Simultaneous electroencephalography (EEG) and functional Magnetic Resonance imaging (fMRI) is a powerful neuroimaging tool. However, the inside of an MRI scanner forms a difficult environment for EEG data recording and safety must be considered whenever operating EEG equipment inside a scanner. Here, we present an optimised EEG-fMRI data acquisition protocol.
Simultaneous EEG-fMRI allows the excellent temporal resolution of EEG to be combined with the high spatial accuracy of fMRI. The data from these two modalities can be combined in a number of ways, but all rely on the acquisition of high quality EEG and fMRI data. EEG data acquired during simultaneous fMRI are affected by several artifacts, including the gradient artefact (due to the changing magnetic field gradients required for fMRI), the pulse artefact (linked to the cardiac cycle) and movement artifacts (resulting from movements in the strong magnetic field of the scanner, and muscle activity). Post-processing methods for successfully correcting the gradient and pulse artifacts require a number of criteria to be satisfied during data acquisition. Minimizing head motion during EEG-fMRI is also imperative for limiting the generation of artifacts.
Interactions between the radio frequency (RF) pulses required for MRI and the EEG hardware may occur and can cause heating. This is only a significant risk if safety guidelines are not satisfied. Hardware design and set-up, as well as careful selection of which MR sequences are run with the EEG hardware present must therefore be considered.
The above issues highlight the importance of the choice of the experimental protocol employed when performing a simultaneous EEG-fMRI experiment. Based on previous research we describe an optimal experimental set-up. This provides high quality EEG data during simultaneous fMRI when using commercial EEG and fMRI systems, with safety risks to the subject minimized. We demonstrate this set-up in an EEG-fMRI experiment using a simple visual stimulus. However, much more complex stimuli can be used. Here we show the EEG-fMRI set-up using a Brain Products GmbH (Gilching, Germany) MRplus, 32 channel EEG system in conjunction with a Philips Achieva (Best, Netherlands) 3T MR scanner, although many of the techniques are transferable to other systems.
Simultaneous electroencephalography (EEG) and functional Magnetic Resonance imaging (fMRI) allows the excellent temporal resolution of EEG to be combined with the high spatial accuracy of fMRI. There are a number of ways in which the data from these two modalities can be combined 1, but all rely on the acquisition of high quality EEG and fMRI data. To date, simultaneous EEG-fMRI has been used to study the correlation between oscillatory rhythms (measured with EEG) and blood oxygenation responses (using blood oxygenation level dependent (BOLD) fMRI) e.g. 2,3. It has also been used to study whether the characteristics of the evoked signal can explain the variance in the BOLD signal on a trial-by- trial basis4,5. In clinical studies the main use of the technique has been to investigate the foci of interictal epileptic discharges, which can help in surgical planning and are currently difficult to localize non-invasively6,7. To achieve the fusion of EEG and fMRI data which is desired, it is essential to have high quality data from both modalities. However, EEG data acquired during simultaneous fMRI are affected by several artifacts, including the gradient artefact (due to the changing magnetic fields required for fMRI), the pulse artefact (linked to the cardiac cycle) and movement artifacts (resulting from movements in the strong magnetic field of the scanner, as well as muscle activity). These artifacts are significantly larger than the neuronal activity of interest and therefore reduction (at source) and correction of the artifacts (via post-processing) are both needed to enable successful implementation of simultaneous EEG-fMRI.
The post-processing methods currently available for correcting the gradient and pulse artifacts require a number of criteria to be satisfied during data acquisition in order to produce high quality EEG data. Over the previous decade the optimal experimental set-up for recording high quality data has evolved as our understanding of the causes of the artifacts 8-10 has improved and we have learnt how to modify experimental methods so as to reduce the artifacts at source 11,12 and to improve the performance of post-processing correction algorithms. These developments include improving the sampling of the gradient waveforms through synchronization of scanner clocks 13,14 and use of a vectocardiogram15,16 to provide a cleaner cardiac trace than the traditional ECG. The vectocardiogram trace is derived from four electrodes placed on the chest with a stringent low-pass filter employed14-16. As a result the trace is relatively unaffected by gradient artifacts and is insensitive to the blood flow artifact making R-peak detection easier. However, the facility to record a vectocardiogram is not available on all MRI scanners and therefore will only be mentioned briefly in this study. The importance of the minimization of artifacts and stringent cleaning of data has been highlighted by the recent demonstration that motion artifacts recorded in the EEG data can correlate with BOLD activity unrelated to the task of interest, producing spurious results if extreme care is not taken throughout the experimental process 17.
The method presented here represents the current optimal approach for obtaining high quality EEG and fMRI data simultaneously using MR hardware and pulse sequences which are widely available, along with commercially supplied EEG equipment. Implementation of the suggested acquisition method, in conjunction with the use of appropriate post-processing methods, will yield EEG and fMRI data that can be used to answer a number of important neuroscience questions.
1. Preparing the Experimental Setup
2. Subject Arrival
3. Recording Outside of the MR Scanner
(Optional: Only Required if You Wish to Compare EEG Data Quality from Inside and Outside the MR Scanner)
4. Setting Subject up Inside the MR Scanner
5. Recording Inside the Scanner
6. Debriefing the Subject
7. Clearing up at the End of Experiment
8. Analysis
Figure 3 shows the signal quality to be expected when no artifact correction has been performed. It is clear that any neuronal activity is obscured. Figure 3C shows that the gradient artifact occurs at distinct frequencies which are harmonics of the frequency of slice acquisition in the fMRI sequence, spanning the entire frequency range of the recording. Figure 4 shows the pulse artifact which is revealed once the gradient artifact has been removed using the post-processing method of average artifact subtraction in Analyzer 2 (version 2.0.2). It is clear that there is considerable spatial variation of this artifact and that O1, one of the channels of interest for this visual experiment, displays a particularly large pulse artifact. This artifact has a lower frequency than the gradient artifact (mainly below 10 Hz – Figure 4C) and is linked to the cardiac activity. Figure 5 shows the EEG data quality that can be achieved after gradient and pulse artifact correction; here the pulse artifact was corrected using average artifact subtraction in Analyzer 2 and the R-peaks of the cardiac waveform were detected from the ECG trace. It is clear that the amplitude of the remaining signals are far smaller and therefore neuronal signals are no longer obscured, as shown by the evoked responses obtained in Figures 6 and 7. Figure 6 shows a typical evoked response produced by averaging across all 300 stimuli. However, the variability of this response across blocks can be seen in Figure 7 and it is this natural and unpredictable variation in neuronal responses which may be used to interrogate correlations between the BOLD and EEG responses when simultaneous recordings have been performed.
Figure 1. A schematic diagram of the set-up of the EEG equipment and the connections required between hardware, as described in the protocol. Click here to view larger figure.
Figure 2. Fourier transform of the signal collected on a subject lying still with the cryo-pumps on (red) and off (black) for a representative channel (P7).
Figure 3. Ten seconds of raw EEG data recorded during concurrent MRI on 16 different channels (A); focusing on 5 seconds of data from Oz (B); with the associated Fourier transform (C). Click here to view larger figure.
Figure 4. Ten seconds of EEG data recorded on 16 different channels during concurrent MRI shown after gradient artefact correction using AAS on 16 different channels (A); focusing on 5 seconds of data from Oz (B); with the associated Fourier transform (C). Click here to view larger figure.
Figure 5. Ten seconds of EEG data recorded on16 different channels during concurrent fMRI, shown after gradient and pulse artefact correction using AAS (A); focusing on 5 seconds of data from Oz (B); with the associated Fourier transform (C). Click here to view larger figure.
Figure 6. Average EEG evoked response (300 averages) for channels 01 and 02 (left) and associated topographic map for the P120 (right).
Figure 7. Variation of evoked response across blocks for channel O1 (responses have been averaged within 30 sec blocks).
General Advice Since the physical layout of all scanner rooms is different we recognize that you may not be able to position your EEG amplifiers outside the bore of the magnet. In this case a good compromise is to place the amplifiers on a thick rubber pad so as to decouple them from the scanner vibrations as much possible. If you find that the gradient artifact correction is not working well, then check the times between volume or slice markers, as it is likely in this case that the TR that has been input to the MR console is not precisely the TR that is being generated. In this case you will need to contact the relevant MR scanner manufacturer for further assistance.
The most important steps in the process of EEG data acquisition during simultaneous fMRI are those taken to ensure that all external noise sources have been minimized (e.g. cyrocooler pumps and vibration of the EEG equipment). To allow optimal gradient artifact correction it is important to ensure that the EEG and MR scanner clocks are synchronized, the slice TR is a multiple of the scanner clock period and that the subject is optimally positioned. To ensure optimal pulse artifact correction many techniques require a clean cardiac trace from which R-peaks can be detected, we suggest that this can be best achieved using a VCG, although it is also possible with a well-positioned ECG lead. If using the ECG then it is recommended to place this at the base of the back to maximize the signal to noise ratio of the R-peak with the added benefit of this being an easier site to access than a position near the heart23. Positioning the ECG lead on the chest results in motion artifacts due to respiration being added to the trace from this lead as well as causing the gradient artifact to vary over time. This can result in the trace saturating and/or gradient artifact correction not working due to template variability and therefore is not recommended.
General Discussion EEG-fMRI is a powerful tool for studying brain function, as the high temporal resolution of EEG can be combined with the high spatial resolution of fMRI. To date, a number of studies have used this multi-modal approach to gain a better understanding of brain function. EEG-fMRI has been applied to healthy volunteers in order to investigate the correlation between oscillatory rhythms (measured with EEG) and blood oxygenation responses (using BOLD fMRI) e.g. 2,3. It has also been used to study whether characteristics of the evoked signal can explain the variance in the BOLD signal on a trial-by- trial basis4,5. In clinical studies the main use of the technique has been to investigate the foci of interictal epileptic discharges which are inherently difficult to localize non-invasively6,7. These examples show the power of this multi-modal imaging tool. However, to enable the study of such phenomena, it is important to have access to the best possible quality of EEG and MRI data. To achieve this inside the MR scanner it is important to have the best experimental set-up and also to choose the most appropriate analysis methods. The optimal analysis methods will to some extent depend upon the research question of interest, as will the correction methods used for removal of artifacts. For example the size and number of movements that have occurred during the recording will determine the most effective combination of algorithms for removing the gradient artifact. However, the optimal experimental set-up of the EEG and fMRI hardware is relatively independent of particular research questions. The guidelines outlined here are therefore of general value and can be followed in experiments using different EEG and MR scanner hardware than we used.
Here we have demonstrated the acquisition methods which should be followed to acquire high quality EEG and fMRI data. We used a visual stimulus based on a previously employed stimulus paradigm 24. However, the same techniques for data acquisition can be applied regardless of the paradigm used to stimulate the brain activity of interest. When choosing your paradigm it should be noted that the quality of the EEG data that can be achieved when recording inside the MR environment with the techniques currently available to users (and described here) still place some limitations on the brain activity which may be studied: there are particular difficulties in recording EEG activity in low (<5 Hz) and high frequency (>80 Hz) bands where residual pulse and gradient artifacts may reside. Additionally, care must be taken when choosing the paradigm so that the possibility of subject movement related to the task is minimized. This is a problem because motion artifacts in the EEG data are often difficult to correct and small artifacts can be difficult to identify clearly, although they still may dominate neuronal signals. These motion artifacts can cause spurious but plausible correlations with the fMRI data17.
Post-processing methods for simultaneous EEG-fMRI are numerous and as such their discussion is beyond the scope of this work. As previously mentioned the gradient and pulse artifact can be removed using a number of techniques which include average artifact subtraction18,19, independent component analysis20,21, optimal basis sets22 and beamformers25. Often a combination of these methods may be employed23 and the performance of the methods is dependent upon factors such as the magnetic field strength and the paradigm used. The optimal post-processing methods for a specific study will also depend on the signals to extract from the data, whether these are oscillatory rhythms or evoked potentials may have an influence on the post-processing methods employed.
Whilst there is considerable on-going research targeting improved data acquisition and analysis methods for simultaneous EEG-fMRI, it is already possible, using the techniques described here, to answer important neuroscience questions which require the combination of the high spatial resolution of fMRI and the excellent temporal resolution of EEG.
The authors have nothing to disclose.
We would like to thank Brain Products GmbH for providing their equipment, expertise and help in producing this work. We would also like to thank Glyn Spencer, University of Nottingham, in assisting with the production of the video. We also thank Engineering and Physical Science Research Council (EPSRC), EP/J006823/1 and University of Nottingham for funding this research.
Name of Reagent/Material | Company | Catalog Number | Comments |
3T MR scanner | Here we use a Philips Achieva but any MR scanner should work. | ||
BrainVision Recorder | Brain Products GmbH | BP-00010 | 1st License item |
BrainVision RecView | Brain Products GmbH | BP-00051 | basis module |
BrainAmp MR plus | Brain Products GmbH | BP-01840 | single amplifier |
BrainAmp USB Adapter | Brain Products GmbH | BP-02041 | BUA64 |
SyncBox | Brain Products GmbH | BP-02675 | SyncBox complete |
Fibre Optic cables and USB connectors | Brain Products GmbH | BP-02300 (FOC5) BP-02310 (FOC20) BP-02042 USB2 Cable) | These come with the above listed equipment. |
BrainCap MR | EASYCAP GmbH | BP-03000-MR | 32 channel EEG cap for use in MR |
Abralyte 2000 conductive Gel | Brain Products GmbH | FMS-060219 | Conductive and abrasive gel to connect electrodes to scalp |
Isopropyl Alcohol BP | Brain Products GmbH | FMS-060224 | To be applied before Abralyte Gel. Isopropylalcohol 70% (60 ml)-for degreasing the skin |
Cotton tipped swab | Brain Products GmbH | FMS-060234 | For application of Abralyte and Isopropyl Alcohol. Cotton Swabs Non-sterile, 100 pieces |