Technical Note
sh
, G
Cha
Berlin School of Mind and Brain & Mind and Brain Institute, Philosophical Institute Humboldt University Berlin, Germany
Article history:
Received 25 November 2008
Revised 13 May 2009
Accepted 8 June 2009
Available online 16 June 2009
NeuroImage 48 (2009) 94–108
Contents lists available at ScienceDirect
NeuroIm
j ourna l homepage: www.e lIntroduction
Simultaneous acquisition of electroencephalography (EEG) and
functional magnetic resonance imaging (fMRI) aims at combining the
strength of both methods and investigating spontaneous and event-
related brain activity with high temporal and spatial precision.
However, MR-related artifacts that are induced in the EEG constitute
a serious challenge (Laufs et al., 2008; Ritter and Villringer, 2006;
Salek-Haddadi et al., 2003). Appropriate removal of those artifacts is
essential for the recovery of the true underlying neuronal signal and
various methods have been developed for that purpose. The
importance of a thorough evaluation of these methods with respect
to artifact removal and deterioration of the underlying signal has been
demonstrated (Grouiller et al., 2007; Ritter et al., 2007). Numerous
types of EEG signatures obtained during fMRI have been investigated
so far, e.g. epileptiform activity (Benar et al., 2006; Hamandi et al.,
2008; Krakow et al., 2001; Lemieux et al., 2001; Salek-Haddadi et al.,
2002), single-trial error-related negativity (Debener et al., 2005),
cortical oscillations such as delta (Czisch et al., 2004; Dang-Vu et al.,Abbreviations: AAS, averaged artifact subtraction; BCG, ballistocardiogram; BOLD,
blood oxygen level dependent; DC, direct current; EEG, electroencephalography; FMRI,
functional magnetic resonance imaging; HFB, high-fr
artifact correction; PCA, principal component analysis;
receiver-operator characteristic; SEP, somatosensory ev
noise ratio; STD, standard deviation.
⁎ Corresponding author. Berlin NeuroImaging Center
Charité Universitaetsmedizin, Charitéplatz 1, 10117 Berl
560 952.
E-mail address: frank.freyer@gmail.com (F. Freyer).
1053-8119/$ – see front matter © 2009 Elsevier Inc. Al
doi:10.1016/j.neuroimage.2009.06.022conclusion, ultrafast EEG oscillations can be continuously monitored during fMRI using the proposed
approach.
© 2009 Elsevier Inc. All rights reserved.a b s t r a c t
Although solutions for imaging-artifact correction in simultaneous EEG–fMRI are improving, residual
artifacts after correction still considerably affect the EEG spectrum in the ultrafast frequency band above
100 Hz. Yet this band contains subtle but valuable physiological signatures such as fast gamma oscillations
or evoked high-frequency (600 Hz) bursts related to spiking of thalamocortical and cortical neurons. Here
we introduce a simultaneous EEG–fMRI approach that integrates hard and software modifications for
continuous acquisition of ultrafast EEG oscillations during fMRI. Our approach is based upon and extends
the established method of averaged artifact subtraction (AAS). Particularly for recovery of ultrahigh-
frequency EEG signatures, AAS requires invariantly sampled and constant imaging-artifact waveforms to
achieve optimal imaging-artifact correction. Consequently, we adjusted our acquisition setup such that both
physiological ultrahigh-frequency EEG and invariantly sampled imaging artifacts were captured. In addition,
we extended the AAS algorithm to cope with other, non-sampling related sources of imaging-artifact
variations such as subject movements. A cascaded principal component analysis finally removed remaining
imaging-artifact residuals. We provide a detailed evaluation of averaged ultrahigh-frequency signals and
unaveraged broadband EEG spectra up to 1 kHz. Evoked nanovolt-sized high-frequency bursts were
successfully recovered during periods of MR data acquisition afflicted by imaging artifacts in the millivolt
range. Compared to periods without imaging artifacts they exhibited the same mean amplitudes, latencies
and waveforms and a signal-to-noise ratio of 72%. Furthermore we identified consistent dipole sources. Ina r t i c l e i n f oUltrahigh-frequency EEG during fMRI: Pu
imaging-artifact correction
Frank Freyer a,c,⁎, Robert Becker a,c, Kimitaka Anami b
a Berlin NeuroImaging Center and Department of Neurology, Charité Universitaetsmedizin,
b Musashino Mental Hospital, Saitama, Japan
c Bernstein Center for Computational Neuroscience, Berlin, Germany
d Max Planck Institute for Brain and Cognitive Sciences, Leipzig, Germany
eequency burst; IAC, imaging-
RMS, root mean square; ROC,
oked potential; SNR, signal-to-
and Department of Neurology,
in, Germany. Fax: +49 30 450
l rights reserved.ing the limits of
abriel Curio a,c, Arno Villringer a,c,d,e, Petra Ritter a,c,e
ritéplatz 1, 10117 Berlin, Germany
age
sev ie r.com/ locate /yn img2008), theta (Sammer et al., 2007; Scheeringa et al., 2008), alpha (de
Munck et al., 2007; Goldman et al., 2002; Goncalves et al., 2006; Laufs
et al., 2003a; Martinez-Montes et al., 2004; Moosmann et al., 2003),
rolandic/mu (Ritter et al., 2008c), or beta (Laufs et al., 2003b)
rhythms, as well as somatosensory- (Iannetti et al., 2005; Schubert et
al., 2008), auditory- (Benar et al., 2007; Eichele et al., 2005; Liebenthal
et al., 2003), and visually-evoked potentials (Becker et al., 2005;

95F. Freyer et al. / NeuroImage 48 (2009) 94–108Bonmassar et al., 1999; Comi et al., 2005). Despite the technical
challenges that complicate particularly the assessment of high-
frequency EEG acquired during fMRI, cortical rhythms up to 50 Hz
related to speech processing (Giraud et al., 2007) and to fMRI resting
state networks (Mantini et al., 2007b) have already been successfully
investigated. However, also EEG oscillations at higher frequencies are
known to be of physiological relevance. For example, attention and
memory processing involve frequencies up to 150 Hz (for detailed
reviews see Jensen and Colgin, 2007; Jensen et al., 2007). Additionally,
high-frequency bursts (HFBs) at around 600 Hz found in human EEG
recordings during somatosensory stimulation (Curio, 2000, 2005;
Hashimoto, 2000) have been shown to reflect spiking activity in
primary somatosensory cortex (Baker et al., 2003). With amplitudes
in the nanovolt-range and durations of a few milliseconds, they are
among the most subtle physiological signals measurable with EEG.
We have previously been able to record HFBs in an interleaved EEG–
fMRI setup (Ritter et al., 2008b), i.e. HFBs were acquired from non
MR-acquisition (‘non-scan’) periods between subsequent MR-acqui-
sition (‘scan’) periods. Using this approach, we showed that
amplitude modulations of different HFB components temporally
separated by milliseconds were associated with anatomically
characteristic and distinct blood oxygen level dependent (BOLD)
signal changes along the thalamocortical pathway. The limitation of
HFB analysis to non-scan periods, however, is an obvious drawback,
reducing fMRI signal-to-noise ratio (SNR) and design flexibility, i.e.
the continuous monitoring of HFBs with simultaneous EEG–fMRI is
desirable. Continuous assessment of EEG signatures above 100 Hz (in
the following referred to as ultrafast/ultrahigh-frequency band)
however, is particularly difficult since imaging artifacts mainly due to
gradient switching predominantly contaminate this spectral range. In
a phantom study it has been recently shown for the first time that
imaging artifacts can be successfully removed up to 150 Hz using an
advanced EEG–fMRI setup (Mandelkow et al., 2006). However,
physiological ultrahigh-frequency EEG signals in the range of
600 Hz have not yet been recovered from fMRI acquisition periods.
The purpose of this study was twofold:
(1) The recovery of ultrahigh-frequency EEG signatures during
fMRI acquisition by introducing a new imaging-artifact correction
(IAC) approach.
(2) The demonstration of applicability of this approach to the EEG
spectrum below the ultrahigh-frequency range.
Purpose (1) was achieved by the following steps:
(a) For data recording, we minimize sampling-based variations of
the imaging-artifact waveform, in order to fulfill the prerequi-
site for optimal IAC based on averaged artifact subtraction (AAS,
Allen et al., 2000).
(b) To cope with remaining variations based on other sources we
optimize our IAC algorithm.
(c) We evaluate the IAC performance in the ultrahigh-frequency
range by comparing HFB features, their corresponding sources,
SNR characteristics as well as unaveraged spectra between MR-
acquisition and non-acquisition periods given all systematic
differences between these two conditions are attributable to
imaging-artifact residuals.
Purpose (2) was achieved by a corresponding evaluation of EEG
features below the ultrahigh-frequency range.
In addition to the imaging artifact which is related to gradient
switching, EEG simultaneously acquiredwith fMRI is also contaminated
by the ballistocardiogram (BCG), which is a physiological cardiac-
related artifact generated by heart rate associated movements of the
electrodes in the static B0 field of the MR scanner (Allen et al., 1998;Bonmassar et al., 2002; Debener et al., 2008). For both our two
purposes, the BCG is of no relevance aswewill explain later in the paper.
Methods
Subjects
Simultaneous EEG–fMRI was acquired from 10 healthy right-
handed subjects (8 female, 21–31 years). All subjects gave written
informed consent according to the Declaration of Helsinki prior to
investigation. The study was conducted in compliance with the
relevant laws and institutional guidelines and approved by the ethics
committee of the Charité University Medicine Berlin. During the
experiment subjects were lying relaxed in a supine position in the
dimly lit bore of the MR scanner. They were instructed to remain
awake with open eyes and to pay attention to the stimulus.
Experimental design
Somatosensory evoked potentials (SEPs) were elicited by transcu-
taneous constant-current electro-stimulation (DS7A, Digitimer, Hert-
fordshire, England) of the right median nerve using square-wave
pulses of 0.2 ms duration and an intensity of 1.5 times above
individual motor threshold. Using ERTS software (BeriSoft Coopera-
tion, Frankfurt, Germany) we employed a blocked stimulation-versus-
rest design. Stimulus repetition rate was set to 8 Hz. When applying
the modified MR sequence (see section fMRI recording parameters)
memory limitations on the MR array processor delimited the number
of volume acquisitions per recording session. Therefore for each
subject two subsequent sessions of 220 volumes were recorded. Each
session lasted 15 min and consisted of 22 blocks (25 s duration per
block, separated by 15 s rest periods, 200 trials per block), resulting in
an overall number of 4400 trials per session.
In order to exclude factors other than the imaging artifact that
might influence HFBs acquired in the MR-environment, in five of the
ten subjects one session outside the MR scanner was additionally
recorded using the same settings.
Synchronization of EEG and fMRI
Synchronizing the clocks of the EEG and fMRI devices prevents
variant sampling of the imaging artifacts and therefore considerably
improves IAC, particularly in the ultrahigh-frequency range (Anami et
al., 2003; Mandelkow et al., 2006; Ritter et al., 2008a). Our IAC
approach used here is based on AAS (Allen et al., 2000), where
imaging artifacts are removed by subtracting an artifact template,
which is estimated by averaging over a number of gradient onset-
locked epochs. Hence consistency across imaging-artifact epochs over
time is a key prerequisite for successful IAC with AAS. In order to
maximize consistency of imaging artifacts through invariant sam-
pling, the internal clocks of the MR scanner and the EEG were
synchronized using a frequency divider (DC-5, Physio Tech, Tokyo,
Japan), which receives the 4 MHz TTL signal from the MR clock and
sends out a 5 kHz signal driving EEG sampling. We assessed the EEG
autocorrelation sequence and ensured that there is no periodic-like
behavior indicating failure of synchronization (Mandelkow et al.,
2006).
EEG recording parameters
EEG recordings were conducted with a 32 channel MR-compatible
EEG system (BrainAmp MR Plus, Brain Products, Munich, Germany)
and an MR-compatible EEG-cap (Easy cap, FMS, Herrsching-Breit-
brunn, Germany), which comprised ring-type sintered silver chloride
electrodes with iron-free copper leads. 21 scalp electrodes were
arranged according to the International 10-20 System with the