A 16-Channel AC/DC Array Coil for Anesthetized Monkey Whole-Brain Imaging at 7T

Abstract

Functional magnetic resonance imaging (fMRI) in monkeys is important for bridging the gap between invasive animal brain studies and non-invasive human brain studies. To resolve the finer functional structure of the monkey brain, ultra-high-field (UHF) MR is essential, and high-performance, close-fitting RF receive coils are typically desired to fully leverage the intrinsic gains provided by UHF MRI. Moreover, static field (B0) inhomogeneity arising from the tissue susceptibility interface is more severe at UHF, presenting an obstacle to achieving high-resolution fMRI. B0 shim of the monkey head is challenging due to its smaller size and more complex sources of B0 offsets in multi-modal imaging tasks. In the present work, we have customized a coil array for lightly-anesthetized monkey fMRI in the 7T human scanner that combines RF and multi-coil (MC) B0 shim functionality (also referred to as AC/DC coils) to provide high imaging SNR and high-spatial-order, rapidly switchable B0-shim capability. Additional space was retained on the coil to render it compatible with monkey multi-modal imaging studies. Both MC global (whole-volume) and dynamic (slice-optimized) shim methods were tested and evaluated, and the benefits of MC shim for fMRI experiments was also studied. A minor reduction in RF coil performance was found after introducing additional B0 shim circuitry. However, the proposed RF coil provided higher image SNR and more uniform contrast compared to a commercially available coil for human knee imaging. Compared with static 2nd-order shim, the B0 inhomogeneity was reduced by 56.8%, and 95-percentile B0 offset was reduced to within 28.2 Hz through MC shim, versus 68.7 Hz with 2nd-order static shim. As a result, functional image quality could be improved, and brain activation can be better detected using the proposed AC/DC monkey coil.

Introduction:

Blood-oxygen-level-dependent (BOLD) functional magnetic resonance imaging (fMRI) of monkeys bridges the gap between non-invasive human research and invasive animal studies in the field of basic neuroscience (J. B. M. Goense et al., 2010; Kim and Ugurbil, 2003; Ku et al., 2011; Li et al., 2019; Logothetis et al., 1999; Tolias et al., 2005). Multimodal fMRI studies in monkeys have shown significant benefits in the exploration of the underlying mechanism of BOLD fMRI (Goense et al., 2012; Hu and Yacoub, 2012; Logothetis et al., 2001; Zaldivar et al., 2014), novel MR-based functional imaging modalities (Goelman et al., 2008; Goense and Logothetis, 2006; Vanduffel et al., 2001) and meso-scale functional connectivity of the brain (Chen et al., 2013; Roe et al., 2015; Teichert et al., 2010; Xu et al., 2019). Notably, fMRI studies in monkeys usually require the capacity to resolve fine functional cortical structures; therefore, a high image signal-to-noise ratio (SNR), functional contrast-to-noise ratio (CNR), and BOLD signal specificity are needed. The common way to achieve these factors is to increase the strength of the static magnetic field (Ciobanu et al., 2015; Duyn, 2012; Ugurbil, 2014). Several pioneering fMRI studies in monkeys have been conducted on small-bore animal ultra-high-field (UHF, >3T) MR scanners, but the small-bore size and reduced accessibility of animal scanners have limited the number of fMRI studies in monkeys. Nevertheless, the recent worldwide growth of installed 7T human MR scanners has led to several promising fMRI studies, including both anesthetized monkeys (Gilbert et al., 2018, 2016; Goelman et al., 2008; Xu et al., 2019; Zitella et al., 2015) and awake monkeys (Gao et al., 2019; Herrmann et al., 2015; Kolster et al., 2007).

While UHF MRI has been shown to be beneficial in fMRI studies in monkeys, realization of the full intrinsic gains provided by the higher static magnetic field depends on the performance of the radiofrequency (RF) transmit and receive coils (Gao et al., 2018; Guérin et al., 2017; Lattanzi et al., 2010; Ohliger et al., 2003; Wiesinger et al., 2004). Due to the lack of commercially available, form-fitting brain RF coils for monkeys, the single-loop receive (Rx) coils and human knee coils are often adopted in fMRI studies in monkeys (Chen et al., 2013; Logothetis et al., 2002). As a result, such non-specialized RF coils can result in suboptimal image SNR. Recently, several customized 7T RF coils for monkeys have been proposed, utilizing size-fitted and massive parallel Rx array designs, and therefore the image SNR can be significantly improved while maintaining large signal coverage. Meanwhile the parallel imaging performance of RF Rx arrays could also be improved (Keil and Wald, 2013), which is particularly beneficial for functional scans with high spatiotemporal resolution. In addition, Rx arrays with a dense element layout (Gao et al., 2019; Gilbert et al., 2018; Zhang et al., 2018) and/or optimized coverage (Gao et al., 2018) have provided significant SNR gains in UHF MRI. All these strategies in coil design for monkeys are promising for submillimeter functional imaging of monkey brains.

Regarding the transmit (Tx) aspect, shorter RF wavelength at UHF MRI makes it difficult to achieve uniform B1+ excitation and contrast, which is especially detrimental to fMRI studies because uniform functional response is usually assumed in wide brain regions in fMRI studies. Due to the smaller dimension of the monkey head compared to human head, conventional transmit coils including large single-loop (Kolster et al., 2007), quadrature-driven loop pairs and TEM antennas (J. Goense et al., 2010) have been successfully used at 7T to ensure whole-brain coverage for macaques; besides, multi-channel Tx coils with more freedom in B1+ management have demonstrated more homogeneous B1+ excitation and uniform contrast in both cortical and deep brain areas such as the cerebellum and brainstem (Adriany et al., 2010; Gilbert et al., 2015; Zitella et al., 2015). However, quadrature-driven birdcage Tx designs that are known to provide good B1+ homogeneity for brain imaging at 3T have not yet been fully explored for monkey brain imaging on a 7T platform.

Compared to 3T, UHF fMRI has the disadvantage of increased subject-specific B0 inhomogeneity (ΔB0) caused by air-tissue magnetic susceptibility interfaces (De Graaf and Juchem, 2016; Stockmann and Wald, 2018). Echo-planar imaging (EPI), which is the most common acquisition strategy for fMRI, is negatively impacted by geometric distortion and blurring caused by phase accumulation down the echo train due to B0 offsets (Stehling et al., 1991). The amount of distortion depends on the echo spacing and parallel imaging undersampling factor, but in practice, limitations on gradient coil performance and achievable undersampling factors prevent the distortion from being completely eliminated. Thus, larger B0 offsets at 7T potentially lead to a more severe voxel shift or image distortion. In addition, B0 inhomogeneity causes intra-voxel dephasing in gradient-echo-based sequences which usually have the highest BOLD sensitivity (Bandettini et al., 1994), leading to signal dropout and reduced T2* in voxels with severe intra-voxel B0 gradients. The segmented EPI acquisition strategy (Wang et al., 2019) is able to reduce image distortion and signal dropout thanks to shortened minimum achievable echo spacing (Hoogenraad et al., 2000); however, it suffers from low temporal resolution and severer physiology noise due to B0 variation between shots.

Among these adverse susceptibility effects, image distortion can be partially corrected by using unwarping techniques based on point spread function mapping (Constable and Zeng, 2002) or the acquisition of spin echo EPI reference images with opposite phase-encoding directions (e.g., TOPUP method (Vickers et al., 2004)). However, these methods have limited efficacy in areas with severe voxel shift, leading to unwanted signal changes in the local voxels and/or image blurring. Other susceptibility effects including signal loss, T2* and BOLD contrast change that are detrimental to fMRI applications are difficult to be corrected or compensated during post-processing. Finally, motion can induce real-time local B0 fluctuations that contribute to temporal noise in functional imaging acquisition and are difficult to eliminate through routine motion correction methods (Logothetis et al., 2007; Moortele et al., 2002), though recently, some progress has been made on this topic (Dymerska et al., 2018).

The most efficient and direct way to eliminate B0 inhomogeneity and the susceptibility effect is through B0 shim, i.e., generating complementary B0 field offsets by using external hardware to null the unwanted ΔB0 fields in the body. The passive shim (Wilson et al., 2002) using magnetic materials (e.g., graphite) is rarely adopted due to the difficulty and complexity in subject-specific configuration of their spatial distribution. In contrast, active and spherical harmonic (SH) shim design has been used in commercial human MR scanners, with each SH coil generating an independent B0 field using specific coil windings. Typically, human MR scanners are equipped with 1^st order (the linear encoding gradients), 2nd-order, and (in some cases) 3rd-order shim terms. Specialized SH insert coils which provide higher-order terms up to 6^th order (Schwerter et al., 2019) are available, but this approach has only been implemented at a limited number of sites. Using the conventional 1st- and 2nd-order SH fields, high-spatial-order ΔB0 fields cannot be adequately nulled in the human brain, particularly in the frontal and temporal lobes. For monkeys, the need for higher-order shim is even more pronounced due to their smaller head size and the severe B0 offsets that arise near regions with brain surgery or implantation in the head.

Multi-coil (MC) shim arrays, which consist of independently driven DC current loops designed either as stand-alone shim arrays (Hsu and Glover, 2005; Juchem et al., 2011, 2010) or integrated RF-B0 arrays (referred to as AC/DC here) (Han et al., 2013; Stockmann, 2013), offer an efficient and practical solution for generating a higher-order B0 field basis set. Because MC shim loops can be placed close to the subject, the number of wire turns in each MC loop and thus their inductance can be reduced compared to conventional SH shim coils. As a result, the low inductance of MC loops allows them to be driven by low-cost and low-voltage current amplifiers that support dynamic current updates during acquisition. AC/DC coil designs potentially save space by integrating RF receiver and DC shim circuitry, thus permitting both high SNR and high shim efficiency. The space savings are favorable for fMRI studies in monkeys with a multi-modal apparatus that must be mounted during scans.

In the present study, we have designed and built a 16-channel AC/DC Rx array equipped with a birdcage Tx coil for signal excitation. Optimal Rx coil geometry and a dense array design were adopted to fully exploit the intrinsic gain under UHF MRI. We have systematically evaluated the coil performance for in vivo monkey brain anatomical and functional imaging at 7T, the interference of B0 shim hardware, and the effectiveness of the proposed MC shim array in mitigating nuisance susceptibility effects.

Methods:

AC/DC coil

RF (AC) Tx coil design

A quadrature-driven, detunable, 16-rung band-pass birdcage Tx-only coil was constructed. A rung length of 240 mm was chosen to ensure a relatively uniform excitation within the macaque brain. The inner diameter of 290 mm provided sufficient space for mounting multi-modal devices and provided adequate space for routing DC shim supply wiring to the AC/DC loops. The birdcage Tx coil was first tuned to 297.2 MHz without a matching circuit by using a standard double-probe method, and two driven ports spaced 90° azimuthally were capacitively matched to 50 ohms and connected to a quadrature hybrid circuit.

RF (AC) Rx coil array design

To maximize the SNR and coil sensitivity distinction in cortical areas, a helmet-like conformal Rx coil array geometry was designed based on a 3D macaque head model. The model was extracted from MPRAGE T1-weighted images collected over a 4 kg female macaque using the software 3D Slicer (Pieper et al., 2005). The coverage of the coil former was designed based on recent electromagnetic simulations (Gao et al., 2018). As shown in Fig. 1a–b and Supplementary Fig. 1, the top of the head was covered as much as possible, with particularly dense coverage of the back of the head for improved SNR and parallel imaging acceleration performance in the occipital cortex, cerebellum, and brainstem. The form of the helmet was also designed to integrate the stereotactic frame, with ear-bar installation on both sides of the head. Four circularly shaped openings (2-3 cm diameter) were left in the helmet to allow intervention of the multi-modal apparatus (e.g., laser optic fiber (Xu et al., 2019)) into brain regions such as the temporal, parietal, and occipital lobes.

Fig. 1.

Coil configurations: (a) coil holder schematic; (b) coil element layout, and grey elements are capacitive decoupled; (c) assembled AC/DC coil array element; (d) assembled transmit coil, AC/DC coil array, and stereotactic frame, and (e) the circuitry of integrated AC/DC coil array element. More conceptional views of the coil design are shown in supplementary Fig. 1.

The coil loops were laid out using 16 AWG wire on a polycarbonate 3D-printed form. Considering a smaller loop diameter usually provides elevated SNR in surface regions, sample-noise-dominated 5 cm-diameter loops were chosen as the building blocks of the Rx array. For tuning, 2~3 segments with distributed discrete capacitors were used to minimize transmission-line effects in the loop. In total, 16 loops were used to spread over the helmet surface, mimicking a soccer ball pattern. Inter-element loop overlap for inductive decoupling was pre-determined using two test loops of a similar size. The optimal gap was chosen by finding the lowest S21 using bench measurement on a vector network analyzer. To leave space for ear-bar installation, two pairs of loops on the left (#5 and #7) and right (#12, #15) sides were decoupled using shared tunable decoupling capacitors instead of inductive decoupling.

Each loop element was tuned and matched before being connected to the preamplifier (Siemens Healthcare, Erlangen, Germany) with 75 ohm characteristic impedance through a 100 mm-long, semi-rigid co-axial cable (UT-85C-FORM, Micro-coax, PA, USA). The blocking efficiency of a PIN-diode-based active detuning circuit was measured using a double probe. The preamplifier daughter board was custom-designed with an integrated DC power pathway, active-detuning pathway, preamplifier matching and decoupling circuit, and input-protection circuit. In addition, four custom-built cable traps were connected at the output of preamplifiers for channels #1, #6, #13, and #14 to minimize common-mode cable currents and interactions with the Tx coil.

B0-shim (DC) hardware

Self-shielded, hand-wound toroid inductors (22 AWG insulated wires, O.D. 12.8 mm, and I.D. 2.7 mm) with the appropriate self-resonant frequency (~320 MHz) and inductance (~340 nH) were used as RF chokes bypassing distributed capacitors on each RF loop. RF interactions of the DC pathway were evaluated on a test single-loop coil (resonating at 297 MHz without matching circuitry) using a double-probe bench measurement (quality factor before and after adding chokes). The contribution of coil loss (Loss_coil) and sample noise (Loss_sample) in a resonant circuit was evaluated with the Q-ratio (quality factor ratio), defined as Q_unload/Q_load = (Loss_sample+Loss_coil)/Loss_coil.

Open-source, low-cost DC current amplifier boards were used to supply a multi-channel B0-shim current (Arango et al., 2016). Less than 30 us was required to update each channel when driving shim coil loads with inductances of approximately 10 uH. The maximum current that could be continuously applied in the DC shim pathway was 2.5 A without the chokes heating to above 40ºC. A microcontroller was configured to dynamically update shim currents during pulse sequences using a fiberoptic trigger from the scanner at the beginning of each TR.

Animal Preparation

Two female macaque subjects (macaque M1: 2 years, 3.5 kg and M2: 3 years, 4.5 kg) were used in the study. All procedures were conducted in accordance with NIH standards and approved by the local Institutional Animal Care Committee. M1 was a normal healthy subject, whereas M2 underwent head surgery and had an optical chamber implanted in the skull over the occipital cortex. Both monkeys were anesthetized with sufentanil administered at 0.04 ml/kg*hr intravenously and isoflurane 0.3%-1.5% before being placed in the stereotaxic frame with the head centered within the coil. Each animal’s physiological state including the heart rate, expired CO₂, oxygen saturation, respiration, and temperature were continuously monitored during the scan. RF coil performance measurements and anatomical image acquisition were conducted only on M1. MC global and slice-optimized B0 shim performance was evaluated on both monkeys. Task BOLD fMRI experiments were conducted only on M2.

Imaging Protocols

Images were acquired on a 7T research scanner (MAGNETOM 7T, Siemens Healthcare, Erlangen, Germany) with a whole-body gradient set (70 mT/m and 200 T/m/s). A 28-channel QED knee coil (Mayfield Village, OH, USA) with 155 mm inner diameter was selected as a reference coil for comparison.

The 3D actual flip angle imaging (AFI-B1) maps (Yarnykh, 2007) (TR1/TR2/TE/α: 20 ms/50 ms/3.45 ms/60°, matrix: 64 × 64, and FOV: 128 × 128 mm²) were acquired to evaluate the transmit efficiency and homogeneity for both coil setups. The normalized root-mean-square error (RMSE, σ_B1⁺) was calculated over the brain region to evaluate the homogeneity of excitation. The Tx voltage for each RF coil was calibrated according to the AFI-B1 for each imaging session.

Spatial SNR was measured on both in-vivo monkey and phantom using root-sum-of-square (rsos) combined proton-density-weighted GRE images (TR/TE/α: 30 ms/6 ms/10°, slice thickness: 3 mm, matrix: 96 × 96, FOV: 96 × 96 mm², and bandwidth: 200 Hz/pixel), and noise correlation information, which was acquired using the same sequence but with no RF excitation(Kellman and McVeigh, 2005; Roemer et al., 1990).

When parallel image acceleration is used, the image SNR was reduced by a factor of $\mathrm{g}\sqrt{\mathrm{R}}$, where g is the g-factor (noise amplification factor) and R is the acceleration factor. The g-factor maps were computed based on the differences in noise level with and without acceleration so as to assess noise amplification in sensitivity-encoding (SENSE) reconstructions of the accelerated images (Pruessmann et al., 1999). Image FOV (96 × 96 mm²) was chosen to tightly fit the head size for evaluation in the worst-case scenario.

High-spatial-resolution structural images were acquired on macaque M1 using a T2*-weighted GRE sequence (TR/TE/α: 2000 ms/25 ms/60°, matrix: 464 × 512, FOV: 92 × 102 mm², slice thickness: 1 mm, GRAPPA acceleration: 2, bandwidth: 30 Hz/px , average: 1, and scan time: 8’7’’), T2-SPACE (TR/TE/α: 2500 ms/140 ms, matrix: 248 × 320, FOV: 74 × 96 mm², slice thickness: 0.3 mm, GRAPPA acceleration: 2, bandwidth: 504 Hz/px, averages: 2 and scan time: 20’40”), and T1-weighted MPRAGE images (TR/TE/TI/α: 2590 ms/2.82 ms/1050 ms/7°, matrix: 200 × 256, FOV: 80 × 102 mm², slice thickness: 0.4 mm, GRAPPA: acceleration 2, bandwidth: 260 Hz/px, averages: 4, and scan time: 19’14’’).

The B0 field map of each DC shim element was acquired on a balloon phantom filled with saline gelatin (Zhang et al., 2014) using a double-echo gradient echo sequence (TR/TE1/TE2/α: 429 ms/3.29 ms/4.31 ms/65°, matrix: 138 × 104 × 35, FOV: 138 × 104 × 70 mm³, and slice thickness: 2 mm). To measure the B0 inhomogeneity in the brain, B0 field maps were also acquired for each monkey subject with different shim settings applied, and the RMSEs over the entire brain and within slices were quantitively calculated respectively.

The MC shim, including global and slice-optimized dynamic shim, was used for whole-brain imaging and was conducted based on the scanner 2^nd order shim. For global shim, whole-brain volume was chosen for shim current optimization; in contrast, dynamic shim minimizes B0-inhomogeneity on a slice-specific basis, and additionally triggers were added in imaging sequence to synchronize and update shim current for each slice. The MATLAB function ‘fmincon’ was used to find the optimal shim current for uniform B0 within target region. The shim current for each channel was subject to a conservative limit of 2A. To increase the robustness of dynamic shim and prevent intra-voxel dephasing caused by B0 gradients along the z-direction, two neighboring slices were included when calculating the shim current for a slice of interest. To evaluate the efficacy of each B0 shim, single-shot EPI (TR/TE/α: 2100 ms/27 ms, matrix: 104 × 138, FOV: 104 × 138 mm², slice thickness: 2 mm, GRAPPA acceleration: 2, echo spacing: 0.79 ms, and bandwidth: 1450 Hz/px, with fat saturation) with added shim triggers were acquired; two triggers were inserted before the RF excitation pulse and after the gradient readout to turn on the shim current during the RF transmit and signal readout and to switch off the current during the fat saturation pulse, respectively. Detailed explanation about the efficacy of shim settings described above is presented in Discussion.

The feasibility and efficacy of the MC shim in BOLD signal detection was first investigated by quantitatively evaluating the tSNR of functional EPI acquisitions. To evaluate the efficacy of MC shim in reducing the nuisance susceptibility effects in the fMRI studies, EPI images using different shim settings were acquired and compared. For a better illustration of the susceptibility effects, GRE images with a short TE (4.3 ms) were acquired to provide anatomic reference images showing the full outline of the brain without apparent signal losses or image distortions, and brain masks were extracted from these images accordingly. To evaluate the benefit of MC shim for fMRI, temporal stability was assessed by computing the temporal SNR (tSNR, mean divided by standard deviation of temporal signal for each voxel) for a time-series of EPI images (TR/TE/α: 2000 ms/27 ms/60°, matrix: 104 × 138, FOV: 104 × 138 mm², slice thickness: 2 mm, GRAPPA acceleration: 2, echo spacing: 0.79 ms, bandwidth: 1450 Hz/px, without fat saturation, and measurements: 70).

Visual-task fMRI was conducted for 2nd-order SH shim and global MC shim using single-shot EPI scans (TR/TE/α: 2000 ms/25 ms/70°, matrix size: 72 × 96, FOV: 72 × 96 mm², slice thickness: 1 mm, GRAPPA acceleration: 2, echo spacing: 0.81 ms, bandwidth: 1488 Hz/px, and measurements: 180). For each block, a black-and-white flickering (1 Hz) checkerboard stimulus was presented for 40 s and then turned off for 20 s, with 6 blocks repeated in total. To minimize the potential confounding factors such as variations in anesthesia level and blood oxygen level, two neighboring runs were performed with 2nd-order shim and global shim, respectively. The post-processing including slice-timing correction, motion correction, volume registration, blurring (kernel size 1) and GLM analysis were performed using AFNI (Cox, 1996) with default settings, and only voxels with p-value < 0.001 were selected to present in the overlaid activation map.

Results

For the test loops, RF chokes caused an average 10.9% (2-segment loop) and 2.5% (3-segment loop) Q-ratio reduction. For all the receive elements, 30 dB active detuning blocking was achieved. Coil decoupling between neighboring elements ranged between −12 dB and −29 dB. For capacitive decoupling, −13 dB and −18 dB were achieved for loop pairs on the left and right side, respectively.

Fig. 2 shows a comparison of RF coil performance in macaque imaging. The addition of RF chokes to enable the DC shim pathway had a moderate nuisance effect on the coil SNR but a negligible effect on inter-element isolation. The proposed 16-channel coil for monkeys showed higher SNR than the 28-channel knee coil, likely attributable to its better filling factor due to its conformal design, which minimized coil-to-subject distance. For the Tx coil, the birdcage coil for monkeys showed reduced B1+ efficiency but slightly higher homogeneity compared to the knee coil.

Fig. 2.

SNR, inter-element coil coupling, and transmit fields are compared among the 16-channel array coil for monkeys, the 16-channel AC/DC array coil for monkeys, and the commercial knee coil. The diameters of the Tx coils are 290 mm (coil for monkeys) and 210 mm (knee coil), and the diameters of the Rx coils are 110 mm (coil for monkeys) and 155 mm (knee coil). The normalized RMSE (σ_B1⁺) of B1+ distribution within the brain region was calculated for each coil. A moderate influence on the SNR, inter-element coupling, transmit coil efficiency, and homogeneity is observed after introducing the DC shim pathway.

Fig. 3 shows the inverse g-factors for the 16-channel AC/DC coil and 28-channel knee coil for a variety of undersampling factors in the anterior-posterior (A-P), right-left (R-L), and head-foot (H-F) directions. At high acceleration factors, the 16-channel coil shows less noise amplification, particularly for the R-L and H-F undersampling directions. Up to an acceleration rate of 3, little noise amplification was introduced by the monkey coil.

Fig. 3.

Inverse g-factors produced by the proposed 16-channel AC/DC coil array and commercial knee coil in three phase-encoding directions with acceleration rates of 3 and 4.

Considering some Rx elements in knee coil was found not in optimal condition, for the data shown in Fig. 2–3, we provided additional SNR and inverse g-factor comparison results on phantom with fully functional knee coil and the 16-channel monkey coil, as shown in Supplementary Fig. 2–3. The monkey coil showed higher SNR in most part of the phantom than the knee coil, which corresponding to the monkey brain region. The inverse g-factor results for both coils acquired on phantom is similar to the results acquired on in-vivo moneky.

As depicted in Fig. 4, all high-resolution structural scans were completed within ~20 minutes with reasonably high quality. Due to the relatively homogeneous excitation provided by the large-diameter birdcage Tx coil, uniform contrast was achieved in T1, T2, and T2* structure images over most parts of the brain, including the cerebellum and brainstem. In addition, in the GRE images, intra-cortical vessels, and the stripe of Gennari at layer IV of the primary visual cortex could be easily found with 200 um in-plane resolution. On the T2-SPACE images, details in the cerebellum and brainstem were clearly shown with 300 um isotropic resolution. The MPRAGE images showed uniform contrast between the grey and white matter over the entire brain, offering better raw data for brain region segmentation. For the T2-SPACE and MPRAGE images, some loss of contrast was seen in the right temporal lobe and cerebellum due to B1+ variations.

Fig. 4.

High-resolution MR microscopy images: 200 μm in-plane resolution GRE, scan time: 8’7’’, (TR/TE/α: 2000 ms/25 ms/60°, matrix size: 464 × 512, FOV: 92 × 102 mm², slice thickness: 1 mm , GRAPPA R: 2, bandwidth: 30 Hz/px , and average: 1); 300 μm isotropic resolution T2-SPACE, scan time: 20’40’’ (TR/TE/α: 2500 ms/140 ms, matrix size: 320*240, GRAPPA R: 2, bandwidth: 260Hz/px, averages: 2); and 400 μm isotropic resolution MPRAGE, scan time: 19’14’’ (TR/TE/TI/α: 2590 ms/2.82 ms/1050 ms/7°, matrix size: 320*248, GRAPPA R: 2, bandwidth: 504 Hz/px, averages: 4). Uniform contrast can be achieved over the whole brain region for T1, T2, and T2* structure scans.

As shown in Fig. 5, B0 inhomogeneity (RMSE) over the whole brain region was reduced by 14% (M1) and 12.8% (M2) using MC global shim compared to static 2nd-order SH shim. Dynamic slice-optimized MC shim reduced the B0 RMSE by 56.8% (M1) and 34.1% (M2). Compared with 2nd-order static shim, the percentage of voxels with less B0 offset were 83.7% (M1) and 79.8% (M2) with the MC global shim, and 80.4% (M1) and 77.6% (M2) through MC dynamic shim. In addition, voxels with a large B0 shift, which contributed to image distortions, were significantly reduced by using the MC shim method. The largest B0 offset, measured as a 95% residual in the B0 distribution across all voxels, could be reduced from 68.7 Hz (M1) and 160.1 Hz (M2) with static 2nd-order shim, to 57.5 Hz (M1) and 131.6 Hz (M2) after applying MC global shim, and 28.2 Hz (M1) and 80.4 Hz (M2) through MC dynamic shim, respectively. Although significant improvements were seen in both M1 and M2 over many brain areas, the most severe B0 offsets near the head implantation in M2 could not be fully mitigated.

Fig. 5.

Experimental comparison of shim performance between 2^nd-order shim, MC global shim, and MC dynamic shim. The upper side shows the ΔB₀ field maps acquired on two monkey subjects (M1 and M2). M1 is a healthy subject, and M2 underwent optical chamber implantation after brain surgery. The standard deviation of the B₀ distribution was calculated over the whole shimmed volume. The lower side shows histograms of the residual ΔB₀ over the brain volume.

There was good agreement between the predicted field maps and acquired field maps within subjects, as suggested in Fig. 6. Obvious signal dropout at the posterior aspect of the brain for both monkey subjects could be seen in the EPI images with the scanner’s 2nd-order shims applied. After applying the MC dynamic shim, the signal loss was significantly reduced, and the brain region in the EPI images was aligned more closely with the brain mask of the GRE anatomic reference. However, signal loss near the head implantation in M2 was not fully recovered, even with MC shim due to the extreme B0 variations near the implant. Nevertheless, EPI image distortion was significantly improved for both monkey subjects. Fewer image voxels appeared outside the brain mask and less voxel offset under opposite phase-encoding directions could be observed.

Fig. 6.

Acquired EPI slices show that MC dynamic shim reduces geometric distortion and signal dropout (EPI: 1 mm in-plane, 2 mm slices, TR/TE: 2100 ms/27 ms, GRAPPA R=2, and echo spacing: 0.79 ms). GRE structure reference images in the upper row with short TE (4.3 ms) are intended to show the signal dropout-free and distortion-free references, respectively. The middle rows show the ΔB₀ distribution over the selected brain slices under each shim condition. The lower rows show the corresponding EPI slices.

As shown in Fig. 7, the tSNR of the EPI time series with both MC global shim and MC dynamic shim was comparable with that under 2nd-order shim, but a more uniform distribution of tSNR could be found on the time series with MC dynamic shim. MC shim also demonstrated its benefit in assisting in the detection of BOLD signals. In visual-task fMRI experiments performed on M2, BOLD signals at the same visual cortical region were better detected with a higher statistical significance after applying MC global shim, as illustrated in Fig. 8 and Supplementary Fig. 4.

Fig. 7.

Dynamic MC shim provides improved temporal SNR over much of the cortex acquired for subject M1 (EPI: 1 mm in-plane, 2 mm slices, TR/TE: 2000 ms/27 ms, GRAPPA R = 2, echo spacing: 0.79 ms, and measurements: 70). Elevated tSNR is attributed to improved T2* over most of the shimmed voxels. The tSNR maps also suggest that no significant instability was introduced by the rapid switching of the shim currents between the acquired slices.

Fig. 8.

MC global shim improves BOLD activation maps in visual task fMRI on subject M1. Maps were generated from the EPI time series using a general linear model. Voxels with p-value < 0.001 are presented.

Discussion

RF Coil Design

Compared to a commercially available knee coil with cylindrical geometry and larger Rx channel count, the proposed head coil for monkeys exhibits a better performance in SNR and parallel imaging acceleration performance. It suggests the importance of receive coil array design in coverage and geometry: although the knee coil was designed with more Rx channels (28 vs. 16), its Rx element size may be larger than that of the proposed monkey coil array due to, e.g., the bigger former dimension and a larger loop array coverage for the cylindrical knee coil; therefore, the Rx coil sensitivity profiles of the knee coil within monkey head are not as distinctive as the proposed monkey coil array with a close-fitting helmet design (Gao et al., 2018). By using the proposed AC/DC coil, regular high-resolution structural scans, e.g., GRE (T2*), SPACE (T2), and MPRAGE (T1), could be completed within 20 minutes, thus benefiting anesthetized macaque experiments with limited scan time controlled within a safe period of anesthesia.

Due to its larger diameter (290mm) and smaller filling factor, the customized birdcage transmit coil showed lower efficiency than the knee coil with an inner diameter of 210 mm. In the meantime, judged by acquired T1-MPRAGE and T2-SPACE anatomical images (see Fig. 4) that are both sensitivity to B1+ homogeneity, the custom-built birdcage coil demonstrated a satisfactory B1+ performance in achieving a relatively uniform excitation within the macaque brain.

The transmit homogeneity was sufficient to produce a uniform white-to-gray matter contrast over most of the monkey brain, including the cerebellum, thalamus, and brainstem. Therefore, the coil is suitable for studying top-down functional connectivity, as well as feed-back and feed-forward networks between the central neural system and peripheral neural system.

Interactions between the RF and DC shim circuitry was systematically investigated using bench measurement for MR image metrics. Only minor signal loss was found in coil quality factor measurements and SNR maps. No adverse effect was found according to the inter-element correlation matrix. The twisted-pair shim wiring and self-shielded toroids were designed to minimize Lorentz forces during shim updating. The tSNR maps demonstrated that any remaining torques on these components are not large enough to impact the EPI image quality. The noise floor in the EPI images was also unchanged when the shim amplifiers were powered on, demonstrating that the amplifiers did not create significant RF noise inside the scanner room.

The mechanical design of the proposed head coil for monkeys considered the integration of a stereotactic frame for head fixation; therefore, the monkey head installation procedure could be simplified, improving efficiency and convenience. While maintaining high SNR and parallel imaging capability, the use of openings in the coil housing further enabled simultaneous neuronal recording, stimulation, intervention, and other effects, which are difficult to implement using existing commercially available RF coils at 7T.

MC Shim

The acquired field maps showed good agreement with the predicted field maps, suggesting a reliable setup of the MC B0 shim system. The proposed 16-channel (5 cm loops) whole-head AC/DC array for monkeys can effectively mitigate B0 inhomogeneity on normal monkey subjects, showing a 14% (global shim) and 56.8% (dynamic shim) reduction in B0 inhomogeneity, respectively. The dynamic shim performance is similar to that of a previously reported human 7T 31-channel AC/DC array, which showed a 56.3% reduction in B0 inhomogeneity over the whole brain (Stockmann, 2013). Less improvement in B0 homogeneity was achieved on the monkey subject (M2) with the head implantation, which we attribute to the highly localized B0 offset in the vicinity of the optical chamber that exceeded the shim capability of our 16-channel array. However, even in this challenging region, MC shim provided some gains in B0 homogeneity and EPI distortion.

In this study, MC global shim showed less efficacy in reducing B0 offset compared with MC dynamic shim, because the associated linear optimization problem in global shim is less determinant than that in dynamic shim. However, because the global-shim experiment setting is easier than that in dynamic shim, one can alternatively add additional smaller shim loops to provide a higher-spatial-order shim capability, even under the global shim setting, facilitating a wider range of applications of the proposed AC/DC array coil.

For dynamic shim in both EPI and field mapping, the time interval between triggers was at the minimum scale of millisecond, which allowed sufficient time for microcontroller to issue update commands to all 16 channels and to allow the switching transients to settle (Arango et al., 2016). Dynamic shim is also compatible with SMS (simultaneous multi-slice) acquisitions (Feinberg and Setsompop, 2013). Simulation studies have shown that MC dynamic shim still works well under an SMS acceleration factor of 2, even though the performance wanes and approaches global shim as the SMS acceleration factor increases (Stockmann and Wald, 2018), and aspects causing such effect remain to be carefully examined to advance the functionality of dynamic shim.

First, because each set of dynamic shim currents is optimized to ensure a uniform B0 distribution within a single slice/slab rather than the whole-brain volume, if a narrow-band saturation pulse is applied (such as fat saturation), sharply varying shim field offsets outside the target shimmed slices can shift brain voxels into the saturation band of the pulse, thereby causing a loss of contrast due to spin history effects. Therefore, for dynamic shim experiments, an additional trigger should be placed before the saturation pulse to inform the micro-controller to turn off the dynamic shim current (or to apply a global shim instead) in a timely manner. A second trigger after the fat saturation pulse then enables slice-optimized shims prior to the RF excitation and spatial-encoding modules of the sequence.

Second, since the dynamic shim minimizes B0 inhomogeneity over each single slice rather than a 3D volume, due to imperfect slice profile that is inherit to a 2D sequence, B0 field measurement may not be accurate in each 2D slice, and therefore B0 gradient may occur across slices after applying dynamic shim current, leading to intra-voxel dephasing and unwanted signal loss, particularly when a thin slab of interest is chosen. Therefore, in the present study, two neighboring slices were included in the dynamic shim volume when computing the optimal shim current. After updating dynamic shim settings each time, as shown in Fig. 5, no apparent through-slice B0 gradients was seen in adjacent sagittal or coronal field maps under dynamic shim, nor any obvious signal loss due to through-slice dephasing was found in the EPI images.

In the present study, we showed the efficacy of MC shim for BOLD fMRI studies. The EPI time series maintained good temporal stability with either global or dynamic MC shim applied, suggesting that the coil and shim amplifiers are stable and have no apparent adverse impact on MR image quality. Moreover, as illustrated in Fig. 6, on both monkey subjects, MC shim reduces signal dropouts induced by intra-voxel dephasing, which is particularly detrimental for GRE-EPI scans used for BOLD fMRI studies at UHF, because the resulting signal voids cannot be recovered with post-processing methods. Furthermore, a longer TE is helpful for enhancing BOLD contrast through accumulating signal dephasing caused by microscopic B0 field changes, and such local B0 disturbance is usually induced by deoxygenated hemoglobin in the blood. But, to ensure less signal loss caused by macroscopic B0 offset and T2 relaxation, a shorter TE (~20 ms) is usually adopted instead. Applying MC shim can potentially make a longer TE feasible and therefore achieve higher BOLD contrast. In this study, we have verified that voxels with significantly higher BOLD activation could be detected in the visual cortical region through applying MC global shim. The elevated BOLD activation is attributable to reduced geometric distortion, recovery of signal loss, and the lessening of nuisance changes in microscopic T2* due to overlaying macroscopic B0 variations. Lastly, better registration of functional and anatomical images can be performed if EPI image distortion is minimized using MC shim, permitting more accurate BOLD fMRI analysis. It should be noted that, the dynamic shim capability of the proposed monkey AC/DC coil is also promising in correcting real-time B0 fluctuations due to respiration, and it would be especially helpful in reducing physiology noise in segmented EPI which is highly sensitive to shot-to-shot B0 variation – to achieve this goal, real-time and accurate measurement of B0 changes requires further investigation.