Safety evaluation of a new setup for transcranial electric stimulation during magnetic resonance imaging

BACKGROUND
Transcranial electric stimulation during MR imaging can introduce safety issues due to coupling of the RF field with the stimulation electrodes and leads.


OBJECTIVE
To optimize the stimulation setup for MR current density imaging (MRCDI) and increase maximum stimulation current, a new low-conductivity (σ = 29.4 S/m) lead wire is designed and tested.


METHOD
The antenna effect was simulated to investigate the effect of lead conductivity. Subsequently, specific absorption rate (SAR) simulations for realistic lead configurations with low-conductivity leads and two electrode types were performed at 128 MHz and 298 MHz being the Larmor frequencies of protons at 3T and 7T. Temperature measurements were performed during MRI using high power deposition sequences to ensure that the electrodes comply with MRI temperature regulations.


RESULTS
The antenna effect was found for copper leads at ¼ RF wavelength and could be reliably eliminated using low-conductivity leads. Realistic lead configurations increased the head SAR and the local head SAR at the electrodes only minimally. The highest temperatures were measured on the rings of center-surround electrodes, while circular electrodes showed little heating. No temperature increase above the safety limit of 39 °C was observed.


CONCLUSION
Coupling to the RF field can be reliably prevented by low-conductivity leads, enabling cable paths optimal for MRCDI. Compared to commercial copper leads with safety resistors, the low-conductivity leads had lower total impedance, enabling the application of higher currents without changing stimulator design. Attention must be paid to electrode pads.


Introduction
Roughly two decades ago, Nitsche and Paulus showed that human motor cortex excitability could be non-invasively modulated by weak electric currents applied through the intact skull by surface electrodes [1]. Since then, the use of transcranial electric stimulation (TES) techniques in neuroscience applications has grown tremendously. There is also increasing interest to apply TES inside magnetic resonance imaging (MRI) scanners. This is motivated by the wish to use functional MRI (fMRI) for characterizing the physiological stimulation effects. More recently, MRI is also applied to shed light on the physical current flow inside the brain. Simulation of the current flow using forward models of the head anatomy [2,3] is feasible, but the accuracies and reliabilities of the results are challenged by a number of factors. For example, the ohmic conductivities of the head tissues at low frequencies are quite uncertain, highlighting the need to validate the simulated fields [4].
MR current density imaging (MRCDI) [5] and MR electrical impedance tomography (MREIT) [6] are two emerging modalities that can indirectly measure the current flow in the brain and the conductivity of the tissue, respectively. These techniques have the potential to improve the accuracy of electric field simulations for TES, as well as for source localization in electro-and magnetoencephalography (EEG and MEG) [7], and can aid in the characterization of pathological tissue [8]. Similar to TES, MRCDI and MREIT use weak currents applied via surface electrodes. The current-induced changes of the static magnetic field are measured and used to determine the current flow or tissue conductivities at low frequency.
The current flow inside the brain changes the magnetic field only slightly, resulting in a low signal-to-noise ratio (SNR) of the measurements and making them prone to artifacts. The effect of the stray fields from the cable currents has previously been studied [9]. Unless the leads are aligned fully parallel to the main magnetic field, the induced stray fields will strongly influence the currentinduced magnetic field changes measured in the brain, and therefore the current density and conductivity reconstruction. This can be corrected for by tracking the cables in MR images and using the Biot-Savart law to subtract induced stray fields [10]. Although the correction method significantly improves the current density reconstruction results, it would be preferable to orient the leads parallel to the main magnetic field to reduce residual errors and increase the robustness of the measurement approach.
For TES, MRCDI and MREIT, the currents are applied through lead wires connected to the subject's scalp via surface electrodes. Extra safety measures have to be taken when conductive materials are used in an MR scanner, especially when in contact with tissue. Many incidences of patient burns caused by coupling between the RF field and lead wires have been reported [11]. However, no burn incidents have been reported for TES-MRI. Heating of leads can be caused by direct electromagnetic induction in wire loops [12] and highly conductive loops must be avoided during MRI. Another important origin of heating is the antenna effect that occurs when wires or other conductors of appropriate length act as "receive antennas" for the RF field. For increasing field strengths, the antenna effect becomes an increasing problem due to the shorter wavelength of the RF field. Half a wavelength is typically found to be the critical length for heating [13,14], but it has been shown experimentally that the length of the lead required to observe the effect also depends on the boundary conditions on each end of the lead [15]. For high impedance at one end (open or connected to a safety resistor or high impedance amplifier) and relatively low impedance at the other end (connected to tissue) ¼ RF wavelength can be critical as well. Therefore, the design of lead wires that reliably prevent the occurrence of electromagnetic induction and antenna effects despite varying boundary conditions is important, but challenging.
Conventional TES devices use highly conductive leads (usually copper) between the stimulator and surface electrodes. For the TES device that is most commonly used in combination with MRI (DC-STIMULATOR MR, neuroCare Group GmbH, München, Germany), most of the cable is realized as twisted pair cable and 5 kU safety resistors are added to each of the two leads to limit the length of highly conductive material near the scanned subject (Fig. 1a). This design improves safety but prevents an optimal cable orientation for MRCDI and MREIT experiments. Because the device supplies a maximal output voltage of 30 V, the safety resistors also limit the maximum possible stimulation current to around 2 mA. Most TES studies so far have used currents up to 2 mA, but there is recent interest to explore higher current strength of up to 4 mA to increase efficacy [16e20]. The safety resistors limit the use of TES-fMRI studies to characterize the physiological effect of higher TES currents.
The aim of this work is to redesign the leads for combined MR and current injection experiments to remove the above restrictions. Specifically, the goal was to develop leads that would allow long straight wire paths parallel to the static magnetic fields and support stimulation currents up to 4 mA with the existing stimulator while not compromising safety. Instead of using highly conductive materials and local safety resistors, we propose to use a distributed resistance by having leads with much lower conductivity, while also having an overall lower total impedance. Carbon fiber leads are routinely used for EEG-MRI and are reported to decrease specific absorption rate (SAR) compared to copper leads [21]. Here, we extend this approach to TES and further minimize the risk for the occurrence of antenna effects by using an even less conductive silicone rubber material (s ¼ 29.4 S/m) for the lead wires. Numerical methods have previously been used to estimate SAR for combined EEG-MRI studies [21e23], as well as for TES-MRI experiments [24,25].
In this study, we use both numerical simulations to estimate SAR as well as experimental temperature measurement. We first simulate a worst-case antenna effect at 298 MHz to investigate the relationship between the antenna effect and conductivity. Secondly, we simulate two electrode types with various lead configurations to ensure safety at both 128 MHz and 298 MHz, corresponding to the proton Larmor frequencies at 3T and 7T magnetic field strength. Lastly, temperature measurements are performed on the electrodes and leads made in-house during in vivo MRI at both field strengths.

Electrode and lead design
We constructed two commonly used TES electrode types inhouse: 1) The circular electrode commonly used for non-focal stimulation in TES or for MRCDI and MREIT (Fig. 1b) and 2) the center-surround electrode used for focal stimulation in TES experiments [26] (Fig. 1c). Both types are 3 mm thick. The circular electrodes are 5 cm in diameter. The center-surround electrodes have an outer ring with an inner and outer diameter of 10 cm and 8 cm, respectively. The diameter of the center electrode is 3 cm. For all electrodes, a 90 cm silicone rubber strip with a cross-sectional area of 10 mm 2 was cut out and used as the lead wire. Both electrodes and leads are made from silicone rubber (ELASTOSIL® R 570/60 RUSS, Wacker, Munich, Germany). The resistance of each of the lead wires is 2 kU ± 200 U. To ensure proper electrical connection and mechanical strength, the rubber leads are sewed on to the electrodes. The other ends of the leads are connected to copper leads with cable crimps. Medical grade touch-proof safety connectors are connected to the copper leads (MS1525-B, St€ aubli, Pf€ affikon, Switzerland). A glass-fiber braided sleeving (GSS6, HellermannTyton, Crawley, Germany) is used for thermal and electrical insulation. The glass-fiber sleeving is also sewed on to the electrode and connected to the copper wire to relieve the silicone rubber lead of any strain. Ten20 conductive EEG paste (D.O. Weaver and Co., Aurora, CO, USA) is used between the electrodes and abraded skin to ensure proper connection.

Simulations
Finite-difference time-domain (FDTD) simulations were performed in Sim4Life (ZMT, Zurich, Switzerland) to obtain specific absorption rate (SAR) results. Simulations were performed with 128 MHz and 298 MHz harmonic excitations. All simulations ran until convergence at À30 dB, tested for steady-state on the lumped elements and sources on the RF coils.

Phantom
The heterogeneous male body model Duke from the IT'IS foundation was used in the simulations [27]. The head was positioned at the centers of the birdcage coils in all simulations. 2 mm isotropic resolution was used for Duke's head and shoulders and 4 mm for the torso. The rest of the body was segmented according to the automatic gridding produced by Sim4Life. This was done to reduce simulation time while still allowing sufficient current flow to obtain accurate simulation results [28].

RF coils
For 128 MHz simulations, a generic body coil was used (Fig. 3). Although the proton Larmor frequency of the scanner used in the experiments is 123 MHz the small difference in frequency will have minimal influence on the results. The 298 MHz coil (Fig. 2a) is a model of a transmit head coil [29] (7T volume T/R, Nova Medical, Wilmington, MA). Both coils are 16 rung high-pass birdcage coils. The dimensions are given in Table 1. The coils have two input ports on the superior end-ring 90 apart shifted 45 relative to the body model. The coils were iteratively tuned to the respective frequencies loaded with Duke with the head placed in the centers of the coils. The coils were driven in quadrature mode with equal input power on both ports. For coil model validation, see S2 in supplementary materials.

SAR evaluation
SAR is a measure of the RF power absorbed by the tissue and is given by (1) where s is the tissue conductivity, r is the density of tissue end E ! is the peak electric field inside the tissue. According to international guidelines IEC 60601-2-33 [30], SAR is limited during MRI to avoid excessive heating of a subject due to absorbed RF power. The two relevant limitations for head MRI are head SAR (SAR averaged over the mass of the head) and local head SAR given as the peak spatial average SAR over 1 g or 10 g of tissue.
To evaluate the influence that the electrodes and leads have on SAR, we compare head SAR and 1 g local head SAR for a reference simulation to simulations that include electrodes. The head SAR and 1 g local head SAR ratios are expressed as SAR is compared for 1 W radiated power as well as for a calibrated B 1 field. The input power P for each simulation is scaled such that the average amplitude of B 1 in the center slice of the coil is the same for all simulations. This is done the following way: P ref is the input power for the reference simulation, set to 1 W, B 1,ref is the average B 1 amplitude for the center slice of the reference simulations, and B 1 is the average B 1 amplitude for the corresponding simulation before normalization. The electrodes, leads, and gel are excluded when R m and R 1g are calculated to include only tissue SAR when averaging is performed. By excluding the electrodes from R m and R 1g calculations, these values only express the changes in SAR on the tissue caused by adding electrodes and not the power loss in the electrodes and the gel. Temperature measurements are used to ensure that the heating caused by the power loss in the electrodes and gel is within regulation limits [30].

Antenna effect simulations
To examine the relationship between the antenna effect and the conductivity of the lead wires, a worst-case simulation was performed with the lead wires parallel to the z-direction with one end connected to the circular electrodes and one end in free space as seen in Fig. 2a. This simulation was only performed at 298 MHz as the antenna effect becomes an increasing problem for higher frequencies. The simulations were performed with varying lead lengths from 0 to 100 cm with 10 cm increments including 25 cm and 75 cm as they are approximately ¼ and ¾ wavelength in air at the proton Larmor frequency 298 MHz. Two conductivities were used for all the incremental lengths of the lead wires, namely 5.8$10 7 S/m for copper and 29.4 S/m for silicone rubber with constant cross-sectional area. For 25 cm (the worst-case length), multiple conductivities were simulated with logarithmic increments from 10 2 S/m to 10 7 S/m. The average power dissipation on the electrodes is used as a measure of the severity of the antenna effect.

Realistic lead configuration simulations
For the realistic lead configurations, simulations were performed for the center-surround and circular electrodes at 128 MHz and 298 MHz, respectively. Four lead configurations were simulated as seen in Fig. 3. A right-left and an anterior-posterior montage were simulated for the circular electrodes ( Fig. 3a and  b). The center-surround electrodes were only simulated for a rightleft montage (Fig. 3c). The lead configurations in Fig. 3aec show the intended use cases. In addition, the lead configuration in Fig. 3d was also simulated to ensure that misplacing the leads will not have critical consequences due to higher E-field close to the RF coil. All three electrode montages were also simulated without leads for both magnetic field strengths. The electrodes and gel are shown in Fig. 2b and c. The add-on subgrid feature using the Acceleware GPU solver (Acceleware, Calgary, Canada) in Sim4Life was used to obtain a fine resolution for electrode, gel and leads while keeping the same grid size in the rest of the simulation space. An isotropic grid size of 0.5 mm was used for electrodes, gel and leads. With this resolution, the smallest structure in any direction is minimum 4 times the grid size.
The conductivity s and the relative permittivity ε r for both the The leads were terminated with an equivalent resistor representing the output impedance of the combined copper cable, filter, and stimulator (DC-STIMULATOR MR, neuroCare Group GmbH, München, Germany). The output impedances at the relevant simulation frequencies were found with a vector network analyzer (VNA). Since the leads are 90 cm long, the equivalent resistors are far outside the effective exposure volume of the coils as seen in Fig. 3.

Experimental setup
Experiments were performed on 3T (MAGNETOM Prisma; Siemens Healthcare, Erlangen, Germany) and 7T (Achieva; Philips Healthcare, Best, The Netherlands) whole-body MRI scanners. Two senior researchers involved in the project were scanned in the experiments. Informed consent was obtained from the participants prior to the MR scans. The touch-proof safety connectors on the electrode leads were connected via a Biopac MECMRI-1 cable to the Biopac MRIRFIF pi filter (BIOPAC Systems, Goleta, USA). The filter reduces noise from the outside and is located in a panel between the scanner room and the control room. For stimulation, a neurostimulator will be connected to the filter on the control room side. The neurostimulator was not used in the experiments as the output impedance of the copper cable and filter remained the same independent of the stimulator. For safety assessment of electrodes and leads, temperature measurements were performed in the 3T and 7T scanners. Image quality assessment and imaging of the leads for stray field correction in MRCDI was performed at 3T. See S1 in the supplementary material for further details and results.

Temperature measurements
For temperature measurements at 3T, the built-in birdcage body coil was used as the transmit coil while at 7T, a birdcage head coil was used for excitation (7T volume T/R, Nova Medical, Wilmington, MA). Fiber-optic probes (Opsens Solutions, Quebec City, Canada) were used to measure the temperature. Four probes were available. The probes were placed in the gel between the electrode and the scalp at various locations indicated in Table 3 and Fig. 6a and b. When a reference probe was used, it was taped to the top of the head of the subject and insulated with a pad to better imitate the scenario of the other probes.
A Rapid Acquisition with Relaxation Enhancement (RARE) sequence was used for both field strengths to obtain a high SAR for the temperature measurements. The sequence parameters were adjusted to obtain approximately 100% reported SAR by the scanner relative to the SAR limit. At 3T, the RARE sequence parameters were repetition time T R ¼ 175 ms, echo time T E ¼ 100 ms, refocusing tip angle ¼ 180 , echo train length ¼ 15, image matrix 512 Â 512 Â 27 and resolution 0.43 Â 0.43 Â 5.2 mm 3 . And at 7T, T R ¼ 3584 ms, T E ¼ 47.54 ms, echo train length ¼ 9, image matrix 768 Â 768 Â 33, resolution 0.28 Â 0.28 Â 3 mm 3 and a varying refocusing tip angle. A Pseudo Continuous Arterial Spin Labeling (pCASL) sequence (T R ¼ 4100 ms, T E ¼ 18 ms, excitation tip angle ¼ 90 , image matrix 73 Â 73 Â 60, resolution 3 Â 3 Â 4 mm 3 , tag duration/tag delay ¼ 1500/1800 ms, tag pulse angle ¼ 24 and tag gradient strength ¼ 7 mT/m) was also used at 3T, as it is a relatively high SAR sequence that will potentially be used for TES-MRI studies. About 50% SAR was reported for the pCASL sequence. The sequences ran for 20 min to achieve sufficient data to accurately model the temperature increase and find the steady-state temperature. The model used is where T(t) is the temperature at time t, T ss is the steady-state temperature, DT is the difference between the start and steadystate temperature, and t C is the time constant of the exponential term.
Stray field comparison An MRCDI experiment with 1 mA current injection was performed to illustrate the change of stray fields from the leads when the improved lead configuration is used. Lead configurations seen in Fig. 1a and our proposed use as seen in Fig. 3a were compared. See G€ oksu et al. [10] for further details on the used MRCDI method. The imaging of silicone rubber used for cable tracking is presented in supplementary material S1.

Antenna effect
Simulations with varying copper lead lengths showed that the antenna effect occurs at odd multiples of ¼ RF wavelength with ¼ being worse than ¾ (Fig. 4a). The same simulations but with lowconductivity silicone rubber showed that the antenna effect is eliminated with this material (Fig. 4a). Further investigation of the relationship between the antenna effect and conductivity at the worst-case length (25 cm) is shown in Fig. 4b. Carbon fiber leads with a conductivity of 6.1$10 4 S/m [31] only reduces the severity of the antenna effect by about 25%, while low-conductivity silicone rubber robustly prevents the occurrence of an antenna effect.

SAR for realistic lead configurations
SAR simulation results for realistic lead configurations are presented in Fig. 5 and Table 2. In Fig. 5, 1 g local head SAR for all three electrode montages with center leads (see Fig. 3aec) are shown. Adding electrodes, gel, and leads to Duke, changes the spatial variation of SAR, especially around the electrodes, but 1 g local head SAR is for all simulations in the same location as for the reference simulation. For 128 MHz simulations, 1 g local head SAR occurs on the skin on the left side of the neck while for 298 MHz it is in the cerebrospinal fluid. As seen in Table 2, only minimal changes to the B 1 field and SAR occur for 128 MHz and 298 MHz with circular electrodes, while the center-surround electrodes have more influence on the B 1 field, and therefore higher SAR after normalization.

Temperature measurements
One of the temperature measurements (indicated in Table 3) including the fitted model is shown in Fig. 6c. The modeled steadystate temperatures for all the measurements are listed in Table 3. Probe positions indicated in Table 3 are presented in Fig. 6a and b for the center-surround and circular electrodes, respectively.
For the circular electrodes at 3T, the highest temperature was observed on the posterior electrode for the anterior-posterior montage, with a temperature of 37.6 C compared to 35 C for the reference probe. For the right-left montage, the maximum temperature on the electrodes was only 1 C higher than for the reference probe.
The center-surround electrodes at 3T showed the highest temperature increase (max 38.5 C), with no observable difference between electrode pads with and without leads. Off-center leads also did not give rise to higher temperatures. For the pCASL sequence, the steady-state temperature was about 1 C lower than for the RARE sequence in the same session, marked with an asterisk in Table 3.
Very limited increase was found for all measurements at 7T. The highest measured difference between the reference probe and a probe on the electrodes was 0.6 C.
Stray field comparison Fig. 7b and e show the fields from the leads calculated with Biot-Savart law using the tracked lead location seen in Fig. 7a and d. Fig. 7c and f are the measured current-induced fields in the MR scanner. The measured fields are both from currents flowing in the leads and in the subject's tissue. It is clear from comparing Fig. 7b and c that DB zc is dominated by stray fields from the leads. With our optimized lead configuration with the silicone rubber leads as seen in Figs. 3a and 7d, where the leads are aligned in the z-direction, the stray field from the leads are greatly reduced (Fig. 7e) and therefore there is no relationship between lead stray field and DB zc that is dominated by tissue currents.

Discussion
TES electrodes with copper leads pose a potential danger to the subject when used during an MRI session. Due to the coupling between the RF field of the scanner and the highly conductive leads, burns of the subject's scalp can occur unless appropriate measures are taken, such as adding safety resistors in a well-considered way. To minimize coupling between the RF field and the leads in general, we propose to use leads made with a low-conductivity material. By that, we gain flexibility to optimize the leads for the intended applications while ensuring safety. Additionally, this makes it easier to safely design more complex electrode configurations with multiple leads, such as the 4x1 montage [2] used for focal stimulation in TES-fMRI experiments. These electrodes and leads can relatively easily be constructed in-house from sheets of conductive silicone rubber.
In simulations, the antenna effect was found for odd multiples of ¼ RF wavelength (Fig. 4a), which is in agreement with previously reported experimental results [15] with the same boundary conditions. The antenna effect is often believed to occur at ½ RF wavelength only, but as pointed out by Balasubramanian et al. [15], this depends on the boundary condition of the leads. With low impedance at one end and high at the other, it occurs at ¼ RF wavelength, whereas with the same boundary condition at each end, e.g. immersed in tissue, antenna effect occurs at ½ RF wavelength.
Simulation results with varying lead conductivity at worst-case length (Fig. 4b) prove that the antenna effect will not occur for silicone rubber with lead conductivity at 29.4 S/m. This increases the flexibility of the lead configuration and improves the experimental setup for MRCDI experiments. Additionally, no safety resistors are needed, which decreases the overall resistance of the leads compared to the conventional setup and allows for higher stimulation currents. This enables the use of increased stimulation current to study immediate and after-effects on BOLD activity using a standard stimulator. Also, safety resistors enforce nodes in electromagnetic waves, which may cause high local fields causing heating in nearby material and even resistor damage [24]. Careful design is needed to limit these effects and ensure appropriate distance from tissue. The simulations also show that carbon leads, which are often used as a safer alternative to copper leads, only reduce the severity of the antenna effect in our simulations by about 25%. Therefore, using carbon leads can provide a false sense of safety and has to be considered carefully for each specific case. Previous simulation work on a 256-electrode EEG cap has shown consistent results by comparing peak local SAR for varying lead conductivities [22]. The authors reported a 6-fold increase in peak 1 g local head SAR for high conductivities (including carbon) and no increase for conductivities below 100 S/m. The study was not for a specific resonance condition as in our case. Our results may therefore also be relevant for the EEG-MRI community.
Only very limited changes were observed for R m and R 1g for all simulations with realistic lead configurations as seen in Table 2. In most cases the B 1 was slightly lower than B 1,ref due to the additional load on the coil when electrodes and leads were included. For 1 W input power, this is also reflected in the slightly lower SAR for some simulations, especially with leads. For the same reason, all simulations with leads have lower SAR and more influence on B 1 than simulation without leads for 1 W input power. Slight changes to the spatial distribution of the RF field caused by electrodes and leads can also influence R m and especially R 1g .
Overall, a very small change and mostly reduction in SAR is seen before B 1 normalization, while some increase in SAR is reported after normalizing. This is not seen as a problem since, if the scanner increases the input power, then the calculated SAR will be adjusted accordingly and the SAR safety limits will be reached earlier. In worst case, this will have an influence on the available ranges of sequence parameters, but not on safety.
Although the electrodes have an influence on local SAR values in the proximity to the electrodes as seen in Fig. 5, the 1 g local head SAR close to the electrodes does not exceed peak 1 g local head SAR already present in the reference simulation. The peak 1 g local head SAR was also in all simulations at the same location as for the reference simulations. Therefore, the local head SAR limits imposed by the scanner will still ensure conformance with the safety regulations. Additionally, in the 298 MHz simulations, the centersurround electrodes are located close to the locations of peak local head SAR without negative effect.  Fig. 3aec). Insets in the corner for g), h) and j) show the surface SAR around the electrodes. Fig. 6. In a) and b) the numbers on the electrodes indicate the probe position referred to in Fig. 6c and Table 3 c) Temperature measurement for center-surround electrodes with center leads (Fig. 3c). L in the legend indicates that it is a measurement on the left electrode. The black curves are the fitted models (Eqn (4)).
The highest measured temperature for the circular electrodes was on the posterior electrode. This is most likely not due to higher power dissipation on the electrode, but rather better thermal insulation as the head rests on the electrode and cushions.
At 3T, more heating was observed for the center-surround electrodes than the circular electrode. The heating was independent of lead position as well as whether leads were attached or not. In agreement with a previous study by Kozlov et al. [25], this indicates that the shape and size of the electrodes have high influence on heating. Therefore, care must be taken when designing new electrodes for TES-MRI experiments. The impedance around the ring of the center-surround electrode is about 300 U. Higher impedance would cause less heating, but it is a tradeoff between heating and homogeneous stimulation currents for focal stimulation. It has to be pointed out that the heating observed is with a high-SAR RARE sequence, which is not a recommended sequence for TES-MRI experiments. Usually, low-SAR echo planar imaging (EPI) or gradient-echo sequences will be used or in worst-case the pCASL sequence also tested in the temperature experiments. The pCASL sequence showed about 1 C less heating than the RARE sequence. In contrast to our findings at 3T, a previous study reported much less heating on the center-surround electrodes [32]. The unspecified conductivity of the electrode material has high influence on the heating, but more importantly only an EPI sequence was used for the heat test. Since EPI used for fMRI experiments is usually a very low-SAR sequence at 3T, insignificant heating would be expected.
At 7T no considerable heating was observed. This is attributed to the fact that SAR is already higher at 7T for similar sequences, and the input power is therefore more restricted compared to 3T.
Although noticeable heating was measured on the centersurround electrodes at 3T, the temperatures were always lower than limits imposed by the international guidelines IEC 60601-2-33 [30] stating that the maximum tissue temperature has to be limited to 39 C.
To further ensure safety and reduce the risk of resistor damage, the conventional electrodes and leads used in this work are limited to use with EPI sequences. The manufacturer requires removal of the cables when other sequences are used. Under the conditions evaluated, this is not necessary with low conductivity silicone rubber leads since no safety resistors are used and safety tests have been performed for high SAR sequences. Table 2 Head SAR and local head SAR (1 g average) ratios (R m and R 1g ) for both field strengths and all simulated electrode montages compared to reference simulations. SAR is given for 1 W input power as well as normalized for the B 1 field in the center slice of the coil. The ratio between B 1,ref and B 1 for the corresponding simulation is used for normalization. All local head SAR maxima were at the same location as for the reference simulation.

Frequency
Electrodes Setup Normalized to 1 W input  Table 3 Modeled steady state temperature from all measurements with various setups and probe positions. The numbers for the positions are indicated in Fig. 6aeb and L, R, A and P refer to electrode position (left, right, anterior and posterior). Ref is the reference probe on top of the head away from the electrodes. For the RARE sequence, the SAR was approximately 100% and for ASL, it was approximately 50%, varying slightly with subjects. The asterisks (*) indicate the same scan sessions for a RARE and pCASL sequence with electrodes and temperature probes in the same location. The double asterisks (**) indicate the data shown in Fig. 6c The rubber leads being visible on MR recordings without adding additional material is a practical benefit when doing stray field correction in MRCDI [9] (see supplementary material S1). Conventional copper leads and insulation are not visible on MRI, and therefore additional material needs to be attached around the leads before use. Attention must be paid to image distortions caused by chemical shift, gradient non-linearity, and concomitant fields, however.
The effect of the stray field on the measured DB zc demonstrated in our MRCDI experiment necessitates correction by applying the Biot-Savart-Law [9,10]. However, any errors and inaccuracies of lead position estimation will have far less influence for the optimized cable configuration made possible by using silicone rubber leads.
Additionally, the lead location cannot be tracked during DB zc measurements. Therefore, any movements during the experiment will have detrimental effect on the stray field correction for the non-optimized setup.
The electrodes and leads have very limited and only superficial influence on B 1 and B 0 maps. However, when calculating the signalto-fluctuation-noise ratio for an EPI time series, image artifacts having long ranging effects (centimeters) were found when copper leads and safety resistors were used. These artifacts were not found with silicone rubber leads. (see supplementary material S1).

Conclusion
We have proposed to use low-conductivity silicone rubber as leads for current injection electrodes in the MR scanner. This eliminates the potential safety hazard that comes with the coupling between high-conductivity materials and the RF field, such as the antenna effect that is not necessarily eliminated with carbon cables or safety resistors. For our setup, the simulations showed no increase in head SAR and head local SAR for both field strengths. Additionally, no temperature above the safety limits was recorded. Due to the increased flexibility of lead configurations, these electrodes offer an advantage for MRCDI experiments due to the reduction of compromising stray fields. For TES-MRI experiments, the maximum stimulation currents can be increased for voltagelimited stimulation devices due to lower overall resistance.

Declaration of competing interest
There are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced its outcome.