A novel whole-head RF coil design tailored for concurrent multichannel brain stimulation and imaging at 3T

Abstract

Purpose:

Multichannel Transcranial Magnetic Stimulation (mTMS) [1] is a novel non-invasive brain stimulation technique allowing multiple sites to be stimulated simultaneously or sequentially under electronic control without movement of the coils. To enable simultaneous mTMS and MR imaging, we have designed and constructed a whole-head 28-channel receive-only RF coil at 3T.

Methods:

A helmet-shaped structure was designed considering a specific layout for a mTMS system with holes for positioning the TMS units next to the scalp. Diameter of the TMS units defined the diameter of RF loops. The placement of the preamplifiers was designed to minimize possible interactions and to allow straightforward positioning of the mTMS units around the RF coil. Interactions between TMS-MRI were analyzed for the whole-head system extending the results presented in previous publications [2]. Both SNR- and g-factors maps were obtained to compare the imaging performance of the coil with commercial head coils.

Results:

Sensitivity losses for the RF elements containing TMS units show a well-defined spatial pattern. Simulations indicate that the losses are predominantly caused by eddy currents on the coil wire windings. The average SNR performance of the TMSMR 28-channel coil is about 66% and 86% of the SNR of the 32/20-channel head coil respectively. The g-factor values of the TMSMR 28-channel coil are similar to the 32-channel coil and significantly better than the 20-channel coil.

Conclusion:

We present the TMSMR 28-channel coil, a head RF coil array to be integrated with a multichannel 3-axisTMS coil system, a novel tool that will enable causal mapping of human brain function.

1. Introduction

Understanding the causal aspects of human brain function requires precise tools to discern how brain activity and behavior are linked by changing neural circuit dynamics [3]. One key feature of such tools is the capability to concurrently perturb ongoing regional processing as well as to record the elicited changes across the network nodes. Hence, combining Transcranial Magnetic Stimulation [4] (TMS) with functional magnetic resonance imaging (fMRI) appears a highly promising avenue for novel non-invasive studies. The combination of TMS and fMRI was presented for the first time two decades ago [5,6]. After showing the feasibility of combining non-invasive brain stimulation and functional imaging several groups have performed concurrent TMS/fMRI studies [7–9] yielding important data characterizing effects of TMS on the ongoing neural processes immediately after stimulus delivery.

However, two major limitations need to be addressed to enable precise causal mapping of whole-brain networks using TMS-fMRI technology. First, the use of MRI-based neuronavigation systems for positioning the TMS coil has shown to be critical for reproducible targeting [10]. In the early TMS-fMRI studies, the use of those systems was not an option due to the lack of space in the head volume coils used for functional imaging (see Fig. 1. Column 2). In more recent studies, the use of a dedicated RF coil array placed under the TMS [11] (see Fig. 1. Column 4) has allowed a more controlled method for placing the stimulation coil in the bore using neuronavigation. Even though methods have been proposed to achieve a post-hoc placement control of the stimulation coil based on the functional images acquired [12], it is still not possible to correct for possible subject/coil movement during the fMRI acquisition. This increases the spatial variability of the stimuli within each fMRI run and necessitates moving the subject out of the scanner bore for coil re-positioning. Second, the existing hardware to acquire functional images while stimulating the brain is quite limited (see Fig. 1). Achieving whole brain imaging with high sensitivity currently remains infeasible for combined TMS-fMRI. The main reason is that both the RF coil arrays used for fast fMRI acquisition and the TMS coil need to be placed close to the head for optimal operation, limiting the usability of state-of-the-art rigid imaging helmets (Fig. 1, Column 1). For this reason, the birdcage RF coil has been the state-of-the-art instrumentation used for experiments requiring whole-head coverage (Fig. 1 Column 2) since it provides enough room for placing the TMS coil and the subject’s head. The drawback of this set up is the much lower sensitivity compared to the standard 32/64-channel RF head coils. The use of the dedicated RF coil array [11] fixated under the TMS coil provides a very high signal to noise ratio (SNR) in the target region (Fig. 1 Column 4). Combining two such RF coil arrays (one for the stimulation site and another on the contralateral side) extends coverage over much of the brain with adequate imaging sensitivity [13]. However, whole brain imaging with comparable SNR to standard 32-channel RF coil arrays is still not possible with the existing solutions. To address these issues currently hampering the TMS/fMRI stimulation/imaging technology, a system needs to be developed that is fully capable of (i) to apply stimulation pulses within a defined region with high spatial accuracy in the MR scanner, (ii) monitor the target position during the experiment to compensate coil/subject’s movements without any physical movement of the stimulation device and (iii) to allow stimulation on the subject’s head while acquiring MRI receive signals with sensitivity comparable to a commercial high-density head RF coil array.

Fig. 1.

Summary of available instrumentation for MR acquisitions and its capability for concurrent TMS/fMRI acquisitions.

Multichannel TMS [1] is a novel non-invasive brain stimulation technique. The use of multiple TMS units in an array configuration enables shifting the location TMS ‘hot spot’ electronically without any mechanical movement, making this technique a powerful solution for the stimulus re-focusing problem inside MR environment. The electronic shifting of the “hot spot” is achieved by computationally determining the current amplitudes to be passed to each of the coil elements to synthesize a desired target field pattern [14,15]. While the mTMS concept is in principle very simple and initially presented more than two decades ago, construction of the first large-scale systems is just currently under way [14–16]. The solution we propose [15] is based on a modular design, where each single TMS unit is a 3-axis TMS coil; other groups have proposed different configurations for accurate electronic steering of the stimuli [14,16]. The magnetic field production and cooling capability (heating per pulse) of the 3-axis TMS coil have been previously described [15]. In our previous study we presented a non-MR compatible design for the 3-axis TMS coil. More recently, we have developed the concept further to achieve sufficient structural strength to enable use in the 3T MRI field environment [17]. Our proposed mTMS system can realize a large variety of different E-field ‘hot spot’ distributions enabling new stimulation approaches that are not feasible with commercially available TMS systems inside the MRI bore (see, e.g., Fig. S1.)

Finally, to allow efficient brain stimulation while acquiring high-sensitivity functional MR images, we have designed and constructed a 28-channel RF coil array for head imaging. This RF coil which can be integrated with our proposed mTMS solution, will be described in this article in detail including the analysis of the effects of the mTMS on the $B_1^{-}$ of the constructed RF coil. Interactions of the transmit coil field $B_1^{+}$ and SAR with the mTMS have been previously analyzed and reported by the authors [2]. Other interactions between a TMS coil and a receive-only RF loop tuned to the Larmor frequency at 3T were carefully investigated in previous work [11], demonstrating minimal effects between the systems due to their different operating frequency range. We also demonstrate the achieved signal-to-noise ratio (SNR) and parallel imaging capabilities of the RF coil array by acquiring anatomical images and EPI based resting-state fMRI data obtained from human subjects. An illustration of the envisioned complete system inside the MR environment is shown in Fig. 2A.

Fig. 2.

A) Proposed 3-axis mTMS system integrated with a whole head 28-channel RF coil array. The mTMS system will be placed on designated holes in the constructed TMSMR 28-channel coil B) 3D-CAD model of the designed housing for the TMSMR 28-channel coil (different views). Dimensions are annotated on the figures.

2. Methods

2.1. Helmet design

The RF coil housing design consists of 3 parts: 1) a 3D-printed helmet to place the RF loop elements on, 2) the base under the helmet to place all interface components to the scanner plug and 3) the covers of the helmet with holes for the TMS units to slide through. The helmet was designed to have a slight incline of 12.5° to allow the placement of stimulation elements in the posterior region. The layout of the mTMS system was chosen to cover both hemispheres in a symmetrical way reaching all possible positions and minimizing the spaces between TMS units while considering the head anatomy. Fig. 2B shows different views of the 3D CAD model and the dimensions of the helmet. The dimensions were chosen to fit 95% men’s and 100% women’s heads based on the head data sizes in Ref. [18]. The housing was designed using 3D CAD Software (Rhinoceros6.0, McNeel&Associates, Seattle, USA) and printed using Acrylonitrile Butadiene Styrene (ABS) for the base and Polycarbonate (PCA) for covers and helmet.

2.2. RF coil design and construction

The choice of the RF loops’ diameter was constrained by two factors: (i) the RF coil target depth which is in the range of 6–8 cm and (ii) the requirement that the TMS units need to be placed through the RF coil elements. To meet both these requirements, a diameter of 7 cm was chosen for each loop to obtain an optimal SNR for the desired target [19] while at the same time allowing 6 cm diameter holes inside them. For most of the elements that do not contain a TMS unit, we kept the same diameter for sake of simplicity. The layout of the RF coil is presented in Fig. 3A. We used transformer decoupling strategy with each neighbor for the RF loops that have TMS units inside [20](see, Fig. 3A orange positions) as well for some elements without stimulation. In addition, the geometric overlapping method was used whenever possible [21].

Fig. 3.

A) Layout of the TMSMR 28-channel RF coil array. Orange blocks show where preamplifier decoupling was implemented. Green blocks show the PCBs for matching/active detuning. Blue circles indicate stand-offs for preamplifiers. B) Circuit diagram of the PCBs for matching/active detuning for each single RF coil element. The values of the components are listed in Supplementary Table S1. C) Details of the construction of the RF coil. Photograph of the finalized RF coil. D) Left: Photograph of the MR-compatible 3-axis TMS coil, TMS unit of the proposed mTMS system. Right: Phantom with “dummy” TMS units used to analyze the effects of the TMS units on the SNR of the constructed RF coil. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

All receive elements were tuned to 123.25 MHz, which corresponds to the 3T scanner’s Larmor frequency (Magnetom, Skyra, Siemens Healthineers AG, Erlangen, Germany). Out of the total 28 receive elements, 16 included a hole to allow a 3-axis TMS coil to slide through the center of the RF coil loop and come to contact with the scalp. These 16 RF elements (channels 1 to 16) were tuned and matched with the TMS coil placed inside, to account for detuning effects of the TMS units [22]. Fig. 3B shows the circuit diagrams used for interfacing the RF coils to the preamplifiers using a second order matching network defined by Reykowski et al. [23]. To isolate the receiver elements from the RF transmit fields, an active detuning network was implemented in each loop (C_det and L_det are the components of the active detuning circuit) using a PIN Diode (Macom, MA4P7470F-1072T, Lowell, MA, USA). The diode is biased through an RF choke (CoilCraft, 1812 CS, Cary, Illinois, US) using a dedicated cable. Additionally, we included a fuse in each RF coil loop with 315 mA nominal fusing current as a safety feature. We moved the preamplifiers towards the sides and rear of the coil former to facilitate an overall open coil topology for positioning the mTMS system (see Fig. 3C). Cable lengths connecting the RF coils to the preamplifiers were minimized. We routed all coaxial cables from the preamplifier outputs to the base of the coil (in black) through designated holes on the base arranged in two groups of 7 cables. We wound the two final composites of coaxial cables around a 3D printed spiral form to construct an additional cable trap to minimize common modes. Finally, we covered the coil with the two-half housing and a thin strip above the eyes to isolate all electronic components from the user and to protect the preamplifiers on the sides and the posterior side of the coil.

2.3. Bench measurements

Bench measurements were performed to adjust all components of the RF coil (i.e., tuning, matching, mutual and preamplifier decoupling), as well as to investigate the influence of the TMS units when placed in their corresponding holes over the coil.

Our measurements were done using a vector network analyzer (EB071C, Agilent, Santa Clara, CA, USA). Preamplifier decoupling was tested with a double-loop probe according to the method proposed in Ref. [23]. The measurements were performed after loading the RF coil with an in-house built, bullet-shaped gel phantom (0.9% NaCl, 0.09% Gadolinium and 2% Agar) that covers most of the imaging volume.

A picture of the final prototype of the MR-compatible version of the 3-axis TMS coil is shown in Fig. 3D left. A simplified version of the z-element of the 3-axis TMS coil (referred to as the “dummy” TMS coil in what follows) that is expected to have the largest influence on the tuning and matching due to its parallel positioning with respect to the RF loop element, was fabricated to facilitate the tuning and matching process. A total of 16 “dummy” TMS units were positioned over the phantom’s head following the stimulation layout to model the ultimate 3-axis TMS coil array. The elements were subsequently fixated with clamps on their position to acquire reliable measurements (see, Fig. 3D right).

2.4. Simulations of the interactions between a TMS unit and an RF loop

To analyze the effects on the RF coil loop sensitivity when TMS units are inserted through them, simulations were carried out following the previously published methods [24,25] using Ansys Electronic Desktop 2021 R2 (Ansys, Canonsburg, PA, USA). First, a 7 cm diameter RF loop at 10 mm from the head phantom was modelled and loaded just with a head phantom of permittivity ${\xi }^{\mathrm{\text{'}}}=79$ and conductivity $\mathrm{\sigma }=0.69\mathrm{S}/\mathrm{m}$. Afterwards a simplified “dummy coil”, which emulates the “dummy” TMS we described above, was modelled and inserted through the RF loop to investigate the effects on the coil sensitivity when changing the distance between the TMS units and the head phantom. We simulated (i) no distance, (ii) 5 mm and (iii) 17 mm distance between the “dummy” TMS coil and the anthropomorphic phantom (see simulation set up in Fig. 4A Top). The EM simulations were run on a Dell Precision 7920 Tower server with 767 GB of RAM and 2 Intel Xeon Gold 6240, 18-core Processor. Electric and magnetic fields were obtained as well as the S-parameter matrix corresponding to the RF coil and the load. The calculated multiport S-parameter matrix was used in the integrated circuit design simulator. The set of simulations was planned to be compared with measurements as directly as possible. Hence, we tuned and matched each set up to achieve similar values as the ones we measured at the bench for the corresponding case. For the set up without the TMS unit selected as reference, the RF loop was tuned and matched to −6dB at Larmor frequency as the value measured at the bench when the TMS unit was removed from the RF loop element on the left side of the TMSMR 28-channel. For the following set ups with the “dummy” coil placed inside the RF loop, this was tuned and matched to achieve an S₁₁ of (i) −15 dB for no distance and (ii) −11 dB for distances of 5 mm and 17 mm. For each case we compared $B_1^{-}$ obtained with the one obtained for the reference loop with no TMS unit inside.

Fig. 4.

Simulation and measurement results of the effects on the RF loop’s sensitivity caused by varying the TMS coil’s distance from the phantom A) Top row: The leftmost panel shows a diagram of the set ups for the reference for both cases (simulations and measurements). The following panels show the set up used for each of the measurements. Center row. Results of the simulations. The results are shown as percentage loss or gain in $B_1^{+}$ with respect to the reference case depicted in the reference panel. For the reference case, the matching and tuning was set to achieve in a S₁₁ ~ −6 dB, to emulate the measurement case (see, methods). Results of the simulations in the first column are shown in μT. Bottom row. Results of the measurements performed using the bullet phantom shown in the top right of this figure. The results are shown as percentage loss or gain of the calculated SNR with respect to the reference case. For the reference case, the S₁₁ measured with no “dummy” TMS coil in that RF element was about −6dB. Results of the SNR measurement in the first column are shown in arbitrary units. B) Plot of the course of the sensitivity (either B1− or SNR) for both cases, simulations and measurement along the black dash line shown in A.

2.5. Measurements of the interactions between a TMS unit (“dummy” TMS coil) and an RF loop of the constructed TMSMR 28-channel coil

To validate the results of the simulations described above, we calculated normalized SNR maps for similar experimental cases using the constructed TMSMR 28-channel coil without any “dummy” TMS coils as the reference. The cases simulated corresponded then to the TMSMR 28-channel coil when placing only a “dummy” TMS coil on the left lateral position (inside an RF loop) and changing the distance to the phantom from (i) no distance but the housing, to (ii) 5 mm from the phantom and (iii) 17 mm from the phantom. The RF loop geometry was similar to the ones constructed in the TMSMR 28-channel coil. For each case, we compared normalized SNR of the RF coil with and without a TMS coil inserted. To calculate the normalized SNR, we acquired 2D Gradient Echo (GRE) images (all MR sequences are described in Table 1) of the bullet phantom (shown in Fig. 4B) using the same transmit voltage for both measurements. The vendor-provided standard flip angle (FA) mapping sequence was used. To isolate and analyze only the effects of the stimulation system on $B_1^{-}$, the SNR maps calculated for the optimal coil combination [26] were normalized by sin(FA) to remove $B_1^{+}$ effects produced by the TMS units [2]. Data was presented as percentage gain or loss (the reference always being the measurement with no TMS) that was calculated as follows:

$$SN{R}_{\mathrm{\%}}=\frac{\left(SN{R}_{TMS}-SN{R}_{\mathit{NoTMS}}\right)}{SN{R}_{\mathit{NoTMS}}}•100\mathrm{\%}$$

Table 1.

MR Sequences used in the evaluation of the performance of the TMSMR 28-channel coil.

MR Sequences	Paramenters
2D GRE	2 mm in-plane resolution, Slice thickness = 3 mm, 45 slices, Matrix = 96 × 110, TE = 2.76 ms, TR = 889 ms, FA = 90°
3D GRE	2 mm isotropic, FA = 90°, 72 slices, Matrix = 128 × 128, TE = 2.6 ms, TR = 50 ms
FA	2 mm in-plane resolution, Slice thickness = 3 mm, 45 slices, TE = 2.24 ms TR = 19720 ms, Matrix = 96 × 96, FA = 8°
B0	2 mm in-plane resolution, SL = 2 mm, 60 slices, TE = 8.42 ms TR = 962 ms, Matrix = 106 × 120 and FA = 90°
MPRAGE	TR/TE/inversion time (TI) = 2300 ms/4.14 ms/900 ms, FA = 9°, 172 slices, 1 mm isotropic, FOV = 265 × 283 mm2, MA = 230 × 256, and BW = 238 Hz/pixel
EPI	TE = 35 ms, TR = 1130 ms, 2.5 mm in-plane resolution, Slice thickness = 2 mm, FA = 90°

2.6. MRI data acquisition and reconstruction for in-vivo measurements

Before we acquired imaging data in-vivo, the coil array was subjected to the safety tests described by Keil et al. [27] with all “dummy” coils in place. The volunteer gave informed consent prior to participating in the study, and the imaging protocol was reviewed by the IRB.

2.6.1. SNR performance, g-factor maps and flip angle maps

To compare the performance of the new constructed coil with the commercially available 32-channel and 20-channel head coils, we acquired in vivo images with “dummy” TMS units on the TMSMR 28-channel coil using 3D GRE images (and flip angle maps). SNR maps obtained from in vivo images were calculated for all cases using the same methods as presented above, but without correcting for $B_1^{+}$ effect to capture the overall system performance. The SNR maps were masked to eliminate the skull and registered to each other to obtain quantitative values for each slice as the average of the SNR gain defined as:

$$SN{R}_{\mathit{gain}}=\frac{SN{R}_{TMSMR28Ch}}{SN{R}_{\mathit{commercial}}}$$

The g-factor maps for a middle transverse slice were calculated as defined in Ref. [28]. For the Flip Angle maps, we acquired the data using the vendor-provided standard flip angle (FA) mapping sequence (details listed in Table 1). The images were co-register using FSL (Analysis Group, FMRIB, Oxford, UK) and displayed in MATLAB (Mathworks, Natick, MA, USA) with the same brain mask that was used for the SNR maps.

2.6.2. B₀ maps and anatomical images

For an assessment of the effects of the mTMS system on the B₀ homogeneity, we acquired in-vivo B₀ maps with the TMSMR 28-channel coil with “dummy” TMS units on the subject’s brain. A vendor-provided two-echo gradient echo sequence was used for B₀ mapping. A magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence was performed to obtain anatomical images of the subject.

2.6.3. Echo-planar images (EPI) acquisition and temporal SNR (tSNR) calculations

To show the quality of functional images acquired with the constructed hardware in vivo using state-of-the-art fast imaging methods, we acquired 2D simultaneous multi-slice EPI (SMS-2) using in-plane GRAPPA (R = 2), 32 slices and 200 volumes. The subject was asked to keep his/her eyes open and remain relaxed and as still as possible (resting state paradigm). The temporal SNR was calculated as the mean of the time-series signal divided by the standard deviation after the raw data was corrected for motion and de-trended.

2.7. Concurrent TMS/fMRI experiment

To show the feasibility of the constructed RF coil to acquire functional images while TMS pulses are applied, we performed a concurrent TMS/fMRI experiment on an spherical phantom (165 mm diameter, 1.25 g NiSO₄ × 6H₂O per 1000 g H₂O dist., Siemens Model No 10496625) using a z-coil element of the 3-axis TMS coil that was potted in epoxy (see Fig. 9A) and placed in the position 5 (see RF Coil layout in Fig. 3A)

Fig. 9.

Concurrent TMS/fMRI feasibility experiment A) Left: Experimental setup. An epoxied z-coil was used to deliver pulses on a spherical phantom (not shown) on Position 5 of the TMSMR 28-channel coil. Right: Photograph of the epoxied z-coil used in the experiment. B) Oscilloscope trace of the timing signal acquired with the pick-up coil placed in the bore. The green line shows the slice trigger provided by the scanner/sequence. The yellow lightings symbols indicate when the TMS pulses were applied with respect to the acquired signal using the external probe in the scanner. C) Timeseries from the selected voxel of the phantom from the baseline (in black), from the images when TMS pulses were applied 20 ms after the trigger (in blue) and 60 ms after the trigger (in dashed purple). The images along the timeseries show the full image of the slice at that volume number. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

First, we acquired baseline functional images of the phantom using the same sequence described above (EPI acquisition) but acquiring only 45 volumes. Then, TMS pulses were applied in 5 blocks (1 pulse per volume in a pre-defined slice acquisition window) following a 10-volume period with no TMS pulses. The intensity of the stimulator was chosen to be 15% maximum stimulator output (MSO). A simple pick-up loop was placed in the center of the bore to acquire detailed sequence timing data.

The TMS pulse triggering was implemented using Presentation software (Neurobehavioural Systems, Berkeley, CA, US) synchronized with the slice timing triggers produced by the scanner. The TMS pulses were applied with two delays (20 ms and 60 ms).

To record the trigger signals and the signal produced by the pick-up loop, we used a digital oscilloscope (RTB2004, Rhode&Schwarz, Munich, Germany).

3. Results

3.1. MR coil properties

All RF elements of the coil were tuned and matched to 123.25 MHz achieving a S₁₁ < −15 dB. RF elements 1–16 which were tuned and matched with the “dummy” TMS units in place showed a frequency shift up by an average of 4.6 MHz and a standard deviation of 0.93 MHz, similar to values reported in the literature [11].

The active detuning efficiency was measured to be better than 40 dB for all elements. These results were verified during our safety tests, with variations of the flip angle near the RF coil elements of below 5% in the absence of the dummy TMS units (not shown). Noise correlation values ranged from 0.2% to 46.0%, with an average of 8.4%. RF elements with TMS units that were inductively decoupled using transformer decoupling had an average noise correlation of 10.1% and the RF elements with no TMS units that were geometrically decoupled presented an average noise correlation of 17.4%. No significant changes were observed in the noise correlation matrix when we acquired noise without the “dummy” TMS units in the TMSMR 28-channel coil.

3.2. Interactions between a TMS unit and an RF loop: simulations

The results of this analysis are shown in Fig. 4A, top row. The effects of the stimulation coil on the sensitivity of the RF loop decrease with the distance of the TMS unit to the phantom, nearly disappearing when the “dummy” TMS coil is 17 mm away from the phantom. Fig. 4B top shows the spatial profile of the sensitivity changes along the dashed line depicted in Fig. 4A. The gain observed at the positions further away from the RF elements stems from the fact that the reference simulation, using the RF coil only, is not tuned and matched optimally as it was described in the methods.

3.3. Interactions between a TMS unit and an RF loop: measurements

The results of the measurements to investigate the effects on the SNR when changing the TMS unit’s distance to the phantom are shown in Fig. 4A, bottom row. The effects of the “dummy” TMS coil on the measured SNR of the RF loop decrease with the distance of the TMS unit to the phantom in a very similar way as in the simulation, being minimal when the TMS unit is 17 mm further than the first position shown on the left. Fig. 4B bottom shows the course of the SNR changes along the dash line depicted in Fig. 4A. The gain observed in the positions further away from the RF elements comes from the fact that in the reference measurement (the RF coil only) the elements are not tuned and matched optimally as the RF loops were constructed to be optimally tuned and matched when a TMS coil is inserted as described in the methods.

3.4. In Vivo comparison with existing clinical RF coils

3.4.1. SNR performance, g-factor maps and flip angle maps

SNR maps of the constructed TMSMR 28-channel coil and the commercial 32- channel and the 20-channel head coils over twenty slices covering the whole brain are shown in Fig. 5. The average SNR gain of the TMSMR 28-channel coil when compared with the 32/20-channel head coils ranged from 0.58/0.72 to 0.74/.93 respectively, varying with location in the head. The SNR performance of the TMSMR 28-channel coil is closer to the SNR performance of the 20-channel head coil than to the 32-channel coil. This effect can largely be explained by the SNR losses produced (especially on the surface of the brain) when TMS units are present, but also by the bigger size of the designed one-piece helmet to accommodate large variance of head sizes and the TMS array.

Fig. 5.

SNR performance in-vivo. Left) SNR maps of the commercial 32 channel head coil. Center) SNR maps of the constructed TMSMR 28-channel coil. Measurements were performed with the “dummy” TMS units in place. Right) SNR maps of the commercial 20 channel head coil.

The g-factor maps of the central transversal slice for different in-plane acceleration factors are shown in Fig. 6. The average 1/g values (proportional to retained SNR) over the slices shown are reported in Table 2. The TMSMR 28-channel coil shows significantly better parallel imaging performance than the 20-channel coil and comparable to the 32-channel coil. This can be explained by the low noise correlation achieved by the constructed TMSMR 28-channel coil (average noise correlation of only 8% compared to 21% of the 32-channel coil and 13.4% of the 20-channel coil) as well as the higher channel count.

Fig. 6.

Comparison of the 1/g maps of the commercial 32/20-channel coils and the novel TMSMR 28-channel coil over the central transversal slice. Left column on each panel shows the 1/g map for single acceleration factors and the right column for combined acceleration factors.

Table 2.

Summary of the averaged 1/g values of the central transversal slice shown in Fig. 6.

R	TMSMR
	32-channel	28channel	20-channel
2	0.976	0.961	0.93
3	0.842	0.774	0.675
4	0.614	0.527	0.392
5	0.365	0.316	0.206
6	0.233	0.215	0.139
2×2	0.961	0.939	0.888
3×3	0.735	0.646	0.381
4×4	0.402	0.323	0.136
5×5	0.146	0.092	0.001

The Flip Angle maps acquired with the 32 channel and our TMSMR 28-channel coil with the dummy coils are shown in Fig. S2. In the same figure, we have plotted the difference between both maps to highlight the effects of the mTMS on the excitation field. The difference ranged from −15° to 15°.

3.4.2. B₀ maps and anatomical images

The calculated B₀ maps are shown in Fig. 7A (left panel). The standard deviation of the slices shown in the figure ranged from 58.2 to 96.4 Hz, with higher values in the top slices where the TMS units have a stronger impact. The values for each slice are shown in Fig. 7A (right panel). EPI distortion resulting from these local static B₀ fields could be partially corrected in post-processing. The anatomical images of the subject acquired with the TMSMR 28-channel coil are presented in Fig. 7B.

Fig. 7.

A) In vivo B₀ maps along the head showing the values converted to Hz (Left panel). Calculated standard deviation for each slice shown in the figure (Right panel). B) Anatomical images acquired with the constructed TMSMR 28-channel coil.

3.4.3. Functional images

The acquired functional images are shown in Fig. 8. The images shown represent the “raw” fMRI time series after being corrected for motion and de-trended. Some geometrical distortions can be observed especially on the top slices due to the high B₀ deviations discussed above. For selected voxels on three top slices, the timeseries are shown and the calculated tSNR value is reported in the figure.

Fig. 8.

Top) Functional images acquired with the constructed TMSMR 28-channel coil. Bottom) Timeseries from selected voxels are shown. The shown data has been corrected for motion and de-trended. The calculated tSNR is reported in each figure.

3.4.4. Concurrent TMS/fMRI experiment

A summary of the results of the concurrent TMS/fMRI experiment is shown in Fig. 9B and C. When the TMS pulses are delivered at 20 ms after the trigger, close to the fat saturation pulses, we do not see artifacts on the acquired images or a degradation of the tSNR. However, if the pulses are close to the imaging gradients, we observed large dropouts. The results obtained in this experiment demonstrate the feasibility of applying TMS pulses during the fMRI acquisition even with no substantial gap in the imaging protocol. We note that a simple pick-up loop or probe similar to the one we used is an invaluable tool to measure the sequence timing to determine appropriate times to deliver the TMS pluses in continuous fast fMRI acquisition scenarios. Even though the TMS pulses are very short (~300 μs), they can have a drastic effect on the acquired MR signals especially when placed close to the RF pulses, EPI navigators or imaging gradients.

4. Discussion

We presented the design and construction of a 28-channel whole head RF receive-only coil array to be integrated with a mTMS system and evaluated its performance. The design was optimized to meet all requirements for seamless integration and usability with the novel 3-axis TMS multichannel system that is currently being developed to enable whole brain mTMS concurrently with fMRI. The most important requirement for the RF coil design was to incorporate holes over the head to enable insertion of the TMS units on top of the scalp for efficient brain stimulation. Additionally, the mechanical design of the housing and distribution of the electronics had to be carefully planned to leave open access to allow incorporation of a fixation system that will hold the TMS units in place. For this reason, a Direct Connect plug (Siemens, Erlangen, Germany) integrated in a rectangular base was a critical element of the overall design. The placement of the preamplifiers on the side and rear part of the coil was to both (i) minimize the possible interactions with the high magnetic fields of the TMS units and (ii) optimize the cable length of the RF elements to the preamplifiers to minimize additional losses to the RF system. Another requirement for the defined placement was to leave appropriate space for placing the TMS units to cover the convex part of the head. Furthermore, the dimensions of the one-piece helmet had to be estimated not only to meet the wide range of adult head sizes but also to enable the capability of shifting the stimulation coils radially to place them on scalp for all subjects. Additionally, a small tilt of the head was incorporated to allow the positioning of the most posterior TMS units in a comfortable manner.

In our previous work [2], we extensively analyzed the effects of the planned 3-axis TMS multichannel array on the whole-body birdcage $B_1^{+}$ integrated in clinical 3T scanners. During the construction and development of the receive-only array reported here, we have evaluated the different effects of the TMS units on the sensitivity $B_1^{-}$ of the RF elements. First, the resonance frequency of the RF loop is shifted up by about 4.6 MHz. This detuning effect is due to the inductive coupling between the RF loop and the stimulation coil, which reduces the equivalent total inductance of the resonator, thus increasing the resonance frequency [29]. Second, from the simulations and measurements we performed it is evident that the TMS units produce sensitivity losses with a very well-defined spatial pattern. The effect observed can likely be explained by eddy currents induced on the TMS units by the MR signal, which appear as losses and shielding effects in the receive circuit. Additionally, we conclude that the losses decay very rapidly when the TMS unit is moved further away from the scalp. However, for the actual application of the mTMS system, the TMS units must be placed very close to the head to optimize the efficiency of the stimulation. Nevertheless, the mechanistic insights obtained from the presented data may support further design efforts to minimize these effects.

Despite the extensive work analyzing the interactions between the RF elements and the mTMS system, we must mention the limitations of our results due to fact that the whole stimulation system (mTMS) is still not yet integrated with the RF coil. Especially the effects of the bulky cables feeding the 3-axis TMS coils in the bore, which cannot be simulated easily, could still have an impact on the final SNR. However, our whole system has been designed to use RF filters to connect the mTMS to the stimulator placed in the system room to reduce the RF noise introduced by the cables going through the waveguide into the MR environment. These RF filters used for the 3-axis TMS coils were designed by MagVenture (Farum, Denmark) and are just extensions of the ones that have been previously used successfully for commercial TMS/MR set ups. We believe that this measure will reduce drastically the noise contribution to the system.

From the SNR performance comparison in vivo with the 32/20-channel commercial coils, we concluded that the TMSMR 28-channel coil achieves about 66% SNR compared to the 32-channel head coil and about 86% SNR compared to the 20-channel head coil (average of the gains over the presented slices in Fig. 5). From the channel count, one would expect that the performance would have been about 87% of the 32-channel head coil and a better SNR performance of around 140% compared to the 20-channel head coil. However, these values were influenced by both the high losses in some cortical regions due to the presence of the TMS units as discussed above as well as the overall larger distance from the RF loop subject’s brain due to the increased size of our one-piece design. The larger element size of the 20-channel head coil also explains the better SNR performance of this particular head coil compared to our constructed RF coil. However, for the parallel imaging performance, we obtained similar g-factor maps with the TMSMR 28-channel coil and the 32-channel coil and largely improved metrics compared with the 20-channel head coil. In this case, the low noise correlation factors achieved by our hardware due to a combination of a gap and overlap layout explains the good parallel imaging capabilities offered by the constructed coil. In light of the performance metrics compared against commercial coils, it is important to keep in mind two facts: (i) none of the existing commercial coils could be used with the mTMS system and (ii) to date there are no RF coil arrays capable of achieving full head coverage and high parallel imaging performance favoring the use of advanced acceleration techniques for combined TMS/fMRI neuroimaging studies. Taken together, our results show that the proposed coil provides highly competitive imaging performance when factoring in the additional design constraints imposed by the mTMS system integration.

The moderate B₀ effects reported above, produced by the susceptibility interface between the TMS units and the head, show that the homogeneity of the B₀ field is indeed disturbed. The values presented here are for an example subject and acquired with a standard shim offered by the commercial 3T system which could be potentially optimized. These values are similar to the ones reported in the literature [30,31] for inferior slices but about twice to three times more for superior slices, due to the proximity of the mTMS elements. B₀ inhomogeneity can disrupt spectrally-selective MR acquisition approaches such as fat saturation and water–fat separation [32]. Standard field mapping methods [33,34] can be used to correct for those distortions as long as excessive signal dropouts within the brain volume do not occur during the acquisition. Additionally, another highly successful method to correct for the inhomogeneities is to shim the field directly during the acquisition using dedicated hardware [31]. Following this idea, we have presented promising results using the TMS units as shim coils to “self-correct” for these effects [35]. Since imaging will only take place when no stimulation pulses are applied, it should be relatively straightforward to add this multi-coil shimming capability to our mTMS system. This would solve the increased B₀ inhomogeneity and improve the EPI acquisition quality and the approach is currently being investigated further.

To conclude, we present the RF receive array component of a versatile novel tool which will have the potential to perform non-invasive high-resolution causal mapping of human brain functions. We envision that the multichannel 3-axis TMS coil system integrated with the TMSMR 28-channel RF coil will become a powerful and precise tool to help disentangle the various mechanisms of action for network-level brain stimulation that can be utilized in human neuroscience and ultimately in clinical applications as well.

Resource and data sharing

The scripts to calculate SNR maps and g-factors can be accessed at https://github.com/lunade/MATLAB_SNR_gfactors_tools. The scripts to process all data and create the figures shown in this paper are available at https://github.com/lunade/Scripts_BS_TMSMR_28Ch. The imaging human data and phantom data will be available at https://nda.nih.gov/edit_collection.html?id=2530. The 3DCAD Model and simulation model for HFFS will be shared by petition.