Simultaneous EEG–fMRI at ultra-high field: Artifact prevention and safety assessment


by João Jorge1,2
1Laboratory for Functional and Metabolic Imaging, EPFL, Lausanne, Switzerland
2Institute for Systems and Robotics, IST, Lisbon, Portugal


Simultaneous EEG-fMRI acquisitions can offer valuable insights for the non-invasive study of human brain function (Britz et al., 2010; Gotman and Pittau, 2011; Scheeringa et al., 2011). Concurrently, the benefits offered by high-field imaging, yielding super-linear gains in BOLD sensitivity (van der Zwaag et al., 2009), have attracted considerable interest towards simultaneous EEG-fMRI at higher field strengths (Neuner et al., 2013). Unfortunately, simultaneous acquisitions are subject to problematic interactions that can compromise data quality and subject safety. Safety concerns arise due to the possible generation of electric currents along the EEG wires, induced by the MRI gradients or RF pulses (Dempsey and Condon, 2001). This is increasingly problematic at higher field strengths such as 7T, where the RF pulse energy is higher and the wavelength becomes smaller than the sample size (Eggenschwiler et al., 2012), increasing the risk of resonant antenna effects. Regarding data quality, high-field acquisitions are likewise affected by increasingly stronger artifacts. EEG signals are particularly heavily degraded by magnetic induction effects, with previously less concerning environment noise sources, such as the He compression systems, exhibiting important contributions in recordings at 7T (Mullinger et al., 2008).

Reducing noise during acquisition is crucial to improve EEG data quality, especially at higher fields. This can be done by reducing the areas formed by electrode leads between each channel and the reference, thereby reducing magnetic induction effects. In this work, we assessed the importance of EEG cable length and geometry on noise sensitivity, at 7T, at the level of transmission between the cap and amplifiers. On a phantom model, the effects of different cable configurations were assessed, with specific attention to He coldhead contributions (study I). An optimized EEG setup with short bundled cables (approximately 12cm from cap to amplifiers) was implemented, and a series of safety tests were conducted, including EM simulations and surface temperature measurements on a phantom during fMRI acquisition (study II). Finally, the setup was employed for simultaneous EEG-fMRI acquisition on 5 healthy volunteers undergoing an eyes-open/eyes-closed task and a VEP run (study III).


Optimized EEG-fMRI setup: All measurements were performed on an actively-shielded Magnetom 7T head scanner (Siemens, Erlangen, Germany) equipped with a custom-built 8-channel Tx/Rx loop head array (Rapid Biomedical, Rimpar, Germany). For studies II and III, EEG data were recorded using two 32-channel BrainAmp MR plus amplifiers and a customized 64-electrode BrainCap MR model. The cap was designed with shortened leads terminating in two connectors at approximately 2cm from the cap surface. The cap connectors were linked to the EEG amplifiers via two 12cm bundled cables, with the amplifiers placed just outside the head RF array (Fig. 1).

After bandpass filtering (0.016 – 250Hz) and digitization (0.5μV), the EEG signals were transmitted to the control room via fiber optic cables. EEG sampling was performed at 5kHz, synchronized with the scanner clock. Abralyte gel was used to reduce electrode impedances. The scanner He coldheads were kept in function at all times in study II and study III.

Fig.1. The optimized EEG-fMRI setup implemented in this work.

Fig.1. The optimized EEG-fMRI setup implemented in this work.

Study I – EEG cable noise contributions: This study assessed EEG noise sensitivity depending on the length and geometry of the ribbon cables connecting the cap to the amplifiers. EEG recordings were performed on a phantom, without MRI acquisition, both with and without the coldheads in function. Data were recorded using a single amplifier connected via a ribbon cable to an MR-compatible tester box, which was fixed to the phantom (to capture strictly cable-related noise contributions). Six cable configurations were tested, comprising 3 different lengths (100, 50, and 12cm) and 2 geometries: the typical flat ribbon configuration, and a bundled configuration where all channels are bunched together in a cylindrical shape (Fig. 1d).

Study II – safety testing: Here, a series of tests were performed to evaluate the impact of the optimized setup (12cm bundled configuration) on subject safety and EEG amplifier integrity. EM simulations were performed to study the effects on B1+ and SAR distributions across the head, as generated by the head RF array used in this work. The measurement was simulated in a realistic human model; EEG electrodes, gel, resistors, wire branches and connector positions were modeled according to the real cap. Temperature measurements were conducted on a phantom fitted with the EEG cap. Two probes were placed on electrodes AF8 and FT9, one in between the EEG amplifiers, and another above the phantom, for reference. Temperature fluctuations were assessed during a 16min session comprising two fMRI runs: a sinusoidal GE-EPI sequence (69% of SAR limit) and a SE-EPI sequence (91% of SAR limit).

Study III – human acquisitions: Human tests intended to assess BOLD and EEG data quality using the optimized setup, particularly in terms of functional sensitivity. Five healthy volunteers underwent an eyes-open/eyes-closed run mediated by auditory cues, and a VEP run with reversing-checkerboard stimuli. After acquisition, EEG data underwent gradient and pulse artifact reduction (Allen et al., 2000; Niazy et al., 2005), downsampling, bad channel interpolation and re-referencing to the average reference. For the eyes-open/closed run, data were decomposed via ICA, and then reconstructed by manual selection of the relevant components. For the VEP run, data were first bandpass filtered to 4–30Hz, then ICA-decomposed, and finally reconstructed by selection of VEP-related components (Arrubla et al., 2013). FMRI data analysis comprised motion correction, slice-timing adjustments, brain segmentation, spatial smoothing (2mm) and temporal detrending (Smith et al., 2004). The datasets were then analyzed with a GLM approach (Worsley and Friston, 1995), modeling both paradigms as block designs.


Study I – EEG cable noise contributions: Based on preliminary tests, the patient ventilation system was found to produce relevant noise contributions at frequencies below 30Hz, but could be switched off throughout all recordings (Nierhaus et al., 2013). With the He coldheads in function, using a 100cm conventional (flat) ribbon cable, most channels clearly displayed a stationary noise pattern of high frequency oscillations (> 20Hz), with a fundamental period of approximately 1s (Fig. 2a). A progressive increase in noise amplitude was clearly seen for channels running farther away from the reference, a trend which became greatly attenuated in the bundled geometry. Comparing all the tested cable configurations, the influence of cable length and geometry on noise power was found highly statistically significant, as was the impact of coldhead contributions (p < 0.01). Over all tested lengths, bundled cables yielded reductions of 0.2 – 69% in total noise power relative to flat cables, with the coldheads OFF, and 43 – 63% with the coldheads ON. Conversely, over the two geometry types, shortening from 100 to 12cm yielded reductions of 44 – 70% in noise power with the coldheads OFF, and 58 – 62% with the coldheads ON. Overall, the combination of cable bundling and shortening (from 100 to 12cm) led to a reduction of 84% in total noise power and of 91% in inter-channel noise power variability, with the coldheads in function (Fig. 2b).

Fig.2. Left: EEG recordings using a 100cm flat ribbon cable, with the He coldheads in function. Right: average EEG noise power for different cable configurations, with and without the coldheads in function.

Fig.2. Left: EEG recordings using a 100cm flat ribbon cable, with the He coldheads in function. Right: average EEG noise power for different cable configurations, with and without the coldheads in function.

Study II – safety testing: in EM simulations, the presence of the EEG materials led to a general loss in B1+ amplitude of approximately 8.0%. The general field distribution was roughly maintained, but a number of local accentuated effects were observed in superior regions, mostly restricted to the scalp. SAR maps expressed similar trends, with an overall decrease of approximately 7.9%. A few local increases could be observed in superior-anterior regions, close to the skin, pushing the peak value from 0.39W/Kg to 0.43W/Kg with the cap. Regarding the temperature measurements, during the 16min session, temperature increases in either the reference probe or the two probes placed on EEG electrodes were found to be minimal (below 1°C). The sensor placed on the EEG amplifiers did measure stronger heating effects (from 21.4 to 27.9°C), although remaining well within their operating range.

Study III – human acquisitions: None of the volunteers reported any unusual skin heating effects, and the EEG amplifiers operated normally throughout all acquisitions. The ICA-reconstructed EEG data from the eyes-open/closed run revealed accentuated alpha modulation in occipital channels, with clear alpha power increases during most of the eyes-closed periods. In the fMRI data, significant negative BOLD signal changes were detected for eyes-closed periods in occipital regions, with average signal changes of -3.9% to -3.7%. For the VEP run, all subjects exhibited an average EEG response in occipital regions dominated by a positive peak occurring approximately 100ms after stimulus onset (Fig. 3a). This P100 peak reflected an anterior-posterior dipole, dominating the average GFP response at the same latency. On a single-trial scale, occipital responses were considerably noisier. Nevertheless, a trial-by-trial regression analysis showed that statistically significant responses (p < 0.05) were found in 164 – 177 trials out of 312 for this subject group. In the fMRI data, significant positive signal changes (+3.0% to +3.8%), correlated with checkerboard stimulation periods, were detected in occipital regions for all subjects (Fig. 3b).

Fig.3. EEG (a) and fMRI (b) responses to the VEP run utilizing reversing-checkerboard stimulation.

Fig.3. EEG (a) and fMRI (b) responses to the VEP run utilizing reversing-checkerboard stimulation.


The results obtained in this work demonstrate important benefits of careful optimization of the EEG signal chain for simultaneous EEG-fMRI. Focusing on the transmission stage between the cap and amplifiers, we have confirmed that both cable shortening and bundling effectively help reducing cable noise contributions to large extents. The optimized setup exhibited no significant safety concerns for subjects or amplifiers, the latter probably owing to the use of a Tx/Rx head array for more localized RF transmission. Based on human recordings, we conclude that alpha-wave modulation, VEPs and the concomitant BOLD signal changes can be detected with favorable sensitivity at 7T. Overall, setup improvements such as those here proposed, together with denoising approaches specifically tailored for simultaneous EEG-fMRI, steadily aid to bring this multimodal approach to satisfactory standards of signal quality, allowing for the full exploit of the benefits offered by high-field imaging.

[1] Allen, P.J., Josephs, O., Turner, R., 2000. A method for removing imaging artifact from continuous EEG recorded during functional MRI. Neuroimage 12, 230-239.

[2] Arrubla, J., Neuner, I., Hahn, D., Boers, F., Shah, N.J., 2013. Recording visual evoked potentials and auditory evoked P300 at 9.4T static magnetic field. PLoS One 8, e62915.
[3] Britz, J., Van De Ville, D., Michel, C.M., 2010. BOLD correlates of EEG topography reveal rapid resting-state network dynamics. Neuroimage 52, 1162-1170.
[4] Dempsey, M.F., Condon, B., 2001. Thermal injuries associated with MRI. Clin Radiol 56, 457-465.
[5] Eggenschwiler, F., Kober, T., Magill, A.W., Gruetter, R., Marques, J.P., 2012. SA2RAGE: a new sequence for fast B1+ -mapping. Magn Reson Med 67, 1609-1619.
[6] Gotman, J., Pittau, F., 2011. Combining EEG and fMRI in the study of epileptic discharges. Epilepsia 52 Suppl 4, 38-42.
[7] Mullinger, K., Brookes, M., Stevenson, C., Morgan, P., Bowtell, R., 2008. Exploring the feasibility of simultaneous electroencephalography/functional magnetic resonance imaging at 7 T. Magn Reson Imaging 26, 968-977.
[8] Neuner, I., Arrubla, J., Felder, J., Shah, N.J., 2013. Simultaneous EEG-fMRI acquisition at low, high and ultra-high magnetic fields up to 9.4T: Perspectives and challenges. Neuroimage.
[9] Niazy, R.K., Beckmann, C.F., Iannetti, G.D., Brady, J.M., Smith, S.M., 2005. Removal of FMRI environment artifacts from EEG data using optimal basis sets. Neuroimage 28, 720-737.
[10] Nierhaus, T., Gundlach, C., Goltz, D., Thiel, S.D., Pleger, B., Villringer, A., 2013. Internal ventilation system of MR scanners induces specific EEG artifact during simultaneous EEG-fMRI. Neuroimage 74, 70-76.
[11] Scheeringa, R., Fries, P., Petersson, K.M., Oostenveld, R., Grothe, I., Norris, D.G., Hagoort, P., Bastiaansen, M.C., 2011. Neuronal dynamics underlying high- and low-frequency EEG oscillations contribute independently to the human BOLD signal. Neuron 69, 572-583.
[12] Smith, S.M., Jenkinson, M., Woolrich, M.W., Beckmann, C.F., Behrens, T.E., Johansen-Berg, H., Bannister, P.R., De Luca, M., Drobnjak, I., Flitney, D.E., Niazy, R.K., Saunders, J., Vickers, J., Zhang, Y., De Stefano, N., Brady, J.M., Matthews, P.M., 2004. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 23 Suppl 1, S208-219.
[13] van der Zwaag, W., Francis, S., Head, K., Peters, A., Gowland, P., Morris, P., Bowtell, R., 2009. fMRI at 1.5, 3 and 7 T: characterising BOLD signal changes. Neuroimage 47, 1425-1434.
[14] Worsley, K.J., Friston, K.J., 1995. Analysis of fMRI time-series revisited–again. Neuroimage 2, 173-181.

©Brain Products GmbH 2015

2 0