Segmenting electroencephalography wires reduces radiofrequency shielding artifacts in simultaneous electroencephalography and functional magnetic resonance imaging at 7 T

Purpose Simultaneous scalp electroencephalography and functional magnetic resonance imaging (EEG‐fMRI) enable noninvasive assessment of brain function with high spatial and temporal resolution. However, at ultra‐high field, the data quality of both modalities is degraded by mutual interactions. Here, we thoroughly investigated the radiofrequency (RF) shielding artifact of a state‐of‐the‐art EEG‐fMRI setup, at 7 T, and design a practical solution to limit this issue. Methods Electromagnetic field simulations and MR measurements assessed the shielding effect of the EEG setup, more specifically the EEG wiring. The effectiveness of segmenting the wiring with resistors to reduce the transmit field disruption was evaluated on a wire‐only EEG model and a simulation model of the EEG cap. Results The EEG wiring was found to exert a dominant effect on the disruption of the transmit field, whose intensity varied periodically as a function of the wire length. Breaking the electrical continuity of the EEG wires into segments shorter than one quarter RF wavelength in air (25 cm at 7 T) reduced significantly the RF shielding artifacts. Simulations of the EEG cap with segmented wires indicated similar improvements for a moderate increase of the power deposition. Conclusion We demonstrated that segmenting the EEG wiring into shorter lengths using commercially available nonmagnetic resistors is effective at reducing RF shielding artifacts in simultaneous EEG‐fMRI. This prevents the formation of RF‐induced standing waves, without substantial specific absorption rate (SAR) penalties, and thereby enables benefiting from the functional sensitivity boosts achievable at ultra‐high field.

S2 Head-shaped agar-gel phantom The agar-gel phantom is composed of three distinct parts with the following composition: • Top part: 3.0 g/l agarose, 9g/l NaCl and 0.13mM Gd-DO3A-butrol in 1850ml H 2 O • Middle and bottom part: 6.46 g/l agarose, 9g/l NaCl and 0.24mM Gd-DO3A-butrol in 2000ml Each part was poured into a plastic bag and cast inside a plastic shell in the shape of a human head.
The interface between the bags can be seen on anatomical images (Fig.S2). The dielectric properties of the phantom were measured on separate mixtures at 300 MHz and ambient temperature (23 • C) using a dielectric probe (DAK-3.5, SPEAG, Switzerland). For both mixtures, the conductivity was σ=1.68 S m −1 and the relative permittivity (dielectric constant) was ϵ r = 76.0.

S5 Different human model
The EEG-fMRI setup was simulated with the female Ella human model. The EEG wiring and electrodes position were adapted to the smaller and different head shape. The simulation model was discretized into 133.0 MCells with non-uniform mesh steps from 0.27 × 0.25 × 0.25 mm 3 to 57 × 70 × 70 mm 3 .
The transmit field and SAR distribution without EEG, with EEG cap and with the segmented EEG cap are shown in Figure S5. The simulation with EEG depicts a similarly shielding pattern compared to the model with Duke, with a strong attenuation in superior regions of the head, local dropout regions close to the wires and electrodes, and intense transmit field close to the wire bundles.
However little attenuation is observed in posterior regions of the head. In overall, the B + 1 attenuation is attenuated by only 13.7% with Ella model compared to 31% with Duke model.
In the simulation with the segmented EEG cap, the EEG-induced RF shielding is substantially reduced, with only 4.35% B + 1 amplitude losses compared to no EEG. A lower power deposition in the upper part of the head was observed with the EEG cap, while the segmented EEG cap caused a higher power deposition close to the scalp. The peak SAR 10g was Similarly to the method described in the main paper, the wires were sewn to a subtemporal cap placed on the agar-gel phantom, and the transmit field maps were acquired in a 7T MR scanner.
Matching EM simulations were performed, in which the 32 wires were set as lossy conductors, and the thick or thin electric insulation layer defined accordingly.
The results are summarized in figure S6.   Figure S6: Transmit field maps acquired in the agar-gel phantom, or simulated in the human model with wire-only EEG models. All wiring models produced strong transmit field shielding artifacts compared to no wires, notably at the back of the imaging subject where the wires were bundled. Constantan and nichrome presented a slightly different transmit field disruption pattern related to their thinner insulation layer that affected the propagation and interaction of EM waves, but the attenuation and inhomogeneity were similar to copper. Carbon fiber wires, with the highest linear resistance, presented higher attenuation but a better homogeneity compared to copper.

S7 Voltage across segmentation resistors
The distribution of the peak voltage across segmentation resistors, normalized to 1 W of input power, is plotted in Fig.S7. The maximal peak voltage is 2.35 VW −1/2 .
On our MRI scanner, the peak power at the coil plug is 5.1 kW (combined 8x1 kW RF amplifiers, with 37% losses between the amplifiers and the coil plug 4 ). Neglecting the losses between the plug and coil elements (including the power splitter and TR switches), the maximal voltage across segmentation resistors would be 167 V at full power, which is below the working voltage of these resistors (200 V).
In typical measurements, the coil operates using about 3 kW out of 8 kW combined power at the output of the amplifiers, therefore 1.9kW at the coil plug, resulting in a maximal voltage of 102 V across segmentation resistors.

S8 Detailed RF coil models
The RF coil model used in EM simulations, described in SI 1, Fig S1 and Table S1, is a generic 8-loop coil with loops of similar dimensions compared to the physical coil used in MR measurements. The coil model was simplified, in particular each loop was modeled using only three distributed capacitors, and without the concentric shields around each loop. Additional sets of EM simulations using the following two RF coil models were performed to determine whether the simplifications substantially affect the overall B + 1 field distribution and EEG-induced RF shielding artefacts: Simulations with the human model: EM simulations with the above-mentioned coils were performed using the realistic Duke human model as the imaging subject and three EEG configurations (No EEG, Full EEG cap and Segmented full EEG cap). The results are presented in Fig. S10, together with those obtained using the most simple coil model (3 capacitors per loop, described in S1, results from Figs.2 and 7) and MR measurements on the human volunteer for reference (same measurements as on Fig.2). For a given EEG configuration, a similar transmit field distribution was observed together with small local discrepancies, such as with the shielded coil model where the B + 1 was locally attenuated at the back of the head (arrows 1), and slightly more intense at the front (arrows 2). Nevertheless, with all three coil models, similar EEG-induced RF shielding patterns were observed, with a practically identical attenuation and inhomogeneity, and similar SAR efficiency (Table S3).
Simulations with the agar-gel phantom : EM simulations with the above-mentioned coils were performed using the digitized agar-gel phantom as the imaging subject and two EEG configurations (No EEG, Full EEG cap). The results are presented in Fig. S11, together with those obtained using the most simple coil model (3 capacitors per loop, results from Fig.3). In overall, a similar transmit field pattern is observed across all three coil models for a given EEG configuration. The full EEG cap cause similar RF shielding patterns with all three coils with slight differences, particularly with the shielded coil for which the B + 1 at the back of the phantom with EEG is stronger compared to results with the other coil models.
Compared to the more detailed RF coil models, the simplified coil model (SI S1) provided a similar transmit field distribution for a given EEG configuration, and more specifically similar EEG-induced RF shielding patterns. This suggests that the simplified coil model used in this study is a valid approximation of the more detailed RF coil models for the purpose of assessing EEG-induced RF shielding artifacts.      Figure S10: Transmit field maps simulated with three coil models, and different EEG configurations using the realistic human model Duke as imaging subject. MR measurement results from Fig.2 were added for reference on the right-hand side. The transmit field is expressed as a fraction of the nominal flip angle. To enable visual comparison, simulation results were normalized to achieve the same nominal flip angle at the center of the imaging subject in the "No EEG cap" case compared to MR measurements. The same normalization was kept for all simulations with a given coil model. In overall, a similar B + 1 distribution was observed with all coil models, with local discrepancies, especially with the shielded coil model where a lower and higher B + 1 are depicted at the back (arrows 1) and front (arrows 2) of the head respectively. Similar RF shielding patterns are observed independently of the coil model. No EEG Full EEG Cap Figure S11: Transmit field maps simulated with three coil models, with and without the full EEG cap, and using the digitized agar-gel phantom as imaging subject. MR measurement results from Fig.3 were added for reference on the right-hand side. All transmit field maps are expressed as a fraction of the nominal flip angle. Simulation results were normalized to achieve the same nominal flip angle at the center of the phantom compared to measurements. In overall, all coil models achieve a similar transmit field distribution with similar RF shielding patterns with all three RF coil models. There are quantitative differences, such as the stronger B + 1 at the back of the phantom with EEG while using the shielded coil model.