Scanning strategies for simultaneous EEG–fMRI evoked potential studies at 3 T

Available online at www.sciencedirect.com International Journal of Psychophysiology 67 (2008) 169 – 177 www.elsevier.com/locate/ijpsycho Scanning strategies for simultaneous EEG–fMRI evoked potential studies at 3 T Tracy Warbrick 1 , Andrew P. Bagshaw ⁎ School of Psychology and Birmingham University Imaging Centre (BUIC), University of Birmingham, Birmingham, B15 2TT, United Kingdom Received 27 February 2007; accepted 25 May 2007 Available online 12 July 2007 Abstract There are two basic strategies for applying simultaneous EEG–fMRI: either the fMRI data are acquired continuously, or the stimulus is presented during a brief gap in scanning when the EEG data is clear of gradient artefact. The former has the advantage that the protocol for the fMRI data acquisition is not affected by the presence of EEG. This study investigated the effect of these different strategies and the subsequent ballistocardiogram artefact removal methods (Average Artefact Subtraction (AAS) and Optimal Basis Set (OBS)) on EEG data quality recorded in response to a visual stimulus. Continuous scanning generally resulted in VEPs that were no worse, and in some cases were better, than those measured during a gap in scanning. The AAS and OBS methods lead to comparable results at the level of the grand average visual evoked potential (VEP), although when examined at the level of the single trial the OBS method was more effective. The spectral quality of the data was similar across scanning protocols, as demonstrated by the proportion of spectral power in each frequency band, although there was an effect of the artefact removal method on the overall spectral power. Some differences in the VEPs were also noted when a TR of 1.5 s was used relative to a TR of 3 s. The results indicate improved EEG quality when fMRI scanning is continuous and BCG artefacts are removed using the OBS method, confirming that EEG can be added to an fMRI experiment with minimal change to the experimental protocol. © 2007 Elsevier B.V. All rights reserved. Keywords: Simultaneous EEG–fMRI; EEG data quality; BCG artefact removal; VEPs 1. Introduction The combination of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) is becoming increasingly popular as a neuroscientific method because it promises improved spatial and temporal resolution, and with it an improved ability to decode the dynamic processes that are at the heart of brain function (Horwitz & Poeppel, 2002; Dale et al., 2000; Debener et al., 2006; Ritter & Villringer, 2006). Simultaneous EEG–fMRI recordings provide additional experimental benefits such as equal environmental factors, no training effects on behavioural tasks, and no differences in the subjects' cognitive or emotional state between recording sessions. However, there are a number of problems inherent in recording EEG in the ⁎ Corresponding author. Tel.: +44 121 414 3683; fax: +44 121 414 4897. E-mail addresses: t.warbrick@bham.ac.uk (T. Warbrick), a.p.bagshaw@bham.ac.uk (A.P. Bagshaw). 1 Tel.: +44 121 414 8836; fax: +44 121 414 4897. 0167-8760/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpsycho.2007.05.014 MRI environment. The EEG data are susceptible to contamination from artefacts, specifically from gradient switching during scanning (the gradient artefact, Allen et al., 2000) and from currents induced by pulse-associated movements which are amplified by the static magnetic field (the ballistocardiogram (BCG) artefact, Allen et al., 1998). These artefacts need to be dealt with appropriately to retrieve meaningful information from EEG data recording during scanning. At least three solutions to the problem of the gradient artefact have been proposed. The first is to have a gap during each volume acquisition to provide a period free of gradient switching when relatively clean EEG data can be collected. The second is to scan continuously and use an appropriate algorithm for artefact removal. While collecting EEG data during a gap in scanning has proved successful (Bonmassar et al., 2002; Kruggel et al., 2000) there is an obvious benefit to scanning continuously since EEG data can be recorded without modifying the scanning protocol. In practice, this means that more slices can be acquired per unit time, leading to more complete brain coverage or 170 T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 increased spatial resolution. Since an fMRI EPI sequence typically has a very regular pattern of gradient switching it thus induces very regular artefact patterns which can be removed by subtraction of the average artefact, provided it is reproduced faithfully in the EEG (Allen et al., 2000). This can be ensured by using hardware with an appropriate dynamic range and sampling rate. A third solution also employs continuous scanning but works by sampling the EEG at specific time points when the gradient artefact is smallest (Anami et al., 2003). This requires much closer integration of the EEG acquisition system with the MRI scanner, and modification of the imaging sequence, but residual artefacts are considerably smaller. Removal of the BCG artefact can be more problematic. Since it is caused by slight movements of the electrodes and leads inside the static magnetic field (see Allen et al., 1998 for more details), an average artefact template is more difficult to calculate than for the gradient artefact due to variations in shape, amplitude and scale over time (Niazy et al., 2005). The effects of the BCG artefact can be reduced by ensuring all electrodes and electrode leads are secure (Allen et al., 1998; Bénar et al., 2003), and as it is dependent on the static magnetic field the effects are more pronounced at 3 T compared to 1.5 T. For example, when scanning at 1.5 T Becker et al. (2005) found no impact of the BCG artefact on their ability to recover event-related potentials (ERPs), whereas at 3 T contamination from BCG artefact is readily apparent (Debener et al., 2007). A number of methods have been proposed for effective removal of BCG artefact, from average artefact subtraction (AAS, Allen et al., 1998) to spatial independent and principal component analysis (ICA and PCA, Bénar et al., 2003; Mantini et al., 2007; Nakamura et al., 2006). A recently developed method is based on the use of temporal PCA to construct an optimal basis set (OBS) which allows variability in the artefact template (Niazy et al., 2005). It has been shown to result in good quality EEG data (Iannetti et al., 2005; Debener et al., 2007). The purpose of the current study was to investigate the combination of different strategies for dealing with gradient and BCG artefacts in simultaneous EEG–fMRI data acquired at 3 T. Although the effect of the alternative methods has been investigated in isolation (e.g., gap vs. continuous scanning (Becker et al., 2005), AAS vs. OBS (Niazy et al., 2005)), the optimal combination of approaches is still not clear. An additional aim was to determine whether the results are dependent on the repetition time (TR) of the fMRI scanning protocol. This has not previously been investigated since earlier studies have used a single TR. Since artefact removal could be affected by TR, it is important to determine the extent to which results can be generalised independent of the scanning protocol. To this end, subjects were scanned with continuous EEG–fMRI using two different TRs (3 s and 1.5 s), with and without a 1 s gap in each volume acquisition. Data were analysed with two methods of BCG artefact removal (AAS and OBS). 2. Method 2.1. Subjects Five healthy volunteers (one female) participated in the study. The mean age of the subjects was 28.4 (±3.8) y. The study was approved by the Birmingham University Imaging Centre (BUIC) research ethics committee and written informed consent was obtained from all subjects. 2.2. Experimental protocol EEG and fMRI data were recorded concurrently while subjects viewed visual stimuli via a mirror attached to the head coil. Visual stimuli were presented using Presentation v10.1 (Neurobehavioral Systems Inc, CA, USA) and consisted of a full field, high contrast checkerboard reversing at 2 Hz. Stimuli were presented for 1 s every 15 s, and were delivered in four runs of twenty stimuli, one run per MRI acquisition protocol (see below). Each run lasted approximately 5 min. Subjects were asked to remain as still as possible and to minimise eye blinks during data acquisition. To help minimise eye movements and to ensure that subjects paid attention to the task, they were asked to fixate on a circle in the middle of the visual field in the break between stimuli. 2.3. FMRI data acquisition Subjects were scanned in a 3 T Philips Achieva scanner (Philips Medical Systems, Best, Netherlands) with an eight element phased array receive head coil. A standard T1-weighted anatomical scan was initially acquired (1 mm isotropic voxels). Four different fMRI acquisition protocols were used: TR = 3 s continuous (49 slices), TR = 3 s gap (33 slices), TR = 1.5 s continuous (24 slices), TR = 1.5 s gap (8 slices). The protocols differed only in the number of slices acquired, with all other imaging parameters kept constant (voxel size 2.5× 2.5 × 2.5 mm, TE = 35 ms, SENSE factor 2 (Pruessman et al., 1998), flip angles 85° and 75° for TR = 3 s and 1.5 s respectively). The length of the gap was 1 s with the first checkerboard reversal occurring 100 ms into the gap. Slices were oriented parallel to the calcarine sulcus. 2.4. EEG data acquisition EEG data were recorded using a 32 channel MR compatible EEG system (BrainProducts, Munich, Germany), which incorporates current limiting resistors of 5 kΩ at the amplifier input and in each electrode. The EEG cap consisted of 30 scalp electrodes distributed according to the 10–20 system and two additional electrodes, one of which was attached approximately 2 cm below the left collarbone for recording ECG, while the other was attached below the left eye (on the lower orbital portion of the orbicularis oculi muscle) for detection of eyeblink artefacts. Data were sampled at 5000 Hz, with a bandpass of 0.016–250 Hz. Impedance at all recording electrodes was less than 10 kΩ. 2.5. Data analysis Raw EEG data were subjected to gradient artefact correction in BrainVision Analyzer (BVA) (BrainProducts, Munich, Germany). An event marker was sent from the scanner to the EEG recording at the start of every volume acquisition; this marker was then used to identify the onset of the gradient artefact in order to create a template for subtraction (Allen et al., 1998). Data were then T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 171 downsampled to 500 Hz. Once free of gradient artefact, the data underwent two different methods of BCG artefact correction: AAS, as implemented in BVA, and OBS (Niazy et al., 2005, Centre for Functional MRI of the Brain (FMRIB), Oxford University, UK), which is available as a plug-in to EEGLAB (http://sccn.ucsd.edu/eeglab/, Delorme & Makeig, 2004). The AAS method applied using BVA uses the repetition of the artefact across time to construct the artefact template, as such it requires accurate detection of R peaks. In this study the R peaks were detected semi-automatically, with manual adjustment for peaks misidentified by the software. To average the artefact in the EEG channels the R peaks are transferred from the ECG to the EEG over a selectable time delay. In this study the time delay was calculated for each subject using the CBC Delay Power macro in BrainVision Analyzer. The average artefact was then subtracted from the EEG. For a more detailed description of this method see Allen et al. (1998). The OBS method developed by Niazy et al. (2005) involves automatic detection of the QRS events. Basis functions describing temporal variation in the artefact are identified using temporal Principal Components Analysis (PCA). These basis functions are then fitted to and subtracted from the EEG. For further details of this method see Niazy et al. (2005). Once gradient and BCG artefacts had been removed the data were segmented into 450 ms epochs (−50 ms to 400 ms) for the purposes of ERP analysis. The artefact-subtracted raw data were also segmented into 800 ms epochs (0–800 ms) for frequency analysis to assess spectral quality of the data. The longer epoch for spectral analysis started at the first checkerboard reversal, and was hence 100 ms into the gap. It thus covered the majority of the gap while allowing 100 ms at the beginning and end to avoid contamination by gradient artefacts or their residuals. Both sets of epochs were inspected for artefacts resulting from eye blink or other muscular sources, and any epoch containing a voltage change of more than 150 μV was rejected. For the 450 ms epochs averages were created, resulting in visual evoked potentials (VEPs) for each subject for each condition. P100 and N140 peaks were then manually detected in each averaged VEP. Finally, a grand average VEP was created for each condition. For the 800 ms epochs, the power density function in each frequency band was computed using FFT (Hanning window, window length 10%) and then an average was created for each subject for each condition. Wilcoxon signed ranks tests were used to examine differences in P100–N140 amplitude between conditions. The data for one subject for continuous scanning with a TR of 3 s was excluded from the analysis due the lack of a clear VEP. 3. Results 3.1. Effect of a gap in scanning For both TRs and BCG artefact removal techniques, visual inspection of the grand average VEPs showed P100–N140 amplitudes to be slightly larger for continuous scanning compared to when there was a gap in scanning (Fig. 1). Qualitatively, the VEPs were also much clearer and the peaks easier to identify for continuous scanning. Furthermore, this effect can be seen Fig. 1. ERPs at O2 for continuous scanning (heavy line) and a gap in scanning (light line) for a TR of 1.5 s (left) and 3 s (right) and BCG artefact rejection performed using Average Artefact Subtraction (top) and an Optimal Basis Set (bottom). The P100–N140 is clearer for continuous scanning in most cases. 172 T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 Fig. 2. Average ERPs at Cz for 2 individual subjects (AB and MK) for a TR of 3 s and BCG artefact rejection using OBS. ERPs recorded during continuous scanning are qualitatively cleaner and larger in amplitude than those recorded during a gap in scanning at the level of the individual subject. consistently in individual subjects (Fig. 2). For example when comparing continuous scanning and a gap in scanning for the OBS artefact removal technique at both TRs, visual inspection of the individual average VEPs shows the amplitude of the VEP to be similar for both methods (e.g. subject AB in Fig. 2) or larger in amplitude for continuous than a gap in scanning (e.g. subject MK in Fig. 2) for 9 of the 10 comparisons made. The clarity of VEPs recorded during continuous scanning is emphasised by examining the data at the level of the single trial (ST). Fig. 3 takes the data from one representative subject (TR = 3 s) and presents it in the form of a stacked plot or ERP image (Jung et al., 2001), where each trial is shown with the amplitude colour coded. It can be seen that Fig. 3. Stacked plots of single ERP trials at O2 for a gap in scanning (left) and continuous scanning (right) for BCG artefact rejection using Average Artefact Subtraction (top) or an Optimal Basis Set (bottom). ERPs are sorted vertically in order of occurrence, the first trial is at the bottom and the last trial is at the top. Continuous scanning with OBS artefact removal leads to the clearest ST P100 responses (bottom right). T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 Table 1 Mean and range for P100–N140 amplitude and P100 and N140 latency Scan type Continuous BCG artefact removal method AAS OBS Gap AAS OBS TR (s) 1.5 3 1.5 3 1.5 3 1.5 3 P100–N140 amp. (mV) 11.6 12.5 11.3 13.2 11.6 13.7 8.4 8.9 P100–N140 range (mV) 6.4–17.6 7.2–15.1 6.6–14.5 5.6–19.4 4.2–17.4 6.7–18.8 2.3–14.5 3.2–16.9 P100 lat (ms) 104.4 102.4 108.8 108 97.5 96 102.4 113.2 P100 lat range (ms) 86–122 76–122 98–130 90–149 88–108 80–108 94–118 96–164 N140 lat (ms) 174.8 158 170.4 168 153.6 172 157.6 149.6 173 N140 lat range (ms) 144–214 130–198 142–218 136–216 132–174 140–204 134–198 126–174 This illustrates the differences in mean P100–N140 amplitude between continuous and gap in scanning and BCG artefact removal methods. In addition, the variability in amplitude and latency across subjects is demonstrated by the inclusion of the range for each measure. continuous scanning and OBS BCG artefact removal leads to considerably clearer ST data, as demonstrated by the consistent positivity at approximately 100 ms (i.e. the ST P100). To gain a quantitative measure of VEP amplitudes the P100– N140 peak-to-peak amplitude was measured for each subject. The mean amplitude of the P100–N140 complex and the mean latencies of the P100 and N140 components can be seen in Table 1. Furthermore, examples of averaged VEPs for each individual subject are displayed in Fig. 4 to illustrate amplitude and latency range. Wilcoxon signed rank tests showed no significant differences between continuous scanning and a gap in scanning for the P100–N140 amplitude (p N .05). 3.2. Effect of BCG artefact rejection technique The analysis above indicated that good quality ERPs could be gained from continuous scanning periods. In order to examine the effect of BCG artefact removal in more detail, data from continuous scanning were analysed further. The need for BCG artefact correction at 3 T is clearly illustrated in Fig. 5, while Fig. 6 shows the VEPs for each BCG artefact correction method at both TRs. The peak to peak amplitude was similar for AAS and OBS when the TR was 1.5 s. However, with a TR of 3 s peak to peak amplitude was larger for OBS. This point is also clearly made in Fig. 3, where the stacked plots indicate that at a ST level OBS is more effective in dealing with BCG artefact than AAS. Wilcoxon signed rank tests showed no significant differences in peak to peak amplitude between AAS and OBS at either TR (p N .05). 3.3. Effect of TR Fig. 7 shows the same data as for Fig. 6 but illustrates the difference in TR more effectively; visual inspection of the VEPs shows that peak to peak amplitude was similar for TR = 1.5 s and TR = 3 s for both BCG artefact correction methods. However, the ERPs peaks appeared to be cleaner for TR = 1.5 s than TR = 3 s, and the effect of BCG artefact removal method was more pronounced at TR = 1.5 s, this effect was consistent across subjects. However, Wilcoxon signed rank test showed that this difference was not significant (p N .05). 3.4. Spectral quality of the data To assess the spectral quality of data collected under each condition FFT was performed on the 800 ms epoch data. Both scanning protocols with a TR of 3 s were assessed for each BCG artefact rejection technique. Fig. 8 shows the power spectral density for a representative subject for each condition. The spectral quality of the data appears to be similar across scanning protocols, as demonstrated by the proportion of spectral power in each frequency band. However, the two BCG artefact rejection methods result in power spectra of markedly different amplitudes. The mean activity in all frequency bands is larger in amplitude for AAS compared to OBS. This was seen in all 5 subjects. The mean activity for each frequency band in each condition is presented in Table 2. 4. Discussion The primary aim of the study was to determine the optimum combination of fMRI scanning protocol (continuous or with a gap) and BCG artefact removal method (AAS or OBS) in order to optimise the quality of EEG data recorded in a 3 T Philips Achieva MRI scanner. To date, both continuous (Debener et al., 2007; Iannetti et al., 2005; Wan et al., 2006) and gap (Bonmassar et al., 1999; Foucher et al., 2003; Kruggel et al., 2000; Otzenberger et al., Fig. 4. Average VEPs for all 5 subjects for continuous scanning with a TR of 1.5 s and BCG artefact removal using OBS, illustrating that clean responses were seen in all of the individual data that forms the basis of the group averages. 174 T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 Fig. 5. Raw EEG data from one representative subject illustrating contamination from BCG artefact (a) and the same 10 s of data after BCG artefact correction using an Optimal Basis Set (b). Part c shows the grand average Visual Evoked Potentials (continuous scanning, TR =3 s) from uncorrected data, and following correction using an OBS and AAS. Fig. 6. Visual Evoked Potentials (VEPs) at O2 for continuous scanning with both TRs and both BCG artefact rejection using AAS and an OBS. The OBS method results in clearer VEPs with a TR of 3 s, however, there is little difference in the amplitude or clarity of the VEP with a TR of 1.5 s. T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 175 Fig. 7. Visual Evoked Potentials for TR = 1.5 and TR = 3 (continuous scanning) across BCG artefact rejection methods. The VEP peaks appear to be cleaner for TR = 1.5 s than TR = 3 s, suggesting that the effect of BCG artefact removal method was more pronounced at TR = 1.5 s. 2005) protocols have successfully been used to examine the relationship between ERP and fMRI measures of stimulus response, but clearly it is desirable to scan continuously in order to maximise the efficiency of scanning and to simplify the design of the experimental procedure. From this point of view, it is not necessary for the EEG data acquired from continuous scanning epochs to be significantly improved for the method to be preferred, rather it is necessary to demonstrate that EEG data quality is no worse. The results presented here confirm that continuous scanning is a viable strategy to allow measurement of good quality VEPs, independent of the method used to remove the BCG artefact and the TR of the fMRI protocol. The method used to remove the gradient artefact, which is based on Allen et al. (1998), relies on accurate timing of the trigger pulses sent from the scanner to the EEG acquisition computer at the start of each volume. With the MRI scanner used in the current study (3 T Philips Achieva), this was the case and gradient artefact removal was successfully Fig. 8. Power spectral density at O2 for one representative subject for each scanning protocol (TR = 3 s) after BCG artefact correction using Average Artefact Subtraction (AAS) and an Optimal Basis Set (OBS). The mean activity in all frequency bands is larger in amplitude for AAS compared to OBS. 176 T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 Table 2 Mean activity (mV) at O2 for each frequency band after BCG artefact correction using Average Artefact Subtraction (AAS) and an Optimal Basis Set (OBS), illustrating that mean activity was generally higher for AAS compared to OBS Continuous AAS Delta Theta Alpha Beta Mean SD Mean SD Mean SD Mean SD 19.21 3.56 45.59 25.12 14.11 4.89 2.83 1.69 OBS 8.34 6.18 6.52 2.72 3.04 0.91 1.11 0.20 Gap AAS 23.05 21.74 26.95 30.12 3.94 2.49 0.95 0.64 OBS 13.01 7.68 9.65 4.97 5.07 3.42 1.52 0.83 The frequency bands used were: delta =1–3.9 Hz, theta=4–7.9 Hz, alpha =8– 12.9 Hz and beta =13–35 Hz. accomplished. It has been suggested that the method may be improved by synchronising the EEG and fMRI acquisition, which has been shown to result in a larger usable bandwidth (Mandelkow et al., 2006). In the current study, visual inspection and quantification of the peak amplitudes and latencies of the P100–N140 components show that continuous scanning results in VEPs that do not differ in amplitude and latency from those recorded during a gap in scanning. Indeed in some cases, Figs. 1 and 3 for examples, the VEP recorded during continuous scanning is larger in amplitude and has clearer peaks than those recorded during a gap in scanning. This result is consistent with previous work at 1.5 T (Becker et al., 2005; Comi et al., 2005; Sammer et al., 2005), but required further validation to ensure that the same conclusion could be extended to data recorded at 3 T. Furthermore, the time-course and topography of task related ERP component modulation has been shown to be consistent in EEG data collected outside and inside of the scanner during continuous scanning (Bregadze and Lavric, 2006), providing additional evidence that the hostile MRI environment does not preclude the recording of good quality EEG data either from the technical or physiological points of view. The current study also made a comparison between two BCG artefact rejection techniques: Average Artefact Subtraction (AAS), as implemented in BrainVision Analyzer, and an Optimal Basis Set (OBS), accessed via the EEGLAB plug-in. The grand average VEPs (Fig. 4) show little difference between BCG artefact correction methods with a TR of 1.5 s, suggesting that the BCG artefact rejection technique used has less influence on the quality of the ERP. However, with a TR of 3 s there is a greater difference between the results for AAS and OBS, with the VEP being larger in amplitude and generally clearer after BCG artefact correction using OBS (Table 1, Fig. 5). The reason for this effect of TR is unclear, especially since the data that underwent the two BCG artefact removal techniques had already had the gradient artefact removed. Previous studies have not examined EEG data quality when scanning with different TRs, and more work will be needed to determine whether this is a reproducible finding. If so, the choice of TR would need to be made with EEG data quality in mind. The stacked plots of Fig. 3 show that at a single trial (ST) level continuous scanning and OBS BCG artefact removal leads to considerably cleaner data. This has implications for ST EEG–fMRI recording, a method that has recently been proposed for combining EEG and fMRI data (Bénar et al., in press; Debener et al., 2005, 2006, Eichele et al., 2005). ST EEG–fMRI requires estimates of the amplitude and/or latency of a particular ERP component for every trial. These are then convolved with the haemodynamic response function and used as a regressor in the fMRI analysis. The results presented in Fig. 3 demonstrate a considerable advantage of OBS in this respect. Fig. 3 also highlights the need to draw conclusions concerning the most appropriate artefact correction method based on the application being undertaken. Although grand average VEPs were of reasonable quality with either BCG artefact removal technique (Fig. 4), it seems unlikely that good quality ST component amplitudes could be extracted independently of the BCG removal method (Fig. 3). Choosing an artefact correction method from grand average VEP quality and generalising to a study concerning single trial VEPs would therefore lead to sub-optimal results, since it has been demonstrated in this study that OBS leads to considerably better single trial data. On a practical note, a further advantage of the OBS method is the robust procedure for automatically identifying the QRS complexes of the ECG (Niazy et al., 2005, based on a modification of an earlier method of Christov 2004), which makes the process of artefact removal much less time consuming. Regarding the spectral quality of the data, the amplitude in each frequency band is considerable larger when data are subjected to BCG artefact correction using AAS than when using OBS. The reason for this is unclear, but is likely is due to the subtraction of different components of the EEG signal in each method. Both AAS and OBS subtract artefact templates, the primary difference being that OBS allows more variation in the template to take account of differences in the shape and amplitude of the artefact from pulse to pulse. The current data does not allow us to conclude which method is optimal, only that the two BCG artefact removal methods result in different amplitude power spectra. Further investigation of the spectral quality of the data, and effects of the artefact correction techniques, is important as differences in EEG data collected inside and outside of the scanner have been observed. For example, Sammer et al. (2005) found non-specific amplitude differences between data recorded inside and outside the scanner. These differences were not related to the effects of interest and their source could not be identified. However, effects of this type, particularly if they are frequency dependent, could have a considerable effect on the use of EEG–fMRI to study oscillatory EEG activity (Laufs et al., 2003; Parkes et al., 2006). This will need to be examined using tasks that can induce power changes in specific bands and, as with the ST data, the conclusions drawn from average ERPs may not generalise well. In conclusion, it is possible to retrieve good quality ERPs from EEG data recorded during continuous scanning at 3 T. The OBS method appears to be more effective than the AAS method for removing BCG artefact, particularly when analysis at the single trial level is required. The results provide further confirmation that the technical aspects of recording EEG data in the MRI canner can be readily overcome, and that with techniques for artefact removal the benefits of the integration of EEG and fMRI data can be applied to a wide range of problems in neuroscience. T. Warbrick, A.P. Bagshaw / International Journal of Psychophysiology 67 (2008) 169–177 177 References Allen, P., Polizz, G., Krakow, K., Fish, D.R., Lemieux, L., 1998. Identification of EEG Events in the MR scanner: The problem of pulse artifact and a method for its subtraction. NeuroImage 8, 229–239. Allen, P., Josephs, O., Turner, R., 2000. A method for removing imaging artifact from continuous EEG recorded during functional MRI. NeuroImage 1, 230–239. Anami, K., Mori, T., Tanaka, F., Kawagoe, Y., Okamoto, J., Yarita, M., Ohnishi, T., Yumoto, M., Matsuda, H., Osamu, S., 2003. Stepping stone sampling for retrieving artefact-free electroencephalogram during functional magnetic resonance imaging. NeuroImage 19, 281–295. Becker, R., Ritter, P., Moosmann, M., Villringer, A., 2005. Visual evoked potentials recovered from fMRI scan periods. Hum. Brain Mapp. 26, 221–230. Bénar, C.-G., Schön, D., Grimault, S., Nazarian, B., Burle, B., Roth, M., Badier, J.-M., Marquis, P., Liegeois, C., & Anton, J.-L. In press. Single-trial analysis of oddball event related potentials in simultaneous EEG–fMRI. Hum. Brain Mapp. Bénar, C.-G., Aghakhani, Y., Wang, Y., Izenberg, A., Al-Asmi, A., Dubeau, F., Gotman, J., 2003. Quality of EEG in simultaneous EEG–fMRI for epilepsy. Clin. Neurophysiol. 114, 569–580. Bonmassar, G., Anami, K., Ives, J., Belliveau, J.W., 1999. Visual evoked potential (VEP) measured by simultaneous 64-channel EEG and 3 T fMRI. NeuroReport 1, 1893–1897. Bonmassar, G., Purdon, P.L., Jaaskelainen, I.P., Chiappa, K., Solo, V., Brown, E., Belliveau, J.W., 2002. Motion and ballistocardiogram artifact removal for interleaved recording of EEG and EPs during MRI. NeuroImage 16, 1127–1141. Bregadze, N., Lavric, A., 2006. ERP differences with vs. without concurrent f MRI. Int. J. Psychophysiol. 62, 54–59. Christov, I.I., 2004. Real-time electrocardiogram QRS detection using combined adaptive threshold. Biomed. Eng. Online 3 (1), 28. Comi, E., Annovazzi, P., Martins Silva, A., Cursi, M., Blasi, V., Cadioli, M., Inuggi, A., Falini, A., Comi, G., Leocani, L., 2005. Visual evoked potentials may be recorded simultaneously with f MRI scanning: A validation study. Hum. Brain Mapp. 24, 291–298. Dale, A.M., Liu, A.K., Fischl, B.R., Buckner, R.L., Belliveau, J.W., Lewine, J.D., Hapgren, E., 2000. Dynamic statistical parametric mapping: combining fMRI and MEG for high-resolution imaging of cortical activity. Neuron 2, 55–67. Debener, S., Ullsperger, M., Siegel, M., Fiehler, K., von Cramon, D., Engel, A.K., 2005. Trial-by-trial coupling of concurrent electroencephalogram and functional magnetic resonance imaging identifies the dynamics of performance monitoring. J. Neurosci. 25, 11730–11737. Debener, S., Ullsperger, M., Siegel, M., Engel, A.K., 2006. Single trial EEG–fMRI reveals the dynamics of cognitive function. Trends Cogn. Sci. 10, 558–563. Debener, S., Strobel, A., Sorger, B., Peters, J., Kranczioch, C., Engel, A.K., Goebel, R., 2007. Improved quality of auditory event-related potentials recorded simultaneously with 3-T fMRI: Removal of the ballistocardiogram artefact. NeuroImage 34, 587–597. Makeig, A., Makeig, S., 2004. EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. J. Neurosci. Methods 134, 9–21. Eichele, T., Specht, K., Moosmann, M., Jongsma, M.L.A., Quiroga, R.Q., Nordby, H., Hugdahl, K., 2005. Assessing the spatiotemporal evolution of neuronal activation with single-trial event-related potentials and functional MRI. Proc. Natl. Acad. Sci. U.S.A. 102, 17798–17803. Foucher, J.R., Otzenberger, H., Gounot, D., 2003. The BOLD response and the gamma oscillations respond differently than evoked potentials: an interleaved EEG–fMRI study. BMC Neurology 4 (22). Horwitz, B., Poeppel, D., 2002. How can EEG/MEG and fMRI/PET Data be combined? Hum. Brain Mapp. 1, 1–3. Iannetti, G.D., Niazy, R.K., Wise, R.G., Jezzard, P., Brooks, J.C.W., Zambreanu, L., Vennert, W., Matthews, P.M., Tracey, I., 2005. Simultaneous recording of laser evoked brain potentials and continuous, high-field functional magnetic resonance imaging in humans. NeuroImage 28, 708–719. Makeig, T.-P., Makeig, S., Westerfield, M., Townsend, J., Courchesne, E., Sejnowski, T.J., 2001. Analysis and visualization of single-trial eventrelated potentials. Hum. Brain Mapp. 14, 166–185. Kruggel, F., Wiggins, C.J., Herrman, C.S., von Cramon, D.Y., 2000. Recording of the event related potentials during functional MRI at 3.0 T field strength. Magn. Reson. Med. 44, 277–282. Laufs, H., Kleinschmidt, A., Beyerle, A., Eger, E., Salek-Haddadi, A., Preibisch, C., Krakow, K., 2003. EEG-correlated fMRI of human alpha activity. NeuroImage 1, 1463–1476. Mandelkow, H., Halder, P., Boesiger, P., Brandeis, D., 2006. Synchronization facilitates removal of MRI artefacts from concurrent EEG recordings and increases usable bandwidth. NeuroImage 32, 1120–1126. Mantini, D., Perrucci, M.G., Cugini, S., Ferretti, A., Romani, G.L., Del Gratta, C., 2007. Complete artifact removal for EEG recorded during continuous fMRI using independent component analysis. NeuroImage 34, 598–607. Nakamura, W., Anami, K., Mori, T., Saitoh, O., Cichocki, A., Amari, S., 2006. Removal of ballistocardiogram artifacts from simultaneously recorded EEG and fMRI data using independent component analysis. IEEE Trans. Biomed. Eng. 53, 1294–1308. Niazy, R.K., Beckman, C.F., Iannetti, G.D., Brady, J.M., Smith, S.M., 2005. Removal of fMRI environmental artifacts from EEG data using optimal basis sets. NeuroImage 28, 720–737. Otzenberger, H., Gounot, D., Foucher, J.R., 2005. P300 recordings during eventrelated fMRI: a feasibility study. Cogn. Brain. Res. 23, 306–315. Parkes, L., Bastiaansen, M.C.M., Norris, D.G., 2006. Combining EEG and f MRI to investigate the post-movement beta rebound. NeuroImage 29, 685–696. Pruessman, K., Weiger, M., Scheidegger, M.B., Boesiger, P., 1998. SENSE: sensitivity encoding for fast MRI. Magn. Reson. Med. 4, 952–962. Ritter, P., Villringer, A., 2006. Simultaneous EEG–fMRI. Neurosci. Biobehav. Rev. 30 (6), 823–838. Sammer, G., Blecker, C., Gebhardt, H., Kirsch, P., Stark, R., Vaitl, D., 2005. Acquisition of typical EEG waveforms during fMRI: SSVEP, LRP, and frontal theta. NeuroImage 24, 1012–1024. Wan, X., Riera, J., Iwata, K., Takahashi, M., Wakabayashi, T., Kawashima, R., 2006. The neural basis of the hemodynamic response nonlinearity in human primary visual cortex: implications for neurovascular coupling mechanism. NeuroImage 32, 616–625.