Dynamic multicoil technique (DYNAMITE) MRI on human brain

Abstract

Purpose:

Spatial encoding for MRI is generally based on linear x, y, and z magnetic field gradients generated by a set of dedicated gradient coils. We recently introduced the dynamic multicoil technique (DYNAMITE) for B₀ field control and demonstrated DYNAMITE MRI in a preclinical MR environment. In this study, we report the first realization of DYNAMITE MRI of the in vivo human head.

Methods:

Gradient fields for DYNAMITE MRI were generated with a 28-channel multicoil hardware arranged in 4 rows of 7 coils on a cylindrical surface (length 359 mm, diameter 344 mm, maximum 5 A per coil). DYNAMITE MRIs of a resolution phantom and in vivo human heads were acquired with multislice gradient-echo, multislice spin-echo, and 3D gradient-echo sequences. The resultant image fidelity was compared to that obtained with conventional gradient coil technology.

Results:

DYNAMITE field control enabled the realization of all imaging sequences with average gradient errors ≤ 1%. DYNAMITE MRI provided image quality and sensitivity comparable to conventional gradient technology without any obvious artifacts. Some minor geometric deformations were noticed primarily in the image periphery as the result of regional field imperfections. The imperfections can be readily approximated theoretically through numerical integration of the Biot-Savart law and removed through image distortion correction.

Conclusion:

The first realization of DYNAMITE MRI of the in vivo human head has been presented. The obtained image fidelity is comparable to MRI with conventional gradient coils, paving the way for full-fledged DYNAMITE MRI and B₀ shim systems for human applications.

1 |. INTRODUCTION

Spatial encoding for MRI traditionally rests on linear x, y, and z magnetic B₀ field gradients corresponding to the 3 first-order spherical harmonic (SH) functions. The field shapes are physically produced by a set of 3 dedicated electrical wire patterns labeled x, y, and z gradient coils^1–5 and complemented by higher order SH coils for B₀ shimming.^5–10 As such, an MR system with second SH order B₀ shim capability, for instance, is equipped with 3 gradient coils plus 5 second-order coils for B₀ shimming, that is, a total of 8 nested coils. Similarly, MR systems with third SH order shim capability employ a multilayer structure of 3 + 5 + 7 = 15 nested coils for spatial encoding, that is, MRI, and B₀ shimming.

The multicoil (MC) technique for B₀ magnetic field control introduced in our laboratory enables the synthesis of a multitude of B₀ field shapes relevant to biomedical MR by scaled superposition of nonorthogonal basis shapes provided by a single-layer assembly of localized coils.¹¹ Magnetic field synthesis by such a dense matrix of simple, but individually controlled, DC loops positioned around the object under investigation has been demonstrated to provide a better approximation to the magnetic field distributions encountered in vivo and to outperform the SH framework for B₀ shimming. B₀ shimming with the MC approach and, moreover, the switched application, also referred to as the dynamic multi-coil technique (DYNAMITE) have been demonstrated to provide superior magnetic field homogeneity across the mouse brain,¹² rat brain,¹³ human brain,^14–16 human spine,¹⁷ and human torso.¹⁸ Although originally conceived for B₀ shimming, the versatile magnetic field-shaping capabilities of MC hardware were soon applied for other purposes, including nutation-angle homogenization at high magnetic fields¹⁹ and FOV reduction.²⁰ Low-order SH shapes, including linear field gradients, furthermore pose a subsection of the fields that can be realized with DYNAMITE magnetic field modeling.²¹ As such, DYNAMITE naturally lends itself to spatial encoding, and MRI¹¹ and proof-of-principle MRI have been demonstrated recently in miniaturized rodent-sized setups for both linear²² and nonlinear²³ MRI encoding. Note that with DYNAMITE MRI, all B₀ fields for spatial selection and encoding are purely based on MC hardware, and the conventional x, y, and z gradient coils are not used at all. Recently, the superior performance of MC shimming was combined with DYNAMITE MRI to demonstrate MRI in highly inhomogeneous magnetic fields by concurrent spatial encoding and B₀ shimming,²⁴ laying the methodological foundation for an integrated DYNAMITE MRI and B₀ shim system based on a single-layer MC assembly.

In this work, we extend DYNAMITE-based imaging previously described on preclinical animal systems to a MC setup suitable for human brain studies. The first multislice gradient-echo (GE), multislice spin-echo (SE), and 3D gradient-echo (GE) images are acquired to demonstrate the utility of DYNAMITE MRI for human brain imaging. The obtained image fidelity, comprising localization accuracy, signal strength, and contrast behavior, is shown to resemble that of conventional MRI relying on dedicated x, y, and z gradient coils, thereby paving the way for full-fledged human DYNAMITE MRI and B₀ shim systems. Preliminary results of this work have been previously published in abstract form.²⁵

2 |. METHODS

2.1 |. MR system

All studies were performed at the Magnetic Resonance Research Center at Yale University (New Haven, CT) on a 4 Tesla Magnex magnet interfaced to a Bruker Avance III HD spectrometer running on Paravision 6 (Bruker BioSpin Corporation, Billerica, MA) and equipped with gradients capable of switching 30 mT/m in 1150 μs. RF pulse transmission and reception was achieved with a 16-channel stripline RF coil with an inner diameter of 25.6 cm and an outer diameter of 31.2 cm.²⁶ The 4 kW maximum transmit power was equally split by two 8-way splitters and delivered to the coil with a static, fixed phase relation that was optimized for a homogeneous ${\text{B}}_1^{+}$ distribution across the human brain. Experimentation was approved by the Yale School of Medicine Human Investigation Committee, and all participants provided written informed consent prior to scanning.

2.2 |. Multicoil setup

All magnetic fields needed for DYNAMITE MRI were generated with a 28-channel MC setup shown in Figure 1. Individual MC elements were comprised of series-connected inner and outer loops of 100 turns each (AWG 22 insulated copper wire; Belden 8077, Belden Inc., St. Louis, MO). The center dimensions of the rectangular inner and outer loops were 118 × 54 mm and 144 × 80 mm, respectively. Each of the coil bundles had an approximately half-circled cross-section with a 5 to 6 mm height and 9 to 10 mm width. The 28 MC elements were arranged in 4 rows of 7 coils (Figure 1A, spanning 359 mm in the z direction) and mounted on the inside of a nylon cylinder (inner diameter = 344 mm, outer diameter = 414 mm). Mounting was achieved with up to 60 cable ties per coil (Catamount, L-5-50-9-C, ABB Installation Products Inc., Memphis, TN). The exact MC element positions have been defined according to the public multicoil information (PUMCIN) reporting guidelines²⁷ and are available as Supporting Information.

FIGURE 1.

Multicoil hardware. (A) Theoretical design of a 28-channel MC setup for human head imaging. (B) Nylon cylinder with mounting holes visible for series-connected inner and outer loops for each MC element. (C) Nylon cylinder with all 28 MC elements and 6 temperature sensors mounted on the cylinder ID. The MC elements terminate at 1 of 2 connectors, 1 of which is visible at the far end of the cylinder (black square). (D) Final MC setup including a 16-channel stripline RF coil. (E) Multichannel gradient waveform controller showing waveforms of 4 channels on the oscilloscope. A master board (lower left) ensures synchronous operation between up to 8 waveform boards containing 8 channels per board. ID, inner diameter; MC, multicoil

Each MC element is interfaced to a ±5 A constant current amplifier (Resonance Research Inc., Billerica, MA). However, the amplifier architecture limited the combined current per bank of 7 MC elements to ±10 A, thereby reducing the average amplitude per MC element to ±1.4 A. In all magnetic field synthesis calculations, the combined current per bank was taken into account as a hard constraint. The inductance and resistance of individual MC elements (including cables and filters) were measured at 6.2 mH and 4.5 Ω, respectively. Although the MC currents could theoretically be ramped from 0 to maximum in circa 500 μs, for convenience the MC rise times were forced as constant-time slopes equal to the 1150 μs MR system gradient rise times. The MC setup was equipped with six PT100 thermal sensor probes (Omega Engineering Inc., Norwalk, CT) for temperature control. No active air or water cooling was provided to the MC setup.

2.3 |. Multicoil calibration

Exact knowledge of the spatial magnetic field distribution produced by each MC element is crucial for the accurate synthesis of magnetic fields based on a combination of all MC elements. To this end, each MC element was calibrated using a multislice B₀ mapping method. Each B₀ map was acquired in the axial orientation as 30 contiguous 3-mm slices with 60 × 80 pixels over a 180 × 240 mm FOV. Phase evolution was sampled over 5 delays (0, 0.11, 0.33, 1.0, and 3.0 ms) as an extension to a 6 ms minimum TE. Following temporal phase unwrapping⁶ the B₀ magnetic field was calculated on a pixel-by-pixel basis as the linear slope of the phase-time curve. A first level of quality control was implemented during the B₀ map calculation, whereby pixels or time points were rejected when falling below a SNR threshold or when the phase-time curve deviated from linearity. For any given MC element, B₀ maps were acquired at −4, −3, −2, −1, 0.5, 1.5, 2.5, and 3.5% of the maximum current. A final MC calibration (in Hz per 100% current) was obtained by a linear fit of the B₀ amplitude-current curve. A second level of quality control was implemented during the MC calibration step whereby pixels or currents were rejected when the amplitude-current curve deviated from linearity, defined as 2% standard error of the linear regression analysis. Missing pixels in the final MC calibration maps were calculated by cubic spline interpolation from nearby pixels.

2.4 |. Validation of multicoil implementation and calibration

The experimental implementation and calibration of the established MC hardware were first validated in dedicated phantom experiments before they were applied to MRI. This was achieved through application of a set of switched field shapes throughout a large 3D FOV to demonstrate the field generation capability in both space (or shape) and time. Linear field gradients were selected because they constitute the building blocks of the MRI sequences employed in this work and, moreover, because their authenticity can be easily assessed both visually and quantitatively. The direction of an example 100 Hz/cm (0.23 mT/m) magnetic gradient field was increased by an arbitrarily chosen 45° per slice throughout the employed multislice B₀ mapping sequence, that is, a rotating gradient pattern was periodically incremented once per repetition time because slice selection was the inner loop (Figure 2A). The switched application of slice-specific field shapes as temporarily static, or pseudostatic, background fields during imaging similar to dynamic shim updating^28,29 enabled direct quality assurance checks through magnetic field B₀ mapping (Figure 2B). The experimental B₀ maps followed the desired magnetic field distribution closely with (99 ± 2) Hz/cm gradient amplitudes across the 10 slices and orientations considered, thereby validating the overall magnetic field synthesis calculations and demonstrating control of sequence timings and amplifiers. Note that this experimental paradigm aimed at validating key technical functionalities of the established MC implementation such as field control and switching performance rather than specific aspects of the MRI sequences employed later; any other set of distinct field shapes could have been used.

FIGURE 2.

(A) Schematic of example MC-based linear field gradients incrementally rotating by 45° on a slice-by-slice basis for functionality testing. Identical gradient patterns were generated in slices 1 and 9 (−y) as well as 2 and 10 (−x/−y), yet they resulted from independent MC field analyses targeting distinct spatial ROIs and unique combinations of 28 coil currents. (B) The dynamic experimental generation of high-fidelity slice-specific 100 Hz/cm rotating linear field gradients overlaid onto magnitude images of the employed water phantom indicates full control of MC waveforms over a spatial range large enough for MRI of the human head. ROI, region of interest

2.5 |. DYNAMITE MRI

Conventional MRI sequences were converted to DYNAMITE MRI sequences by direct replacement of linear gradient switching by MC switching (Figure 3). The amplitudes and timings of the 28 channels were set by a multichannel gradient waveform controller that was developed in-house.³⁰ The controller is modular with up to 8 channels per board and expandable up to 8 boards or 64 channels. For the present study 28 channels were controlled by 4 boards holding 7 channels per board. Any pulse sequence can be described by a number of magnetic field shapes applied as switching events. The effective magnetic field gradient comprising either individual x, y, or z gradient shapes or superpositions thereof were theoretically synthesized considering the calibrated coil basis information for the MC hardware at hand employing the B0DETOX software developed in-house.³¹ This was repeated for every gradient field of the MRI sequence, resulting in a matrix of current values comprising of 28 individual coil currents per gradient field times the number of gradient shapes (or events) constituting the MRI sequence. The control software of the gradient waveform controller operated in the console room sends the determined current values for all channels and all sequence events to local memory of the gradient waveform controller located in the hardware room via ethernet before the MRI scan is started. A series of sequence-controlled TTL pulses sent from the MR controller to the gradient waveform controller during the MRI scan sequentially strobes out ramped current settings for every channel in concert with RF pulses and delays, thereby assembling the MRI sequence at hand. Individual TTL signals hereby do not contain any amplitude information, but instead result in retrieval of the next set of 28 coil currents from local memory and immediate synchronous application to the amplifier hardware for the physical realization of the next gradient field.

FIGURE 3.

MRI pulse sequence conversion. (A) A conventional multislice GE MRI sequence based on linear x, y, and z magnetic field gradients and (B) its conversion to a MC-based DYNAMITE GE MRI sequence. The DYNAMITE sequence is described by events that are being advanced with TTL pulses. Each event holds the current values for 28 MC elements. The RF pulse, TE, and gradient strengths are identical between (A) and (B). DYNAMITE, dynamic multicoil technique; GE, gradient echo

DYNAMITE MR images were acquired with multislice gradient-echo (TR/TE = 1000/14 ms) and multislice spin-echo (TR/TE = 2500/50 ms) sequences as 128 × 128 matrices over 256 × 256 mm with sixteen 3-mm slices with 3340 Hz/cm (7.8 mT/m) slice selection, 794 Hz/cm (1.9 mT/m) phase encoding, and 1953 Hz/cm (4.6 mT/m) frequency encoding gradients, along with a 50 kHz acquisition bandwidth. Slice-specific z magnetic field gradients were optimized over a series of 15-mm slabs symmetrically surrounding the 3-mm slice at hand, that is, extending it by 6 mm on each side. All 16 z gradients were synthesized at 1% average error over a total range of 60 mm, calculated as mean(abs(target-synthesized))/max(abs(target)), resulting in the 48-mm FOV (= 16 slices × 3 mm slice thickness). In addition, DYNAMITE MRI was performed with a 3D gradient-echo sequence (TR/TE = 25/11 ms, 10° excitation) as a 256 × 256 × 96 matrix over 256 × 256 × 96 mm with 635 Hz/cm (1.5 mT/m) slab selection, 602 Hz/cm (1.4 mT/m) phase encoding, and 1953 Hz/cm (4.6 mT/m) frequency encoding gradients leading to an acquisition bandwidth of 50 kHz. The z gradient for slab selection was generated over a 96-mm range at 1% average error and combined with frequency-selective RF excitation for an effective image FOV of 96 mm. Similarly, z gradients for phase encoding along the z-direction were optimized over the targeted 96 mm FOV at 1% average error. For DYNAMITE MRI, the total number of events amounted to 18,432, 22,528 and 172,032 for multislice GE, multislice SE and 3D GE methods, respectively. All MRI sequences allowed selective replacement of DYNAMITE with conventional gradient encoding. MR images acquired with both approaches, therefore, shared the overall appearance including spatial geometry, image contrast, or RF behavior, providing an unobstructed view on aspects related to spatial selection and encoding, that is, the topic at hand. MRI acquired with conventional gradient technology were assumed to be free of artifacts and considered the gold standard in this research.

2.6 |. Data processing

Multislice and 3D-encoded MRI were reconstructed by 2D and 3D inverse Fourier transform, respectively. No zero-filling or filtering was applied, and the identical reconstruction was applied to all MRI irrespective of their origin, that is, whether they were based on encoding with DYNAMITE or conventional gradient technology.

Image distortion correction was implemented for 3D DYNAMITE MRI based on the measured or theoretical deviations from the ideal, linear magnetic fields similar to previous reports³² and applied to selected cases. In short, every pixel in undistorted space was encoded with the nonideal magnetic fields according to the exact sequence timings in order to calculate the position in distorted space. The signal intensity at the distorted position was then translated to the original, undistorted pixel position. State-of-the-art, for instance, nonlinear, transformations were not applied.

3 |. RESULTS

Conventional (Figure 4A) and DYNAMITE (Figure 4B) multislice spin-echo images acquired from the head of a single human subject are of comparable quality and sensitivity without any significant artifacts in the MC-based images. Multislice spin- (Figure 5A) and gradient-echo (Figure 5B) images from multiple human subjects show a high correspondence between conventional and DYNAMITE encoding in all cases without any obvious artifacts. Notably, some dark spots were observed in the periphery of all MRI, that is., both DYNAMITE and conventionally encoded, as a result of limited RF performance between stripline coil elements due to destructive interference.

FIGURE 4.

In vivo multislice spin-echo images of a single subject. Images (TR/TE = 2500/50 ms) acquired using (A) conventional and (B) DYNAMITE magnetic field gradients. Emphasis on the low amplitude range confirms similar fine structure and ghosting behavior in (C) conventional and (D) DYNAMITE MRI. None of the images are corrected for geometric image distortion. For display purposes, only 6 of 16 slices are shown

FIGURE 5.

In vivo multislice gradient and spin-echo images of multiple subjects. (A) Spin-echo (TR/TE = 2500/50 ms) and (B) GE (TR/TE = 1000/14 ms) images acquired using conventional (top) and DYNAMITE (bottom) magnetic field gradients. None of the images are corrected for geometric image distortion. For each subject, a single slice out of 16 slices is shown

However, some minor geometric deformations can be noticed primarily in the skull areas of the DYNAMITE MRI in Figures 4 and 5, that is, in the image periphery, compared to conventional spatial encoding with the system’s built-in gradient system. The attained DYNAMITE field accuracy was high for all gradient field shapes of all sequences with average errors per shape below 1% of the respective maximum apparent field amplitude in all cases. Upon closer inspection of higher amplitude x (602 Hz/cm) (Figure 6A) and y (1953 Hz/cm) (Figure 6B) gradients used in the 3D MRI sequence for phase encoding and frequency encoding, respectively, differences between the desired and actual magnetic field gradients are noticeable (Figure 6A,C). Moreover, these field imperfections are nonlinear and, although the overall gradient amplitude is accurate, regional imperfections are apparent that must be expected to result in suboptimal spatial encoding. The imperfections can be readily approximated by theoretical magnetic fields obtained through numerical integration of the Biot-Savart law for all MC elements (Figure 6B and D). The close agreement between experimental and theoretical magnetic fields indicates that the observed nonlinearity is due to an inherent inability of the 4 rows of 7 MC elements to generate a perfectly linear magnetic field. Whereas the nonlinearity will lead to image distortions, the effects are consistent and relatively small such that they can be expected to be readily removed through image distortion correction.

FIGURE 6.

Detailed analysis of MC field generation of selected linear field gradients employed for phase encoding (A/B, x gradient, 602 Hz/cm) and frequency encoding (C/D, y gradient, 1953 Hz/cm) for a 220 mm diameter spherical volume in a 240 × 240 mm² FOV. (A/C) The difference between the targeted ideal linear field gradient (left) and the experimentally measured linear field gradient (center) represents the apparent residual imperfection (right). (B/D) Difference between MC-based linear field gradients predicted by a theoretical evaluation of the magnetic field produced by each MC element and ideal linear field gradients. Note the 22- and 40-fold enhanced vertical scale in the difference maps compared to both target and MC fields in the x and y gradient analyses, respectively. Equivalence of theoretical and experimental imperfections validates the overall quality of the MC implementation and demonstrates that residual imperfections are not random but well understood, suggesting that the predicted (noise-free and spatially unlimited) theoretical field shapes can be employed for spatial unwarping in postprocessing

This hypothesis was tested in a resolution phantom and subsequently applied to the 3D GE MRI acquired from the in vivo human head (Figure 7). The DYNAMITE-based MRIs are high quality without major image artifacts and largely resemble the conventional MRIs. However, closer inspection reveals modest image distortions, particularly visible around the edge of the phantom (Figure 7C,D; red arrows) and human head (Figure 7I,J; red arrows). With a simple image distortion correction algorithm employing the spatial nonlinearity in the DYNAMITE encoding fields, as shown in Figure 6, the image distortions can be largely removed on phantoms (Figure 7E,F; green arrows) and human head (Figure 7K,L; green arrows). The apparent lower performance of distortion correction on human head is largely due to randomly segmented pixels in the skull suffering from lower SNR. The performance of image distortion-correction of brain-only is on par with the phantom data (87% and 96% correspondence of DYNAMITE and corrected DYNAMITE images with the conventional MRI, respectively). Note that the images of Figures 4 and 5 are not corrected for geometry distortions to emphasize the minor contribution of spatial misregistration during the DYNAMITE acquisitions. However, the residual image geometry distortions in the DYNAMITE images can be largely removed through image distortion correction, as detailed here.

FIGURE 7.

Geometric image distortion correction in vitro and in vivo. (A) Reference image and (B) signal mask acquired with conventional gradients as extracted from a 3D GE dataset. (C) Image and (D) mask obtained with DYNAMITE magnetic field gradients. Small geometric distortions primarily visible at the edge of the phantom (red arrows) lead to an 88% correspondence or overlap between the DYNAMITE and reference MRI pixel positions. (E) DYNAMITE-based image and (F) mask following geometric image distortion correction improves the overlap to 96% (green arrows). Human brain image and mask as extracted from a 3D GE dataset acquired with (G,H) conventional and (I-L) DYNAMITE-based gradients. Before and after geometric image distortion correction, the correspondence between conventional and DYNAMITE MRI was 84% (red arrows) and 93% (green arrows), respectively

Average duty cycles for the multislice GE, multislice SE, and 3D GE MRI sequences were 9.5%, 2.8%, and 6.9%, respectively, relative to an assumed fully dynamic range of ±5 A. Neither active air nor water cooling was applied, and coil temperatures measured from 6 representative coils rose significantly during the 1-, 7- and 10-minute scan durations, respectively, due to resistive heating. However, a maximum absolute coil temperature of 45°C was not exceeded at any time. Notably, there was no contact between study subjects and the MC hardware, thereby avoiding any exposure to the above temperatures.

4 |. DISCUSSION & CONCLUSION

Here, we have presented the first realization of DYNAMITE MRI of the in vivo human head with gradient fields that were purely MC-based. MRI sequences employing several 10 thousand up to 172,032 individual gradient events were successfully synthesized with single-layer MC hardware and played out in a well-timed fashion together with RF pulses and data acquisition to mimic and replace conventional gradient hardware. The DYNAMITE image fidelity was comparable to that of conventional MRI employing dedicated x, y, and z gradient coils.

All DYNAMITE-generated B₀ gradient shapes were of high accuracy (errors ≤ 1%); however, they were not perfect due to the limited number of MC channels employed in this first realization. The resultant image imperfections were minor yet noticeable. Moreover, they could be accurately predicted based on DYNAMITE magnetic field modeling both theoretically using the Biot-Savart law and experimentally employing the calibrated MC basis fields and, as such, corrected well by image postprocessing. MRIs acquired with the systems’ standard x, y, and z gradient system were assumed to be artifact-free in this study and defined as gold standard for geometry referencing. Such assumed idealized gradient performance is unlikely because some level of gradient nonlinearity is common on human MR systems. Similarly, the application of geometry correction is the rule rather the exception on clinical systems, mostly without nonexpert users even realizing, and the magnitude of those corrections is comparable to those applied here.^33,34 The realized DYNAMITE image fidelity has already reached that of conventional coil hardware and is expected to further improve with MC design improvements, employing more channels for enhanced B₀ field modeling capability.

Standard 1.5 to 3 Tesla clinical MR systems provide gradient amplitudes of 30 to 50 mT/m (12.8–21.3 kHz/cm), with dedicated research installations exceeding 100 mT/m.^35,36 Resulting clinical imaging bandwidths range from 50 kHz for fMRI applications to several hundred kilohertz for high-resolution structural MRI. The 7.8 mT/m (3.3 kHz/cm) slice selection and 4.6 mT/m (2.0 kHz/cm) frequency encoding gradients employed in this study were suitable for spatial encoding and MRI. The achieved 50 kHz acquisition bandwidth is respectable; yet, the employed amplitudes are below those typically applied for human MRI. It is important to remember, however, that B₀ field gradients in this study were based on an effective average maximum coil current of 1.4 A compared to several hundred amperes used in conventional gradient coils. A size-matched analysis of MC and conventional gradient hardware found that accuracy and efficiency levels for the generation of x, y, and z gradients over large regions of interest with MC field control are not comparable to dedicated SH wire patterns and, consequently, it seems unlikely that current MC technology has the potential to replace SH gradient coils on regular MR scanners.²¹ However, similar field accuracy and field generation efficiency was achieved with the MC approach when spatial encoding fields were optimized and applied in a region-specific fashion, for instance, for multislice MRI, and equivalent image quality is achieved throughout large 3D regions of interest. In this work, spatial selection and encoding across an almost 10 cm slab covering the human head has been realized with a DYNAMITE 3D gradient echo MRI sequence, demonstrating the suitability of the approach and implementation for field generation and MRI over large regions of interest.

Typical gradient rise times of clinical MR systems are in the order of a few hundred microseconds, that is, significantly shorter than the 1150 μs ramps employed in this study. The inductances of the MC hardware and a voltage rail of the employed power supply system of 72 V resulted in theoretical MC rise times of approximately 500 μs. For convenience and to match the (limited) available x, y, and z gradient performance, MC switching was performed with constant-time slopes at 1150 μs rise time. Reduced rise times will be enabled when the amplifier voltage rail is increased or the inductance of MC elements reduced. We recently presented an MC design tailored to DYNAMITE MRI of the human head in a head-only MR scanner.³⁷ With coil inductances close to 2 times lower than the MC hardware employed here, we expect to reduce the rise times to approximately 300 μs.

The MC array was attached with cable ties to the inside of a nylon tube, and this prototype implementation did not support any type of active coil cooling, for example, with forced air or water. The temperature rise during the MRI acquisition was significant yet tolerable in this proof-of-principle work. However, coil cooling purely based on passive convection is insufficient with applications relying on higher gradient duty cycle, and active temperature management will be necessary in the future. The next generation MC hardware will, therefore, be immersed in epoxy resin and equipped with water-based cooling circuits similar to conventional gradient technology. MC field control in this study was limited to the generation of B₀ field gradients for spatial encoding, that is, MRI, and no B₀ shimming was applied. We presented the first such MC design aimed at combined DYNAMITE MRI and B₀ shimming of the human head in a 1.5 Tesla head-only MR scanner recently, enabling high duty cycle applications including steady-state free precession MRI.^37,38 Whereas for the current study the controller unit has only been used to employ simple linear ramps, the controller is capable of full waveform control on all channels. This feature will become indispensable in future work that uses spatiotemporal encoding to counteract magnetic field inhomogeneity in addition to spatial encoding.^39,40 In this research, DYNAMITE was employed to resemble conventional MRI sequences that are based on standard linear B₀ field gradients and to demonstrate the feasibility of DYNAMITE MRI on human anatomy as a stepping stone toward clinical applications. DYNAMITE stands out by its ability to produce complex B₀ field shapes, thereby naturally lending itself to advanced applications employing B₀ shapes beyond basic linear ones, such as nonlinear spatial selection^20,41–43 or encoding.^23,44,45

The presented MC methodology sets the stage for a new field of research from advanced image encoding to consolidated hardware design.

Comparable image quality is a milestone toward establishing fully functional and purely MC-based MRI and B₀ shim systems. This work forms the basis for DYNAMITE-based image encoding as part of a BRAIN Initiative grant to establish a head-only 1.5 Tesla MR system suitable for non-invasive brain studies with minimal mobility restrictions.⁴⁶ This small size and lightweight MR system developed by a multi-institutional consortium of multidisciplinary researchers will be transportable for imaging brain function and structure in populations and environments almost anywhere. The positioning of only the human head inside the magnet bore puts extreme geometric and engineering constraints on the key system components, including the coil hardware for B₀ field management. Single-layer MC technology will therefore replace conventional multilayer SH coil systems and provide all B₀ field shapes for MRI and B₀ shimming at a total device thickness of only 22 mm.^37,38

Supplementary Material

Text S1

Text S1 Defintion file of coil wire path