with intertrial variability (see Figures 2 and S1 in our
original study for single-trial samples). Thus, even narrower-band
responses should be carefully evaluated if they match other iGBRtb
characteristics, since they could reflect the peak of a spectrally
wider effect that is significant over a narrower range of frequencies.

As we repeatedly emphasized, we do not imply that our results apply to
“all the iGBRs measured by EEG” as Melloni et al. insinuate.
Theoretically, even the iGBRtb could have a contribution from, or
could sum up with, a second, neural source. However as already noted,
because this hypothesis violates parsimony, such claim (of past or
future studies) needs to be supported by direct data-driven evidence
which accounts for saccadic activity. Conveniently, iGBRs related to
saccades have specific signature properties. The characteristic
spatial distribution can be evaluated by looking at the full scalp
distributions, including periocular channels, with different choices
of reference. Single trials can be examined for evidence of
oscillations of certain duration rather than spikes (e.g., Gray et al.
[1989] required three cycles). This should be performed on the
unfiltered signal (since the SP itself causes “ringing” of narrow-band
filters). Eye tracking can be used to examine the experiment-specific
SRM. As already mentioned, different methods may be used to attenuate
the effect of saccades (e.g., CSDs, ICA, Beamforming), but since there
is no perfect filter, these too should be used with great care. Thus,
we gladly join Melloni et al. in encouraging the scientific community
to continue the search for EEG equivalents of neural iGBR, but with
due scientific skepticism.

On Wed, Feb 3, 2010 at 2:57 PM, Kelly <hurtstotouchfire at gmail.com> wrote:
> http://www.springerlink.com/content/a287xx8n15p56x23/
> This guy is coming to our lab for a year.  He did some awesome work
> that demonstrated that the evoked gamma peak that many researchers
> have written about is actually an artifact of high-frequency saccades.
> Also, talking to my supervisor today, she said that the caps she used
> in France were passive, and they got better signal than our active
> caps here at the lab.  She said they came from this company:
> http://www.neuroscan.com/cart/Details.cfm?ProdID=188&category=6
> I find that encouraging because I had all but given up on passive
> electrodes.  I feel like if there is a right way to do it, we should
> be able to figure that out.  At this point, I've basically decided
> that we need to make our own cap, possibly from scratch. But I'm going
> to start by refitting the current cap with shielded wire.
> -Kelly

More information about the EEG mailing list