crs
re f
Revised 23 December 2008
Accepted 12 January 2009
Available online 24 January 2009
NeuroImage 45 (2009) 1144–1150
Contents lists available at ScienceDirect
NeuroIm
e lSimultaneous EEG–fMRI recordings provide a powerful tool to
increase the functional resolution in brain imaging experiments
since they synergistically profit from the temporal accuracy of the
EEG and the high spatial resolution of the fMRI recordings. The
combination of both methods has been used in studies on sleep
(Czisch et al., 2002; Horovitz et al., 2008), on brain oscillations as
alpha (de Munck et al., 2007; Goldman et al., 2002; Goncalves et
al., 2006; Laufs et al., 2003; Mantini et al., 2007; Moosmann et al.,
2003), rolandic beta (Ritter et al., 2008) or theta (Scheeringa et al.,
2008) activity, on event-related potentials (ERP) (Benar et al., 2007;
Debener et al., 2006; Eichele et al., 2005; Mulert et al., 2008) and
on pathological states such as epilepsy (Benar et al., 2002; Krakow
et al., 2001).
or MR signal quality (Krakow et al., 2000) has been reduced to an
acceptable amount (Mullinger et al., 2008a). More troublesome are
the influences of the MR system on the EEG signal quality. Both the
large steady static magnetic field (B0) and the fast time varying
fields generated by the imaging sequence of the MR system induce
artefacts in the EEG signal. First, pulsatile cardiac-related move-
ments of the electrode leads in the high static magnetic field
induce voltages that are superimposed upon the cerebral signals
(Ives et al., 1993). A number of methods have been proposed to
correct this so called ballistocardiogram (BCG) artefact (Allen et al.,
1998; Benar et al., 2003; Bonmassar et al., 2002; Debener et al.,
2007; Niazy et al., 2005). Secondly, switching magnetic fields and
high frequency pulses during the MR image acquisition induces MR
imaging artefacts that are up to 50 times larger in amplitude thanSimultaneous EEG–fMRI measurements
ging because both methods mutually infl
☆ Parts of this study were presented at the 14th Annu
for Human Brain Mapping, June 15–19, 2008, Melbourn
⁎ Corresponding author.
E-mail address: moosmann@gmail.com (M. Moosma
1053-8119/$ – see front matter © 2009 Elsevier Inc. All
doi:10.1016/j.neuroimage.2009.01.024effect of the EEG equipment on patient safety (Lemieux et al., 1997)Introductionduring fMRI acquisition. The correction techniques proposed so far cannot optimally represent transitions, i.e.
when abrupt changes of the artefact properties due to head movements occur. The algorithm developed here
takes the head movement parameters from the fMRI signal into account to calculate adequate EEG artefact
templates and subsequently correct the distorted EEG data. The data reported in this work demonstrate that
the realignment parameter-informed algorithm outperforms the commonly used moving average algorithm
if head movements occur. The superiority is reflected by comparing the residual variance after artefact
correction with either method. The residual variance is lower around head-movements that exceed head
deflections of about 1 mm when applying the realignment parameter-informed algorithm. Additionally, the
signal to noise ratio of a surrogate event-related potential (ERP) increased by 10–40% for head displacements
larger than 1 mm. The algorithm developed here is particularly suited for studies where head movements of
the subject cannot be prevented as in studies with patients, children, or during sleep. Furthermore, the
enhanced signal to noise ratio of a single trial ERP indicates the power of the presented algorithm for single
trial ERP-fMRI studies in which EEG signal quality is a critical factor.
© 2009 Elsevier Inc. All rights reserved.Article history:
Received 28 October 2008In this work we introduce a new algorithm to correct the imaging artefacts in the EEG signal measuredTechnical Note
Realignment parameter-informed artefact
EEG–fMRI recordings☆
Matthias Moosmann a,⁎, Vinzenz H. Schönfelder b, Ka
Helge Nordby a, Kenneth Hugdahl a,d
a Department of Biological and Medical Psychology, University of Bergen, Norway
b Bernstein Center for Computational Neuroscience Berlin, Germany
c Radboud University Nijmegen, Donders Institute for Brain, Cognition and Behaviour, Cent
d Division of Psychiatry, Haukeland University Hospital, Bergen, Norway
a b s t r a c ta r t i c l e i n f o
j ourna l homepage: www.are technically challen-
uence each other. The
al Meeting of the Organization
e, Australia.
nn).
rights reserved.orrection for simultaneous
ten Specht a,d, René Scheeringa c,
or Cognitive Neuroimaging, Nijmegen, Netherlands
age
sev ie r.com/ locate /yn imgphysiological EEG signals. For this reason, the first combined EEG–
fMRI studies employed interleaved EEG recordings where EEG data
was only collected between periods of fMRI acquisition (Seeck et
al., 1998; Warach et al., 1996). Alternatively, the fMRI was only
recorded after the EEG event of interest (Krakow et al., 1999).
Obviously, these methods have drawbacks in terms of efficiency.
The advent of EEG amplifiers with a high dynamic range resolving

small physiological signals as well as large MR scanner related
artefacts principally allowed the correction of MR artefacts and
thus continuous EEG recording.
As a first step to correct the MR-imaging related artefacts,
template subtraction either in the time (Allen et al., 2000) or
frequency (Hoffmann et al., 2000; Sijbers et al., 1999) domain are
most commonly used. An average artefact template is calculated
and then subtracted from the raw EEG signal. Recently, a second
processing step that employs optimal basis sets of orthogonal
components to filter residual imaging artefacts was proposed
(Niazy et al., 2005).
Residual variances after artefact correctionwere identified as being
caused by jitter of the internal clocks of the EEG and fMRI leading to
heterogeneous imaging artefacts. Several methods have been pro-
posed to reduce jitter related residual variance (Goncalves et al., 2007;
Negishi et al., 2004). A significant improvement of the corrected EEG
signal quality was achieved by synchronization of EEG and fMRI clocks
to prevent jittering (Anami et al., 2003; Mandelkow et al., 2006;
Mullinger et al., 2008b).
However, when the head of the subject moves during data
acquisition, the increased heterogeneity of the imaging artefacts
impairs EEG signal quality (Laufs et al., 2008). Head motions alter the
geometry of electrodes and cables in the magnetic field and
consequently the induced artefacts properties change. Whereas
head movements in fMRI data are commonly corrected by the
realignment procedure (Friston et al., 1996), this issue has not been
addressed for MR gradient contaminated EEG data.
The correction techniques proposed so far are adequate for
homogeneous data or slow drifts of the artefact properties but
commonly used algorithm, which is based on artefacts template
subtraction (Allen et al., 2000).
Methods
We compared the commonly used moving average algorithm
which is based on the approach by Allen et al. (2000) with the
realignment parameter informed algorithm (RP-informed) in 18
datasets. Twelve datasets stem from an earlier EEG–fMRI study
(Scheeringa et al., 2008). Additionally, 6 datasets were recordedwhere
subjects deliberately performed head movements. These datasets
were added to cover a larger range of the extent of head-movements.
EEG data were recorded with an MR compatible EEG amplifier
(Brainamp MRplus, Brainproducts, Munich, Germany) and an MR
compatible 32-channel electrode cap (Easy cap, Falk Minow Services,
Herrsching-Breitbrunn, Germany). EEG data were recorded at 5 kHz
and impedances were kept below 10 kΩ. All recordings were done
with Brain Vision Recorder software (Brainproducts, Munich, Ger-
many). The average of a representative set of 6 electrodes (C1, C2, O1,
O2, T7, and T8) was used for further analysis and comparison. No
filters were applied to the EEG signal. Functional imaging was
recorded with a 1.5 T scanner (Magnetom Vision, Siemens, Erlangen,
Germany) using a T2⁎-weighted gradient echo planar imaging
sequence (repetition time TR=2.34/2.1 s, acquisition time TA=2.29/
2.0 s, echo time TE=30/60 ms, flip angle 90°, matrix 64×64, voxel size
3.5/3×3.5/3×3/3mm3, interslice distance 3mm, 33/20 slices, 280/200
scans for the two different datasets). The internal clocks of the EEG
and the MR scanner were synchronized to avoid jitter problems
(Anami et al., 2003; Mandelkow et al., 2006).
a he
ts b
ht).
1145M. Moosmann et al. / NeuroImage 45 (2009) 1144–1150cannot optimally represent transitions, i.e. when abrupt changes of
the artefact properties occur. Here, we propose an artefact
correction method that uses head movement parameters from the
fMRI signal to calculate adequate EEG artefact templates and
subsequently correct distorted EEG data (see Fig. 1 for an illustration
of the method). The newly developed realignment parameter
informed (RP-informed) algorithm is compared to the most
Fig. 1. Illustration of artefact correction methods used in this work. In the event of
schematically indicated on the left. The conventional moving average algorithm correc
realignment parameter informed algorithm (RP-informed) respects the discontinuity (rig
to correct the red artefact.Moving average algorithm
The standard artefact correction method calculates individual
templates which are subtracted from respective artefact periods to
correct the MR-imaging related artefacts. The templates are calculated
from a moving average of a constant number of artefact volumes
ad movement (vertical dashed line) the artefact properties may change abruptly as
y averaging over the discontinuity caused by the head movement (middle) while the
The grey blocks indicate which artefacts are taken into account to calculate the template