Effects of simultaneous EEG recording on MRI data quality at 1.5, 3 and 7 tesla

Available online at www.sciencedirect.com

International Journal of Psychophysiology 67 (2008) 178 – 188 www.elsevier.com/locate/ijpsycho

Effects of simultaneous EEG recording on MRI data quality at 1.5, 3 and 7 tesla Karen Mullinger a , Stefan Debener b , Ronald Coxon a , Richard Bowtell a,⁎ a b

Sir Peter Mansfield Magnetic Resonance Centre, School of Physics and Astronomy, University of Nottingham, Nottingham, NG7 2RD, United Kingdom Institute of Hearing Research, MRC Institute of Hearing Research Southampton Section, Department of Otolaryngology Royal South Hampshire Hospital, Southampton, SO14 0YG, United Kingdom Received 28 February 2007; accepted 11 June 2007 Available online 9 August 2007

Abstract Although the focus of attention on data degradation during simultaneous MRI/EEG recording has to date largely been upon EEG artefacts, the presence of the conducting wires and electrodes of the EEG recording system also causes some degradation of MRI data quality. This may result from magnetic susceptibility effects which lead to signal drop-out and image distortion, as well as the perturbation of the radiofrequency fields, which can cause local signal changes and a global reduction in the signal to noise ratio (SNR) of magnetic resonance images. Here, we quantify the effect of commercially available 32 and 64 electrode caps on the quality of MR images obtained in scanners operating at magnetic fields of 1.5, 3 and 7 T, via the use of MR-based, field-mapping techniques and analysis of the SNR in echo planar image time series. The electrodes are shown to be the dominant source of magnetic field inhomogeneity, although the localised nature of the field perturbation that they produce means that the effect on the signal intensity from the brain is not significant. In the particular EEG caps investigated here, RF inhomogeneity linked to the longer ECG and EOG leads causes some reduction in the signal intensity in images obtained at 3 and 7 T. Measurements of the standard deviation of white matter signal in EPI time series indicates that the introduction of the EEG cap produces a small reduction in the image signal to noise ratio, which increases with the number of electrodes used. © 2007 Elsevier B.V. All rights reserved. Keywords: Electroencephalography; Functional magnetic resonance imaging; MR image artefacts; Field homogeneity

1. Introduction Combined electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) is becoming an increasingly popular tool for the investigation of human brain function. The combination of these techniques has already been shown to have great potential for providing new understanding of the relationship between both spontaneous and evoked electrical activity and the haemodynamic response in the human brain (Goldman et al., 2002; Laufs et al., 2003; Scarff et al., 2004; Debener et al., 2005). The value of this approach in the study of diseases, such as epilepsy, has also become apparent (Lemieux et al., 2001; Lemieux, 2004). The largest challenge in the implementation of simultaneous EEG and fMRI lies in recording ⁎ Corresponding author. E-mail address: [email protected] (R. Bowtell). 0167-8760/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpsycho.2007.06.008

the weak electrical signals from the brain in the presence of the large voltages resulting from the time-varying magnetic field gradients that are needed for MRI, as well as those due to cardiac related movement in the strong static magnetic field of the magnetic resonance (MR) scanner. The focus of most technical work in this area has therefore been on characterisation and elimination of the artefacts in EEG recordings made simultaneously with MRI (Allen et al., 1998, 2000; Anami et al., 2003; Debener et al., 2008-this issue). This work has of course, been crucial in making it possible to record useful EEG data in combined EEG/fMRI experiments. The degradation of MR images due to the presence of the EEG equipment has been less well explored, since it has proved possible to acquire fMRI data with simultaneous EEG recording without making any significant modifications to the MRI techniques or hardware. However, as simultaneous EEG and fMRI is applied across more areas of research, it is important to understand the effect of

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

the combined approach on the data acquired using both modalities. It has previously been shown that the presence of the EEG recording apparatus can have an adverse effect on the quality of MR image data (Krakow et al., 2000; Lazeyras et al., 2001; Scarff et al., 2004; Vasios et al., 2006). Scarff et al. (2004) noted that the signal to noise ratio (SNR) of MR images decreased with increasing numbers of EEG electrodes. Their analysis was carried out on T⁎2 -weighted images acquired using echo planar imaging (EPI), the imaging technique that is most commonly used in fMRI studies, and the SNR measurement was made by comparing the average signal within a brain region to the standard deviation across an area outside the head containing only noise. Some investigative work has also been carried out by Vasios et al. (2006) on the effects of their “InkCap” system on the SNR in echo planar images of the human brain, indicating that this system has very little effect on image quality. The same group has also measured the effect of EEG caps on the quality of retinotopic maps obtained using fMRI (Bonmassar et al., 2001) showing the cap has little effect. The work done by Krakow et al. (2000) was more extensive, involving study of the effects of the different components of the EEG system on images of phantoms and human subjects acquired using EPI. The in-plane spatial extent of the image artefact was used as a measure of the severity of the effect produced by a variety of electrodes, resistors and leads. The largest artefacts were found to be produced by injudiciously chosen resistors. In addition, the artefact due to any combination of materials was found to be “comparable to that of the worst single component” (Krakow et al., 2000). These authors also investigated the artefacts caused by conductive gel and paste/gel mixtures used for conventional EEG recordings. These artefacts were found to be significant, leading to the recommendation that the minimum possible amount of gel should be used. The previous work in this area has concentrated on the characterisation of artefacts due to the EEG system in fMRI data, rather than exploration of the mechanisms underlying the degradation of image quality and has been somewhat qualitative in nature. There are a number of possible sources of degradation of MR images in the presence of the EEG recording apparatus, but the most important effects may be those of (i) the magnetic susceptibility of the cap materials on the homogeneity of the static magnetic field, and, (ii) the electrically conducting elements of the EEG system on the uniformity of the radio frequency (RF) fields involved in excitation and detection of the MR signal. MRI relies on the use of a strong and extremely homogenous magnetic field, B0, which usually varies in strength by less than 1 part per million over the imaging region. Unfortunately, whenever an object is placed inside the magnet its presence disturbs the homogeneity of the field. The magnitude of the field disturbance is of order χB0, where χ is the magnetic susceptibility of the object, although the exact spatial form of the field perturbation depends on the shape of the object and its orientation with respect to the main field. The magnetic susceptibility of water is about − 9 × 10− 6 whilst that of air is approximately zero, with most tissues of the human body

179

having χ-values close to that of water. The materials used in EEG caps are also only weakly magnetic (e.g. for copper, χ = − 9.2 × 10− 6), but unfortunately the high sensitivity of MRI to field perturbations means that field changes of only a few micro-Tesla can cause image artefacts. This is because the frequency of the NMR signal is 42.6 MHz per Tesla of applied field, so that even small changes in field cause significant frequency variation (e.g. 1 μT causes a 42.6 Hz frequency offset). Consequently the presence of the EEG electrodes and leads can introduce artefacts in MR image data due to magnetic field inhomogeneity. These artefacts take two main forms. First, field inhomogeneity leads to image distortion, an effect which becomes important when frequency offsets are similar in size to the frequency separation of pixels in the image. This is a particular problem in EPI since for this sequence the separation of pixels is large (typically 10–50 Hz). Second, if the variation in frequency across the imaging slice is large the signal in gradient echo images will be reduced, leading to regions of signal loss or “drop-out”. This effect is a consequence of the variation of the phase of the NMR signal coming from different positions across the slice and it becomes significant when the range of frequencies across the slice, Δf, is large enough that Δf × TE ∼ 1, where TE is the echo time. It is often the case that signal drop-out is a more significant problem than image distortion in fMRI data. Clearly the extent of signal drop-out and image distortion will depend on the magnetic susceptibilities of the materials used in the electrodes and leads and their spatial arrangement with respect to the subject's head and applied field. Maps showing the spatial variation of field offsets would provide useful information about the dominant sources of artefact. MRI also relies on the use of an RF coil that produces a uniform magnetic field varying at the NMR frequency (e.g. 128 MHz at 3 T) when energised. The strength of the rotating component of this magnetic field which interacts with the nuclear magnetization is usually written as B1. To generate an MR image it is necessary to excite an NMR signal from the sample which is done by applying an RF pulse via the RF coil. The effect of this pulse is to rotate the nuclear magnetization from its equilibrium state of alignment with the applied B0 field, by an angle, θ, that depends on the product of B1 and the duration of the RF pulse. The NMR signal that is used to form an MR image depends on the amount of transverse magnetization generated by the RF pulse, which is proportional to sinθ. Any variation in the strength of B1 over the object being imaged will therefore give rise to unwanted intensity variation in the resulting image. The strength of the signal received from a particular region of the object also depends on the value of B1 in that region (via the principle of reciprocity) giving rise to further intensity variation in the presence of any spatial inhomogeneity of B1. When materials of high electrical conductivity are exposed to RF, large surface current densities which act to screen the RF field from the interior of the material are generated. These currents also disturb the B1-field within regions in close proximity to the conductor. Consequently it can be expected that electrically conducting material in EEG caps will cause some perturbation of the B1-field, leading to

180

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

Fig. 1. B0 field maps (in Hz) acquired from the phantom, shown after removal of large scale field variations at 1.5 T (A–C), 3 T (D–F) and 7 T (G–I) with the 64 electrode cap (left) 32 electrode cap (centre) and no cap (right) on.

artefactual intensity variation in MR images. In addition, the interaction between the RF field and conducting material increases the effective resistance of the RF coil. Since a resistance acts as a source of noise, this effect can lead to a reduction in the SNR of images obtained in combined EEG/ fMRI studies. The aim of the work described in this paper was to characterise the artefacts generated in MR images by the presence of commercially available EEG recording hardware as a function of field strength and number of cap electrodes. Experiments were therefore conducted using 32 and 64 electrode caps on MR systems operating at 1.5, 3 and 7 T to evaluate the level of artefact and the SNR reduction manifested in echo planar images, as are typically used in fMRI experiments, due to the presence of the EEG recording system. Simple analysis of such data does not allow assignment of the source of artefacts specifically to B1 or B0 inhomogeneity. Such an assignment would however be valuable in deciding how to modify EEG cap arrangements so as to reduce the level of artefact. We also therefore carried out B0 and B1 mapping at the three field strengths so as to quantify the field perturbations produced by the 32 and 64 electrode caps.

walled spherical phantom had an 18 cm outer diameter and 17 cm inner diameter. It was filled with saline doped with GdDTPA so as to yield a T1 of approximately 700 ms at 3 T, similar to that of white matter (Lu et al., 2005). Images were acquired in the presence of 32 and 64 electrode MR-compatible EEG lycra caps (EasyCap, Herrsching, Germany) and with no cap in place. Both EEG caps employed sintered Ag/AgCl ring-electrodes arranged in an equidistant montage. While the 32 channel cap covered the upper half of the head sphere, following the international 10–20 system, the 64 channel cap included infracerebral electrode positions, the latter improving EEG source localisation efforts (Hine and Debener, 2007). On the cap, the ECG lead ran from the cable tree in a superior–inferior direction at a posterior location, slightly left of the mid-line, whilst the EOG lead was positioned at an anterior/left position. Electrodes, with a 1.2 cm outer diameter and 0.6 cm inner diameter, including plastic adapters and 5 kΩ resistors, and light-duty, braided copper wires, approximately 0.3 mm thick, were used

2. Methods and analysis

64 electrode cap B0 (Tesla) 95 Standard percentile deviation range (Hz) (Hz)

32 electrode cap

No cap

95 Standard percentile deviation range (Hz) (Hz)

95 Standard percentile deviation range (Hz) (Hz)

1.5 3.0 7.0

33 35 50

16 16 47

Data were acquired from human subjects (5 in total) and a spherical phantom on Philips Achieva MR systems operating at 1.5, 3 and 7 T. The body RF coils were used for signal excitation at 1.5 and 3 T, while eight channel, head RF coils were used for signal reception at these fields. A quadrature birdcage coil was used for signal reception and excitation at 7 T. The plastic-

Table 1 The 95 percentile range and standard deviation of field offset values (in Hz, to 2 significant figures) within the phantom after the removal of low spatial frequency field variations

27 30 57

7.9 9.4 14

9.6 9.7 13

5.5 5.6 12

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

181

Fig. 2. Effects of 32 electrode cap at 3 T on B0 maps (in Hz) (A&B) and flip angle maps (normalised to average flip angle) (C&D). A&C were acquired with the cap on (regions affected are highlighted) and B&D are with no cap.

for connection to BrainAmp MR EEG amplifier(s) (Brainproducts, Munich, Germany) during MR scanning. Abralyte 2000 conductive gel (EasyCap, Herrsching, Germany) was used to apply the electrodes to the subject's scalp or phantom surface. The quantity of gel placed between each electrode and the surface of the phantom was equivalent to the amount typically used when EEG recordings are acquired from human subjects (b1 ml per electrode). A thin layer of conducting gel was spread over the surface of the phantom to mimic the effect of the scalp in human subjects. The ECG and EOG leads were also connected to the “face” region of the phantom. In initial experiments, we did not find significant differences in the effects reported here when the caps were, or were not, connected to the EEG amplifier (switched on or off) and so for the majority of the experiments described here the electrode caps were not connected to the amplifier. Four different imaging procedures were carried out: (i) B0 field perturbations due to magnetic susceptibility effects from the EEG caps were mapped using a three dimensional (3D) spoiled gradient echo sequence (Ericsson et al., 1995). The phase of the signal in images acquired at echo time, TE, using this sequence, is equal to γΔB0TE where γ is the magnetogyric ratio and ΔB0 is the field offset. Maps showing the spatial variation of the signal phase thus carry information about the field offsets in the object that is being imaged. Here, data were acquired from the phantom and a single human subject with 2 mm isotropic resolution, TE = 20 ms and TR = 23 ms at all fields. The resulting phase maps were unwrapped and then scaled by 2πTE to yield maps of the field offset in Hz (1 Hz = 2.35 × 10− 8 T). Field variations occurring on a large length scale (e.g. due to poor shimming) were eliminated

by subtracting a smoothed version of the field map formed from a 2nd order polynomial fit on a plane by plane basis. This emphasized the effects of the EEG electrodes and wires which produce more localised field inhomogeneities that were not eliminated by the fitting procedure. (ii) B1 inhomogeneity was assessed by analysis of 3-D flip angle maps generated using a double-delayed, spoiled gradient echo sequence (Yarnykh, 2007). This uses two equivalent RF pulses per TR period, each of which generates a free induction decay that is phase encoded and then sampled under a read gradient, so as to allow the subsequent formation of two different 3D images. The time between RF pulses in a given TR period is written as TR1, while the time between the 2nd RF pulse in one period and the first RF pulse in the next TR period is written as TR2. In the regime where the longitudinal relaxation time, T1 NN TR1 & TR2 it can be shown that the flip angle, θ, produced by the RF pulse at each location in the image can be simply calculated from the ratio of signal amplitudes in the two images (Yarnykh, 2007). Data were acquired on a single volunteer and on the phantom using this approach with 2 mm isotropic resolution, TE = 3 ms, TR1 = 12 ms and TR2 = 60 ms, and with a nominal flip angle setting of 60°. Each scan took approximately 20 min. Flip angle maps were calculated in Matlab (Mathworks Inc.) to avoid divide by zero problems in calculating the ratio of the signal amplitudes, flip angles were only calculated in regions where the signal in both images was greater than four times the standard deviation in a region of the images lying outside the head or phantom. The resulting flip angle maps were normalised by dividing by the average flip angle calculated over the object of interest thus allowing more straightforward identification of

182

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

Fig. 3. Flip angle maps of the phantom (normalised to average flip angle) with the 64 electrode cap (left), the 32 electrode cap (centre) and no cap (right). A–C show maps acquired at 3 T from similar slices to that shown Fig. 1. The maps shown in the rest of figure are taken from more inferior slices and show the effect of B1 perturbations occurring in proximity to the ECG and EOG wires (arrowed) at 3 T (D–F), 7 T (G–I) and 1.5 T (J–L). A more inferior slice is shown for the 64-electrode cap compared with the 32-electrode cap as the path of the EOG and ECG wires was different on these caps. The thresholding process described in the Methods and analysis section leads to the generation of significantly sized areas in the 7 T maps where the flip angle can not be characterised.

any regions significantly affected by B1 inhomogeneity. The standard deviation of the normalised flip angle over the phantom was also calculated at each field strength. (iii) A standard multi-slice EPI sequence was implemented with a 64 × 64 matrix, TR = 4.2 s, 85° flip angle and isotropic voxel size (phantom: 3 mm; head: 3.3 mm) with no slice gap using echo times that are conventionally employed for fMRI experiments at each field strength (1.5 T: 60 ms; 3 T: 40 ms: 7 T: 25 ms). The number of trans-axial slices was chosen to allow for full head or phantom coverage (45 and 52 slices respectively) and the pixel bandwidth in the phase - encode direction was about 28 Hz at all fields. Data were acquired on 5 human subjects (2 female, 3 male with a mean age of 29) as well as on the phantom, with fifty volumes acquired for each arrangement (32, and 64 electrode caps and no cap). After motion correction, using MCFLIRT within FSL, registering to the central volume

acquired, and high-pass filtering, using a square filter with a cut off at 0.033 Hz to remove scanner drift, the SNR of the data was assessed by pixel-wise calculation of the ratio of the temporal standard deviation to mean. This ratio was then averaged over an appropriately defined region of interest (ROI) giving a measure of the temporal SNR within that region. ROI's of approximately 800 voxels in size (reduced to 400 voxels at 7 T to avoid regions of significant signal drop-out) were positioned Table 2 Standard deviation of relative flip angle within the phantom B0 (Tesla)

64 electrode cap

32 electrode cap

No cap

1.5 3.0 7.0

0.05 0.12 0.44

0.08 0.16 0.43

0.02 0.08 0.34

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

183

Fig. 4. Effect of 32 electrode cap at 3 T on 3D surface rendered MPRAGE images. A&B: Images with 32 electrode cap, highlighted regions are the paths of the EOG (A) and ECG (B) wires. C&D: The same images with no cap in place.

Fig. 5. EPI data of a similar slice of the phantom at 1.5 T (A–C), 3 T (D–F) and 7 T (G–I) with 64 electrode cap (left), 32 electrode cap (centre) and no cap on (right).

184

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

Fig. 6. Representative slice of EPI data for a single subject at 1.5 T (A–C), 3 T (D–F) and 7 T (G–I) with 64 electrode cap (left), 32 electrode cap (centre) and no cap (right) on.

in white matter (WM) in locations showing minimal signal loss due to either B1 or B0 inhomogeneity, thus providing a measure of reduction in the global image SNR due to loading of the RF coil by the conducting material of the EEG electrodes and leads. The extent of signal loss due to field inhomogeneity from the caps was also measured in the EPI data acquired from the phantom. This was done by counting the number of pixels within masked regions whose intensity fell below 5 or 10% of the mean signal within the region. Two different masked regions were employed: the first covered the entire spherical phantom, while the second covered the whole phantom except for a 1.1 cm thick layer at its surface. This second region was chosen to provide an indication of signal drop-out effects in the brain due to EEG electrodes and wires positioned on the scalp, and consequently separated from brain tissue by about 1.1 cm on average (Menon et al., 1997). (iv) The effect of the EEG recording apparatus on anatomical images, acquired using a standard MPRAGE sequence (Mugler and Brookeman, 1990) with a SENSE factor 2, was also assessed. These images had 1.1 mm isotropic resolution and 282 mm FOV in all directions. The slices covered the whole head or phantom. The resulting images were 3D surface rendered to show the effects of the wires and electrodes on the surface of the image.

All imaging procedures were carried out on the phantom for both caps at the three field strengths. EPI and MPRAGE data were acquired from the head at all field strengths for both caps, Table 3 Measurements of signal drop-out in echo planar images of the phantom Masked B0 (Tesla) region

1.5

3.0

7.0

64 electrode cap 32 electrode cap

No cap

Threshold 5%

10%

5%

10%

5%

10%

Whole Phantom Inner Region Whole Phantom Inner Region Whole Phantom Inner Region

0.010

0.027

0.010

0.023

0.012

0.028

0

0.00010 0.00031 0.0014

0.000080 0.00036

0.051

0.078

0.012

0.015

0.029

0.021

0.0024 0.0059

0.00016 0.00085 0.0010

0.0025

0.10

0.16

0.11

0.17

0.027

0.044

0.086

0.14

0.11

0.18

0.014

0.023

The table shows the ratio of the number of pixels lost relative to the total number within the phantom when the images are thresholded to exclude voxels with less than 5% or 10% of the mean pixel intensity. The masked region spanned either the whole phantom or an inner region excluding a 1.1 cm thick spherical shell at the periphery of the phantom.

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188 Table 4 Average ratio of temporal standard:mean in a WM ROI in EPI data B0 No cap (Tesla) 1.5 3.0 7.0

32 electrode cap

% 64 electrode Change cap

0.0166 ± 0.0007 0.0214 ± 0.0005 27 ± 8 0.0097 ± 0.0001 0.0102 ± 0.0004 4 ± 4 0.0076 ± 0.0003 0.0089 ± 0.0002 18 ± 4

% Change

0.0225 ± 0.0004 26 ± 10 0.0106 ± 0.0002 15 ± 5 0.0097 ± 0.0007 28 ± 13

The percentage change indicates the reduction in SNR due to each of the caps. Values of the ratio and percentage change were calculated by averaging measurements over the 5 subjects with the relative error of these values then found.

but B0 and flip angle mapping was not carried out at any field strength using the 64 electrode cap on the human head due to concerns over potential RF heating. For the same reason, mapping was not carried out at 7 T on the human head with the 32 electrode cap in place. 3. Results Images and maps taken from representative slices within the phantom and head have been identified and are displayed in the figures. All relevant data values were used when calculating the quantities detailed in the tables shown in this section. 3.1. B0-maps Fig. 1 shows field maps from similar axial slices which were acquired from the phantom at 1.5, 3 and 7 T in the presence of the 64 and 32 electrode caps and with no cap in place. Localised field inhomogeneity resulting from individual electrodes, which increases in severity with field strength, is evident in these maps. This behaviour is summarised in Table 1 which details the standard deviation of the field offset in Hz and the range of fields within which 95% of pixel values in the phantom lie for the different arrangements. An increase in the field offsets due to the presence of the cap was not as clear in data acquired from the human subject at 3 T (Fig. 2A&B) and it can be seen that susceptibility variations across the tissues of the head are the dominant source of field perturbation in these maps. Even in the phantom, other sources of field inhomogeneity such as air bubbles and the plastic screw used for sealing the sphere cause significant residual field inhomogeneity in the absence of the EEG caps (Fig 1C,F & I).

185

spatial variation of the flip angle. Fig. 3D–L show the flip angle variation in a more inferior axial slice cutting through the path of the EOG and ECG leads. These images revealed a greater cap-induced flip angle inhomogeneity, which became much worse at 7 T. A similar pattern of spatial variation of the flip angle also occurred in the data from the human subject, where a significant reduction in flip angle was seen in occipital and frontal areas (Fig. 2C&D) which underlie the ECG and EOG leads. The corresponding B0 maps (Fig. 2 A&B) did not show large B0 field perturbations in these areas, indicating that any signal reduction found close to these leads is unlikely to be due to B0 field inhomogeneity. 3.3. MPRAGE data The paths of the EOG and ECG leads, as well as individual electrode positions, could be easily identified in the 3 T MPRAGE images of the head, shown in Fig. 4. Signal reduction close to these leads causes artefactual indentations at the surface of the volume rendered images shown in Fig. 4A and B. 3.4. EPI data Fig. 5 shows images of a similarly positioned single slice from EPI data sets acquired from the phantom at 1.5, 3 and 7 T with the 32 and 64 electrode caps in place and with no cap present. It is evident that the ring electrodes caused small regions of signal drop-out at the periphery of the phantom, resulting from the localised field inhomogeneities that can be seen in the field maps shown in Fig. 1. These effects are less evident in Fig. 6, showing echo planar images collected on a single subject at the three different field strengths with and without the caps in place. The dominant effect of the EEG caps in the 7 T data is an exacerbation of the large scale intensity variation, resulting from increased B1 inhomogeneity. Although this effect is less evident in the data acquired at lower field, there was a slightly reduced signal intensity in left occipital lobe at 3 T (arrowed in Fig. 6D&E), resulting from the B1 perturbation due to the ECG lead. Table 3 characterises the signal dropout in the EPI data from the phantom which results from both B1 and B0 inhomogeneity for the different caps and field strengths. Table 4 details the ratio of the temporal standard deviation to average signal averaged over the white matter ROI for each arrangement, along with the percentage change in this ratio due to the presence of the 32 or 64 electrode caps.

3.2. B1-maps 4. Discussion Fig. 3 shows normalised flip angle maps acquired from the phantom at the three different field strengths in the presence of the 32 and 64 electrode caps, and with no cap in place. Fig. 3A– C show maps acquired at 3 T from similar axial slices to those shown in Fig. 1. Comparison of Fig. 3A (64 electrode cap) and B (32 electrode cap) with Fig. 3C (no cap) indicates that the individual electrodes did not generate significant B1 inhomogeneity. However, evaluation of the standard deviation of the normalised flip angle in the phantom (Table 2), indicated that the presence of the caps caused a significant increase in the

The B0 field maps acquired from the phantom (Fig. 1) show that each electrode generates significant local perturbation of the magnetic field, the exact form of which depends on the orientation of the electrode with respect to the field. Similar effects were produced by individual electrodes in the two caps, but a greater proportion of the periphery was affected by the 64 electrode cap because of the larger number of electrodes. The data reported in Table 1 shows that the presence of the EEG cap increases the width of the distribution of field values within the phantom and

186

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

that the width is increased at higher field strength. Although, it might be expected that the values in Table 1 would scale linearly with field strength, the fact that they increase less rapidly with field probably reflects the higher performance of the shimming systems used in the 3 and 7 T scanners, which limits to some extent the increase in field inhomogeneity at the higher fields. The distribution widths are not strongly affected by the number of electrodes. This is probably because the greater overlap of the fields from the different electrodes in the case of the 64 electrode cap slightly reduces the depth into the phantom that the field inhomogeneities extend, counterbalancing the increased surface area of the phantom over which inhomogeneities are produced. Such behaviour results from the dipolar nature of the field perturbation from each individual electrode, producing regions of both increased and reduced field, whose overlap in the case of multiple electrodes leads to a reduction of the depth to which the field perturbation extends. The local gradients of the field that are evident in Fig. 1 generate localised areas of signal drop-out and distortion at the surface of the phantom in the echo planar images shown in Fig. 5, which are most severe in the images acquired at high field. The electrodes do not however cause significant dropout in the echo planar images of the brain shown in Fig. 6. This is a consequence of the localised nature of the field inhomogeneities which consequently do not extend from the site of electrodes at the scalp into the brain. The field maps obtained from the human subject at 3 T (Fig. 2A&B) bear out this observation, as they display little noticeable effects of the electrodes relative to the larger field inhomogeneities generated by the head itself (particularly the sinuses). The effects of B1 inhomogeneity pose more significant problems for image data quality in the case of the EEG caps studied here. Table 2 shows that the introduction of either EEG cap causes a noteworthy increase in the standard deviation of the relative flip angle within the phantom. The flip angle maps shown in Fig. 3 indicate the dominant source of B1 inhomogeneity is not the electrodes, but rather the EOG and ECG leads. This is also evident from the MPRAGE data acquired from the human subject (Fig. 4) where the presence of these leads causes an artefactual indentation of the scalp in the volume-rendered data. The exact reason that these leads cause a much stronger local perturbation of the B1 field than the other leads need further investigation. However any explanation is likely to relate to the main distinguishing feature of these leads compared with those attached to other electrodes, which is their greater length: the ECG and EOG leads used here are 58 and 20 cm longer respectively than the Tp9 lead. The fact that there are significant effects at all three field strengths means that the interaction is unlikely to be a resonant effect where the lead length matches an integral number of half wavelengths of the RF, since this wavelength changes by a factor of 4.67 between 7 and 1.5 T. At 7 T where RF must be applied at 300 MHz it is evident from Fig. 3I that the spherical phantom itself causes significant flip angle inhomogeneity. This is a well known phenomenon at high-field, which results from the increased effects of electrical conductivity and relative permittivity on the applied electromagnetic fields when the RF wavelength approaches the size of the object being imaged (Adriany et al., 2005). Addition of the

conducting material of the cap, and particularly the EOG and ECG leads then causes a further significant degradation of the B1 homogeneity which is clearly seen in Fig. 3G&H. The flip angle maps obtained from the human subject at 3 T (Fig. 2C&D) show that the effects of the EOG and ECG lead on B1 homogeneity are also manifested in the human head, leading to areas of flip angle reduction in occipital and frontal areas. The effect of the ECG lead is seen as an area of reduced signal intensity in the EPI data obtained at 3 T (Fig. 6D&E), an effect which we have noted in a number of other previous studies. The effects of much greater B1 inhomogeneity are evident in the EPI data acquired from the phantom (Fig. 5G–I) and the human head (Fig. 6G–I) at 7 T. In the latter case, introduction of the EEG caps leads to a further increase of the signal intensity at the centre of the head relative to that at the periphery. These effects lead to spatial variation in the sensitivity of detection of activation in fMRI data analysis. In particular a larger fractional BOLD signal change is needed in areas of signal drop-out for identification of statistically significant signal changes. Table 3, which characterises the signal drop-out that occurred in the echo planar images of the spherical phantom (Fig. 5), shows that at 1.5 and 3 T the majority of the signal loss occurs at the periphery of the phantom so that the fraction of pixels effected is much smaller when just the inner region rather than the whole phantom is considered. Signal drop-out at these fields mainly reflects the effect of B0 inhomogeneity, which as discussed above is confined to regions close to the electrodes, and increases with field strength. Table 3 also shows that at 7 T, the signal drop-out caused by the introduction of the caps is more severe and is not limited to regions at the surface of the phantom, since the fraction of voxels lost in the whole phantom is similar to that found when considering just the inner region. This is a consequence of the dominance of the effect of EEGcap-induced B1 inhomogeneity at the higher field. Table 4, which details the ratio of temporal standard deviation to average signal intensity in a white matter ROI, shows that the SNR increases with field strength with or without the EEG cap in place. The increase in SNR is greater between 1.5 T and 3 T compared with 3 T to 7 T, most likely because the relatively large voxel sizes used here mean that at 7 T the increased importance of physiological noise (Kruger and Glover, 2001) moderates the gain in intrinsic SNR. Table 4 also shows that the conducting material in the caps reduces the SNR, with the 64 electrode cap having a significantly larger effect than the 32 electrode cap at 3 and 7 T. The change in SNR due to the presence of the conducting materials of the cap appears to be lowest at 3 T. Assuming that the NMR signal strength is unaffected by the presence of the cap the change in the SNR depends largely on the ratio of the magnitude of the extra noise due to the cap's loading of the RF coil to the magnitude of other noise sources. Both would be expected to increase with field strength, but it appears that between 1.5 and 3 T the latter increases more rapidly, whilst between 3 and 7 T the increase in cap-induced noise is greater. In the case of the data acquired at 7 T, it is possible that introduction of the cap has also somewhat reduced the signal strength due to the direct effects of B1 inhomogeneity or indirect effects in missetting of the flip angle.

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

It is worth noting that the reduction in SNR due to the EEG caps was relatively small even in the worst case (64 electrode cap at 7 T) and that the use of a WM region, in which the physiological noise is lower than in grey matter, along with the relatively long TR used for this experiment to allow for whole head coverage and the large voxel size, gives these data a high base SNR, leading to a high sensitivity to cap-induced changes. The results described here are similar to those previously reported by Vasios et al. (2006) who described a 12% reduction in WM SNR due to the their 32-channel ink-cap system, and Scarff et al. (2004) who showed a reduction of about 30% in SNR in WM at 3 T when comparing MR data acquired with 64 and 256 electrodes in place, using a different EEG cap system. 5. Conclusions Using B0- and B1-mapping techniques we have measured the static and RF magnetic field inhomogeneities generated by commercially available 32 and 64 electrode EEG caps at three different field strengths. The resulting field maps have been used to explain the artefacts produced by the caps in echo planar images of the kind which would normally be used in combined fMRI/EEG experiments. The results indicate that the electrodes are the dominant source of B0 field inhomogeneity, but the localised nature of the field perturbation that they produce means that the effect on the signal intensity from the brain is not significant relative to the field inhomogeneity produced by the head itself. In the particular EEG caps investigated here, RF inhomogeneity linked to the longer EEG and EOG leads does have a weak effect on the signal intensity in the echo planar images of the brain at 3 T and this effect merits further investigation. Use of higher resistance material (e.g. carbon or ink rather than copper) in lead fabrication might be expected to reduce the interaction with the RF field. At 7 T the introduction of either a 32 or 64 electrode caps causes a significant deterioration of the B1 homogeneity, which is already relatively poor even in the absence of the EEG system. This effect also appears to be linked to the longer EEG leads, as the B1 homogeneity is less affected in superior regions. It is worth noting that techniques for the amelioration of the effects of B1 inhomogeneity at high field are actively being developed (Adriany et al., 2005; Katscher and Bornert 2006) and could also be employed to overcome the effects of EEG caps on the B1 field. Measurements of the standard deviation of white matter signal in EPI time series indicate that the introduction of the EEG cap causes a small reduction in the image SNR, which increases with the number of electrodes in agreement with previous findings (Scarff et al., 2004). Despite the increased effect of B0 and B1 inhomogeneities, the SNR in the EPI data obtained at 7 T from the human brain is still greater than that obtained with no cap in place at 3 T, indicating that attempts to implement simultaneous EEG and fMRI recording at 7 T are very worthwhile (Vasios et al., 2006; Mullinger et al., 2007). Acknowledgments The authors would like to thank Andreas Bungert (SPMMRC) and David Foxall (Philips Medical Systems) for

187

their help with this work. This work is supported by an MRC grant (G9900259) and a PhD studentship for KJM provided by EPSRC and Philips Medical Systems. References Adriany, G., Van de Moortele, P.F., Wiesinger, F., Moeller, S., Strupp, J.P., Andersen, P., Snyder, C., Zhang, X.L., Chen, W., Pruessmann, K.P., Boesiger, P., Vaughan, T., Ugurbil, K., 2005. Transmit and receive transmission line arrays for 7 tesla parallel imaging. Magnetic Resonance in Medicine 53 (2), 434–445. Allen, P.J., Poizzi, 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 (3), 229–239. Allen, P.J., Josephs, O., Turner, R., 2000. A method for removing imaging artifact from continuous EEG recorded during functional MRI. NeuroImage 12 (2), 230–239. Anami, K., Mori, T., Tanaka, F., Kawagoe, Y., Okamoto, J., Yarita, M., Ohnishi, T., Yumoto, M., Matsuda, H., Saitoh, O., 2003. Stepping stone sampling for retrieving artifact-free electoencephalogram during functional magnetic resonance imaging. NeuroImage 19 (2), 281–295 (Part 1). Bonmassar, G., Hadjikhani, N., Ives, J.R., Hinton, D., Belliveau, J.W., 2001. Influence of EEG electrodes on the BOLD fMRI signal. Human Brain Mapping 14, 108–115. Debener, S., Ullsperger, M., Siegel, M., Fiehler, K., Yves 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. Journal of Neuroscience 25 (50), 11730–11737. Debener, S., Mullinger, K.J., Niazy, R.K., Bowtell, R.W., 2008. Spatial and temporal dynamics of the ballistocardiogram artefact as revealed by EEG recordings at 1.5, 3 and 7 tesla static magnetic field strength. International Journal of Psychophysiology 67, 189–199 (this issue). doi:10.1016/j. ijpsycho.2007.05.015. Ericsson, A., Weis, J., Hemmingsson, A., Wikstrom, M., Sperber, G.O., 1995. Measurements of magnetic field variations in the human brain using a 3D-ft multiple gradient echo technique. Magnetic Resonance in Medicine 33, 171–177. Goldman, R.I., Stern, J.M., Engel, J., Cohen, M.S., 2002. Simultaneous EEG and fMRI of the alpha rhythm. NeuroReport 13 (18), 2487–2492. Hine, J., Debener, S., 2007. Late auditory evoked potentials asymmetry revisited. Clinical Neurophysiology 118, 1274–1285. Katscher, U., Bornert, P., 2006. Parallel RF transmission in MRI. NMR in Biomedicine 15, 152–157. Krakow, K., Allen, P.J., Symms, M.R., Lemieux, L., Josephs, O., Fish, D.R., 2000. EEG recording during fMRI experiments: image quality. Human Brain Mapping 10, 10–15. Kruger, G., Glover, G., 2001. Physiological noise in oxygenation-sensitive magnetic resonance imaging. Magnetic Resonance in Medicine 46 (4), 631–637. Laufs, H., Krakow, K., Sterzer, P., Eger, E., Beyerle, A., Salek-Haddadi, A., Kleinschmidt, A., 2003. Electroencephalographic signatures of attentional and cognitive default modes in spontaneous brain activity fluctuations at rest. Proceedings of the National Academy of Sciences of the United States of America 100 (19), 11053–11058. Lazeyras, F., Zimine, I., Blanke, O., Perrig, S.H., Seeck, M., 2001. Functional MRI with simultaneous EEG recording: feasibility and application to motor and visual activation. Journal of Magnetic Resonance Imaging 13, 943–948. Lemieux, L., 2004. Electroencephalography-correlated functional MR imaging studies of epileptic activity. Neuroimaging Clinics of North America 14 (3), 487. Lemieux, L., Salek-Haddadi, A., Josephs, O., Allen, P.J., Toms, N., Scott, C., Krakow, K., Turner, R., Fish, D.R., 2001. Event-related fMRI with simultaneous and continuous EEG: description of the method and initial case report. NeuroImage 14, 780–787. Lu, H., Nagae_Poetscher, L.M., Golay, X., Lin, D., Pomper, M., van Zijl, P.C.M., 2005. Routine clinical brain MRI sequences for use at 3.0 Tesla. Journal of Magnetic Resonance Imaging 22, 13–22.

188

K. Mullinger et al. / International Journal of Psychophysiology 67 (2008) 178–188

Menon, V., Ford, J., Lim, K., Glover, G., Pfefferbaum, A., 1997. Combined event-related fMRI and EEG evidence for temporal–parietal cortex activation during target detection. NeuroReport 8 (14), 3029–3037. Mugler, J.P., Brookemann, J.R., 1990. 3-Dimensional magnetization prepared rapid gradient echo (MP-RAGE) magnetic resonance imaging. Magnetic Resonance in Medicine 15, 152–157. Mullinger, K.J., Brookes, M.J., Stevenson, C.M., Clemence, M., Morgan, P.S., Nockowski, C.P., Profusz, S., Bowtell, R.W., 2007. Exploring the feasibility of simultaneous EEG/fMRI at 7 T. Proc. ISMRM 2007.

Scarff, C.J., Reynolds, A., Goodyear, B.G., Ponton, C.W., Dort, J.C., Eggermont, J.J., 2004. Simultaneous 3-T fMRI and high-density recording of human auditory evoked potentials. NeuroImage 23, 1129–1142. Vasios, C.E., Angelone, L.M., Purdon, P.L., Ahveninen, J., Belliveau, J.W., Bonmassar, G., 2006. EEG/(f)MRI measuremtnets at 7 Tesla using a new EEG cap (“InkCap”). NeuroImage 33 (4), 1082–1092. Yarnykh, V.L., 2007. Actual flip-angle imaging in the pulsed steady-state: a method for rapid three-dimensional mapping of the transmitted radiofrequency field. Magnetic Resonance in Medicine 57, 192–200.