32-channel RF coil optimized for brain and cervical spinal cord at 3T

Abstract

Diffusion and functional MRI of the spinal cord remain challenging due to the small cross-sectional size of the cord and susceptibility-related distortions. Although partially addressable through parallel imaging, few highly parallel array coils have been implemented for the cervical cord. Here we developed a 32-channel coil that fully covers the brain and c-spine and characterized its performance in comparison with a commercially available head/neck/spine array. Image SNR and temporal SNR were respectively increased by 2× and 1.8× in the cervical cord. Averaged g-factors at 4× acceleration were lowered by 22% in the brain and by 39% in the spinal cord, enabling 1 mm isotropic R=4 Multi-Echo Magnetization Prepared Gradient Echo (MEMPRAGE) of the full brain and c-spine in 3:20min. Diffusion imaging of the cord at 0.6×0.6×5 mm³ resolution and tractography of the full brain and c-spine at 1.7 mm isotropic resolution were feasible without noticeable distortion. Improvements of this nature potentially enhance numerous basic and clinical research studies focused on spinal and supraspinal regions.

Introduction

Magnetic resonance imaging (MRI) is routinely used for diagnosing pathologies in the brain and spinal cord, such as tumors or multiple sclerosis lesions. However, the ability to detect subtle pathological features is limited by the image contrast as well as the spatial resolution of current clinical systems.

There is growing interest in diffusion-weighted (DW) MRI due to its specificity to white matter structure, providing quantitative markers in neurodegenerative diseases (1). However, DW-MRI is particularly challenging in the spinal cord due to the small cross-sectional size of the cord (~1 cm), which requires high spatial resolution. Furthermore, susceptibility artifacts nearby intervertebral disks create image distortion when using echo-planar imaging (EPI) readout (2).

Functional MRI (fMRI) enables detection of neuronal activity via blood oxygenation level dependent (BOLD) contrast (3). Numerous clinical and neuroscience applications could benefit from non-invasive examination of brain and spinal cord functions, such as the evaluation of the sensorimotor changes following acute trauma, tumor or neurodegenerative disease. Although widely applied to the brain, fMRI of the spinal cord remains challenging for many of the same reasons hampering spinal DW-MRI (4,5). In addition, a recent study reported low temporal SNR (tSNR) values in the spinal cord compared to the brain (6). In contrast to the SNR of a single image, the tSNR is defined as the ratio of the mean signal intensity to its standard deviation over time (7) and is therefore relevant for evaluating coil performance for fMRI studies. TSNR in the spinal cord is considerably lowered by the presence of physiological noise, whose amplitude is higher than that in the brain (6).

Radio frequency (RF) receive-only coil arrays utilize a spatial arrangement of several small surface coil elements and record MR signal from each channel independently (8). This provides two major benefits; first, the sensitivity local to an individual element is increased due to the close coupling between the detector and tissue. Optimized combination of the array elements preserves the sensitivity near a given element and improves the sensitivity for tissue distant from all the elements. Second, simultaneous reception from the coils enables parallel imaging reconstruction methods in which the acquisition time is decreased. In EPI, the increased velocity of k-space traversal in the phase encode direction (achieved by omitting phase encode lines) has the effect of reducing susceptibility-induced distortions in the image, which is particularly valuable for spinal cord imaging.

Although receive arrays were originally developed and demonstrated for spinal cord imaging (8), most highly parallel detectors (≥32 elements) have focused on the brain (9-12) and heart (13-16). Some array designs have also been recently proposed for the spinal cord (17-20). These include arrays designed for full coverage of the spine up to 24 channels (18), and those designed specifically for the cervical spine up to 16 channels (17). For the c-spine, the geometry of the body allows elements to be placed around the neck to improve sensitivity. In typical state-of-the-art clinical coils, the head coil is compatible with the spine array and an anterior neck coil, and all elements can be selected simultaneously for imaging the head and spinal cord. Additionally, commercially available head and cervical arrays are typically designed for comfort and to be used with nearly 100% of the population, and hence are generally not made to fit as tightly to the body as might be desired in a research setting where sensitivity is the primary concern.

Here we report on a newly built close-fitting brain and c-spine coil designed to cover both regions with 32 circular loop elements. The design stops at the thoracic level T1-T2 to allow small elements to be used for maximum sensitivity and accelerated imaging capabilities and places most of the elements on a single rigid former so that coupling between elements can be minimized. We compare the dedicated brain/c-spine array to a commercially available head/neck/spine array coil and evaluate its performance with measures of image and temporal SNR and g-factor maps. We demonstrate the coil capabilities in highly accelerated anatomical images, DW imaging and tractography. Preliminary results were presented in abstract form (21,22).

Materials and Methods

Coil design/building

To obtain a close-fitting of the array to the head and neck, we propose a design based on a 3D surface reconstruction of a segmented anatomical MRI from a representative subject followed by the creation of the model in acrylonitrile butadiene styrene (ABS) plastic using a 3D printer. Figure 1a illustrates the steps used for the design of the posterior former, which housed most array elements (30 elements). Table 1 shows the dimensions of the 32-channel coil with comparison with a standard head/neck/spine array (Siemens). The coil former was tested on 35 individuals and is large enough to fit this population’s head and neck. In addition to the former we constructed two anatomically shaped “paddles” with 2 overlapped elements each such that on or the other (but not both) can be chosen for use with the 30 elements on the posterior former. The neck paddle is designed to cover the anterior part of the neck. The head paddle rests on the subject’s forehead and completes the brain coverage of the array (mostly the frontal lobe area). Hence, depending on the application, the appropriate paddle can be chosen to fully use the 32 receive channels of the clinical scanner. Figure 1b shows photographs of the 32-channel coil with both head and neck paddles.

Figure 1.

a. Methodology for optimal design of the coil using a 3D printer. Key steps were: automatic segmentation of an MPRAGE, reconstruction of 3D surface, design and printing of the coil frame. b. Photographs of the 30-channel coil with head and neck 2-channel paddles (although only one paddle can be combined at a time with conventional 32-channel MRI system). Switching between both paddles takes less than a minute. The coil fits all the 35 human subjects used for testing. c. Back of the coil array with zoomed panel focusing on the head/neck region. Preamplifiers were mounted close to each loop to minimize noise and cross-talk from cable currents. d. Schematics of a single array element, which consisted in a wire loop with four distributed capacitors. C4 was an adjusting capacitor for fine coil tuning.

Table 1.

Dimensions of coils used in this study (in mm). The 24ch spine array is not reported here because only a portion of the coil was used (the 3 elements most rostral elements). AP: Antero-posterior, LR: Left-Right and IS: Infero-superior

Coil dimensions (in mm)
	AP	LR	IS
32ch (head part)	185	195	203
32ch (neck part)	114	135	77
Standard 12ch Head	272	254	303
Standard 4ch Neck	220	200	190

The layout of coil elements consists of an arrangement of hexagonal and pentagonal tiles, similar to that used for the “soccer-ball” 32-channel head coil (10). Figure 1c shows photograph of the coil elements. The region covering the brain has 20 overlapped circular loop elements of diameter 95 mm and the region covering the neck and cervical-spine has a total of 10 overlapped elements of diameter 85 mm. The distance between the center of each element was kept at the critical value of 0.75 times the loop diameter to minimize their mutual inductance (8). However, during the passive decoupling phase, we manually bent each loop to further minimize their mutual inductance. Array elements were made of wire loops (tinned copper wire, 16 Gage) with four capacitors as shown in the schematic of Figure 1d. The tuning capacitor circuit board contains a variable capacitor to fine-tune the loop resonance to 123.2 MHz (corresponding to 3T). Each element was tuned and matched to minimize the noise figure of the Siemens preamp. The active detuning was achieved with a PIN diode in a parallel LC circuit. The preamp decoupling was achieved by transforming the preamplifier input impedance to a short across this diode using a coaxial cable length of 42 mm between the preamplifier and trap circuit. The preamplifier was mounted about 20 mm above the coil elements and was oriented parallel to the B₀ field.

Coil testing on the bench

Bench testing was carried out using a custom-made test rig that provided analog signal control for each of the 32 channels individually. The test rig included the power supply of 10 V for preamplifiers, switches to bias PIN diodes with 100 mA current to test detuning during transmit mode, as well as reverse biasing of the diodes (−30 V) during image acquisition. The test rig had BNC connections to allow the evaluation of each preamplified signal. Using the test rig on the 32-channel coil, we assessed detuning, decoupling, preamplifier decoupling, and element tuning. Additionally, the ratio of unloaded-to-loaded quality factor (Q_unload to Q_load) was measured using a tightly fitted head/neck phantom. Decoupling measurements were performed on the nearest neighboring elements using a direct S₂₁ measurement with cables directly connected to the preamplifier sockets of the two elements under test. When measuring the decoupling between an adjacent pair, all other unused elements of the phased array were actively detuned (PIN diode under forward bias) and had the preamps attached (supplying preamplifier decoupling). Note that because the loops were matched to the impedance needed for optimum noise performance (noise-match) Z_NM, simply connecting them to the 50 Ohm network analyzer would have been problematic in the S₁₂ measurement. Hence, we used an additional impedance transformation to transform the Z_NM to 50 Ohm for all adjustments using the network analyzer. This transformation was done with a series of capacitors in front of the output coax mounted on a “dummy” preamplifier, which were plugged into the array’s preamp socket. Additional tests (active detuning, preamplifier decoupling, Q-ratios, and final tuning) used the S₁₂ measurement between two decoupled (~80 dB) inductive probes lightly coupled to the array element under test. For measurement of the active detuning, the S₁₂ change between decoupled probes was recorded for a given array element with and without forward bias current, applied to the PIN diode trap while the preamplifier was replaced by a power matched load.

For the preamplifier decoupling, the S₁₂ between two decoupled probes lightly coupled to the array element was compared between two states. Firstly with the preamplifier in place (noise-matched) and secondly with the preamplifier replaced by power-matched load. Ratio of Q_unload-to-Q_load was measured in each loop under test surrounded by its non-resonant neighboring elements (each element being actively detuned by the PIN diode under forward bias). We used a phantom to measure Q_load. The coil quality factor was given by Q = f₀ / Δf, where f₀ was the center frequency and Δf was −3 dB bandwidth. The tuning and matching of the individual elements was set by monitoring S₁₁ at the preamp input using a calibrated cable and adjusting to the noise-matched impedance needed for the preamplifier.

Coil testing in the MRI

The 32-channel coil was developed and tested on a 3T scanner with 32 receive channels (Trio, TIM System, Siemens Healthcare, Erlangen, Germany). Five subjects were scanned after obtaining review by the institution’s Human Research Committee. The 32-channel coil was compared with a commercially available 19-channel array (Siemens) comprised of elements from the 12-channel head coil, 4-channel neck coil and 24-channel spine coil (only the most rostral row comprising of the Siemens spine coil was activated; these 3 elements were used in “triple mode” whereby each loop was utilized independently with a separate receiver channel).

Coil stability was evaluated on an agar phantom using single-shot gradient echo EPI time series (FOV=200mm, TR/TE=1000/30ms, BW=2298, flip angle=90°, matrix=64×64, 16 slices of 5mm each, 500 measurements). We measured the peak-to-peak variation in the signal intensity averaged over a 15-pixel square ROI positioned in the center of the phantom after linear and quadratic trends were removed from the time-series (23).

Safety assessment

The coil ran a battery of tests that assessed safety for the subjects. The coil cover was inspected to make sure no conductor was exposed and no sharp angle was present. To check whether active detuning during transmit phase was sufficient we measured the power needed to achieve the adjustment flip angle (180°) in a phantom with and without the receive coil present. The ratio of these two measures was required to be between 0.8 and 1.2. Cable lengths of less than 30 cm (between trap circuits) were required.

We also tested the coil for heating. After switching off the SAR monitor and the gradient stimulation monitor, we measured the temperature increase in the coil caused by RF transmit power being absorbed by the receive circuitry or heating by induced currents from the gradient switching. The detuned coil and phantom were scanned for 15 minutes with a body coil B1 field of 30 μT applied at a 10% duty cycle and repetition time of 60ms; an RF power level well above the SAR limit.

Image SNR and g-factor

Proton density weighted gradient echo images were acquired in the axial, sagittal and coronal planes (TR/TE = 300/6 ms, flip angle = 10°, slice thickness = 5 mm, FOV = 192 mm, matrix = 384×384, Bandwidth = 300 Hz/Pix). Noise covariance was measured in each channel using the same sequence without RF excitation. Given a signal vector S (a vector of image intensities at a given pixel across all coils), S^H the Hermitian transpose of S and Ψ⁻¹ the inverse of the noise covariance matrix Ψ, the image SNR for the root-sum-of-squares (rSoS) combination (24) can be expressed as:

$${\mathrm{SNR}}^{rSoS}=\frac{S^{\mathrm{H}}S}{\sqrt{S^{\mathrm{H}}\Psi S}},$$ [1]

And the image SNR for the noise covariance weighted root-sum-of-squares reconstruction (cov-rSoS) (8) is:

$${\mathrm{SNR}}^{\mathrm{cov}-rSoS}=\sqrt{S^{\mathrm{H}}{\Psi }^{-1}S}$$ [2]

For coil comparison, image SNR maps were calculated with the SNR^cov-rSoS formula above. The correction factors of Kellman and McVeigh (25) were then applied to the resulting SNR maps. SNR gain was quantitatively assessed in specific areas of the central nervous system. Regions of interest (ROI) were defined in the brain, cerebellum, brainstem and cervical spinal cord (Figure 3). SNR^rSoS and SNR^cov-rSoS were averaged within each ROI and then averaged across subjects.

Figure 3.

Left: Sagittal image showing the definition of ROIs used for quantifying the mean SNR in the supratentorial brain (1), cerebellum (2), brainstem (3) and cervical spinal cord (4). Middle, Right: Axial maps showing the ROIs used for quantifying the mean g-factor in the brain and spinal cord.

For a fair assessment of the g-factor characterizations relevant to accelerated imaging, the FOV along the phase-encoding direction should be selected to tightly enclose the object to maximize the aliasing induced by the acceleration (26). Given that the neck is usually smaller than the head in the A-P direction, we evaluated the coil g-factor using two different FOV: one that enclosed the head (aligned with the anterior-commissure / posterior-commissure, FOV = 220 × 220 mm), and one that enclosed the neck (perpendicular to the cord, at level C4, FOV = 180 × 180 mm). In both experiments, phase encoding direction was set to A-P, since L-R is usually avoided in spinal cord imaging to prevent aliasing of the shoulders. We used the head paddle for evaluating the g-factor in the brain, and the neck paddle for evaluating the g-factor in the spinal cord. SENSE g-factor maps were then computed voxel-wise from the individual channel data, noise covariance statistics, and coil sensitivity map information by taking the ratio between the SNR^cov-rSoS and SENSE SNR maps (27). Mean g-factor was measured within the brain and spinal cord using an axial ROI (Figure 3).

Conventional MRI

Two anatomical datasets were acquired. The same acquisition parameters were used for both coil arrays. The first sequence was a 3D T1-weighted multi-echo magnetization prepared rapid gradient echo (MEMPRAGE) (28) with the followings parameters: TR/TI = 2200/1200 ms, TE = 1.64/3.50/5.36/7.22 ms, flip angle = 7°, FOV = 320 mm (70% FOV in the phase encoding direction), resolution = 1×1×1 mm³, R = 4 acceleration in the A-P direction using GRAPPA reconstruction (29), Bandwidth (BW) = 285 Hz/pix, acquisition time = 3:20 min. Images were not corrected for intensity non-uniformity.

The second anatomical sequence was a 2D T2*-weighted with multi echo data imaging combination (MEDIC) that enabled good white/gray mater contrast in the spinal cord (30). Transverse slices were centered at C5 level. Parameters were: TR/TE = 1300/14 ms, flip angle = 30°, FOV = 154 mm, matrix = 256×256, effective resolution = 0.6×0.6×3 mm³ (further interpolated in k-space to reach 0.35×0.35×3 mm³), R = 3, BW = 260 Hz/Pix, with a saturation band covering the posterior neck and an acquisition time = 2:31 min.

Diffusion-weighted MRI

Three DW datasets were acquired. The first and second focused on the spinal cord and were aimed at demonstrating axial acquisition with high in-plane resolution. For these scans, we used the neck paddle. The sequence was based on a single shot spin echo EPI with twice refocusing pulse (31). The first dataset was acquired with 8 slices covering C2-C7 vertebral levels with the following parameters: TR/TE = 1530/102 ms, matrix = 128×128, resolution = 0.8×0.8×5 mm³, R = 2 (with 24 ACS lines), b = 700 s/mm², 30 diffusion-encoding directions, BW = 890 Hz/Pix. The second dataset was acquired with 14 slices covering C2-C6 vertebral levels with TR/TE = 2500/118 ms, matrix = 136×136, resolution = 0.6×0.6×5 mm³, b-value = 700 s/mm², 30 diffusion-encoding directions, BW = 750 Hz/Pix. Both acquisitions were triggered on the cardiac beat using the pulse oximeter. Following acquisition, motion correction was applied slice-by-slice using FSL FLIRT (32) with 3 degrees of freedom (T_x, T_y, R_z) using the b=0 image as the target image for registration. Diffusion tensors, Fractional Anisotropy (FA) and Monte-Carlo samples of principal directions were estimated using FSL BedpostX (33). From the latter output, maps of 95% confidence interval were computed using the method described in (34). These maps provided statistical confidence on the orientation of the first eigenvector of the diffusion tensor. It was therefore a good surrogate for the amount of noise present in DW data.

The third DW dataset aimed at demonstrating full brain and spinal cord tractography using small isotropic voxels. Here we used the head paddle. Parameters were: sagittal orientation, TR/TE = 14280/80 ms, FOV = 286 mm (75% FOV in the phase-encoding direction), 74 slices (no gap), matrix = 168×126, resolution = 1.7×1.7×1.7 mm³, R = 3 (with 32 ACS lines), b-value = 800 s/mm², 30 diffusion-encoding directions, BW = 1653 Hz/Pix. Diffusion tensor estimation and multi-seeded deterministic fiber tractography were performed using MedINRIA software (35).

Temporal SNR

To compute the tSNR and hence evaluate the potential benefits of the 32-channel coil for fMRI of the spinal cord, resting state fMRI time series were acquired at the c-spine level (neck paddle was used here). Slices were oriented perpendicular to the cord and centered at the middle of each vertebral body, which minimized the amount of susceptibility artifacts (6). Images were acquired using a gradient echo EPI sequence (TR/TE = 2500/35 ms, 10 slices, FOV = 192 mm, matrix = 128×128, resolution = 1.5×1.5×3 mm³, phase-encoding = A-P, R = 2, BW = 1502 Hz/Pix, 60 repetitions). TSNR maps were computed voxel-wise by dividing the averaged 60 time points by their standard deviation (36). Then, an axial region of interest (ROI) was defined over the spinal cord between C2 and C5 levels, and tSNR was averaged into this region to allow direct comparison between both coils.

Results

Coil characteristics

Each element showed an unloaded/loaded Q (with a head/neck phantom) ratio of ~240/40. Active detuning (via the PIN diode) provided >30 dB isolation between the tuned and detuned states. Preamp decoupling provided >15 dB reduction in the magnitude of the currents induced in the coil compared to the case where the preamp was replaced by a Z_NM load. Coupling between neighboring elements (S₁₂) ranged from −16 dB to −12 dB. Measure of coil impedance (S₁₁) showed the elements tuned and matched to the appropriate impedance.

All safety tests successfully passed. For assessing active detuning efficiency, the ratio of transmitted power with and without the coil present was 0.91. The temperature increase in the coil caused by RF transmit power being absorbed by the receive circuitry or heating by induced currents from the gradient switching was 1°C.

Stability test indicated a peak-to-peak variation of 0.36% over a 15×15 pixel ROI for 500 time-points for 3×3×5 mm³ resolution EPI (after detrending).

Image SNR noise correlation and g-factor

Figure 2a shows comparison of SNR^cov-rSoS maps between the 32-channel coil with neck paddle (left), the 32-channel coil with head paddle (middle) and the standard 19-channel coil (right). SNR increase was visible in most regions of the brain and spinal cord with maximum gains within the cortex. Results from specific regions of interest were summarized in Table 2 and show SNR increase of about 1.8-fold in the brain and 2-fold in the spinal cord

Figure 2.

a. SNR^cov-rSoS maps for the 32-channel coil with head paddle (left), the 32-channel with neck paddle (middle) and the standard coil (right) in one subject. For the two axial maps (two last rows), slices were oriented along the AC-PC for the brain and centered at C5 for the spine. SNR increase was visible in most regions of the brain and spinal cord with maximum gains within the cortex. SNR gain was quantitatively assessed in the brain, cerebellum, brainstem and cervical spinal cord. Results are showed in Table 2. b. Noise correlation coefficient matrices generated from acquisition without RF excitation. The averaged off-diagonal of the matrix was 7.7% for the 32-channel coil and 8.5% for the standard coil.

Table 2.

SNR comparison. Mean (± standard deviation across five subjects) has been computed in various parts of the central nervous system, as shown in Figure 3

Image SNR measurements
		Supratentorial brain	Cerebellum	Brainstem	Cervical cord
cov-rSoS	32ch (head paddle)	318 ± 19	309 ± 9	163 ± 30	205 ± 35
	32ch (neck paddle)	267 ± 8	304 ± 8	172 ± 14	229 ± 28
	Standard	172 ± 21	151 ± 21	127 ± 22	114 ± 11
	32chhead / Standard	1.8	2.0	1.3	1.8
	32chneck / Standard	1.6	2.0	1.4	2.0
rSoS	32ch (head paddle)	239 ± 10	247 ± 7	148 ± 6	169 ± 25
	32ch (neck paddle)	215 ± 6	244 ± 7	147 ± 11	181 ± 20
	Standard	161 ± 6	142 ± 5	120 ± 11	103 ± 4
	32chhead / Standard	1.5	1.7	1.2	1.6
	32chneck / Standard	1.3	1.7	1.2	1.8

Figure 2b shows noise correlation coefficient matrices for the 32-channel coil and the standard coil. The averaged off-diagonal of the noise correlation coefficient matrix was 7.7% for the 32-channel coil (with head paddle) and 8.5% for the standard coil. Figure 4 shows inverse g-factor maps (1/g) for both coils at various SENSE acceleration factor (R). Inverse g-factor was chosen to allow identical color scaling in all conditions. Overall the 32-channel coil exhibits lower g-factor than the standard coil. Based on ROI analysis, mean g-factor in the brain decreased by 12%, 22%, 19%, 9% and 54% at R=3, 4, 6, 2×2 and 3×3, respectively. Mean g-factor in the spinal cord was decreased by 9%, 39%, 57% and 22% at R=3, 4, 6, and 2×2, respectively.

Figure 4.

Maps of the inverse SENSE g-factor (1/g) at various accelerations for both coils. Closely fitted axial FOV was prescribed for the brain (top, FOV = 220 × 220 mm) and cervical region (bottom, FOV = 180 × 180 mm). For single acceleration direction (R=2 to 8), phase encoding was set to A-P. The peak g-factor is shown below each map. The spinal cord is circled in black. This figure suggests that 3x acceleration was feasible in the brain and spinal cord with relatively small cost in SNR. Note that there were not enough coil elements in the standard coil to accelerate with R=8 or R=3×3 in the cervical region.

Conventional imaging

Figure 5a compares the MEMPRAGE (3:20 min acquisition time) between the 32-channel coil and the standard coil. The R=4 MEMPRAGE showed little SNR reduction and no visible reconstruction artifact (aliasing pattern) with the 32-channel coil. In comparison, the standard coil exhibited lower SNR. Figure 5b compares both coils with the MEDIC sequence within the cervical spinal cord. The 32-channel coil provided higher SNR, enabling better delineation between the white and the gray matter.

Figure 5.

a. MEMPRAGE sequence. FOV = 320×224 mm, resolution = 1×1×1 mm³, TR/TI = 2200/1200ms, flip angle = 7°, R=4 (accelerated along A-P), BW = 285Hz/pix, TA = 3:20min. Images were not corrected for intensity non-uniformity and have the same color scaling. b. MEDIC sequence, transverse slice centered at C5 level. The same slice prescription was used for both coils. FOV = 154 mm, TR/TE = 1300/14 ms, flip angle = 30°, matrix = 256×256, resolution = 0.6×0.6×3 mm³, R=3, BW=260Hz/pix, saturation band covering the posterior neck, TA=2:31min.

Diffusion-weighted MRI

Figure 6 shows the FA and angular confidence maps of the diffusion tensor first eigenvector for the 0.8×0.8 mm² and for the 0.6×0.6 mm² in-plane resolution. Overall the 32-channel coil provided more consistent FA maps (regarding the white/gray matter geometry) and lower angular uncertainty.

Figure 6.

FA and maps of 95% angular confidence computed from FSL BedpostX for the 0.8×0.8 mm2 and 0.6×0.6 mm² in-plane spatial resolution. Overall the 32-channel coil provided lower angular uncertainty in the whole cervical cord.

Figure 7a shows the diffusion-weighted acquisition at 0.6×0.6 mm² centered at C4 level (3 averages). The cord was covered by about 20 voxels in the lateral direction and by about 15 voxels in the A-P direction. Dorsal/ventral roots were nicely delineated on the 2^nd eigenvector, previously shown using q-ball imaging on ex vivo scans (37). Figure 7b shows a full brain + c-spine 1.7 mm isotropic resolution tractography combining large field of view (286 mm × 214 mm), large matrix size (168×126) and high acceleration factor (R = 3). Fiber tractography was feasible without noticeable distortion.

Figure 7.

a. Example of DW images of the spinal cord at 0.6×0.6×5 mm³ spatial resolution. Dorsal/ventral roots were visible on the 2nd eigenvector. b. Full brain + c-spine tractography from 1.7mm isotropic DTI acquisition using the 32-channel coil with head paddle (74 sagittal slices, TR/TE = 14280/80 ms, R=3, b-value = 800 s/mm², 30 diffusion-directions). No distortion correction was applied.

Temporal SNR

Figure 8 shows tSNR maps for the 32-channel (top) and standard (bottom) coils for an axial slice centered at C4. Quantitative comparison was done by averaging the tSNR in an ROI covering the spinal cord from C2 to C5 levels in one subject. Mean tSNR was 24.7 +/− 5.9 for the 32-channel coil and 13.9 +/− 5.8 for the standard coil. Values ranged between 20 and 30 for the 32-channel coil and between 5 and 15 for the standard coil.

Figure 8.

TSNR maps for the 32-channel (top) and standard (bottom) array coils for an axial slice centered at C4.

Discussion

We presented a new 32-channel coil for imaging the brain, brainstem, cerebellum and cervical spinal cord. SNR measurements respectively showed a 1.8x and 2x increase in the supratentorial brain and spinal cord, compared to a commercially available coil (12-channel brain, 4-channel neck, 3-channel spine). Some of the noise correlations were as high as 0.5, which was likely due to the close proximity to the neck tissue. Because of this, the cov-rSoS combination yielded higher SNR than with the rSoS combination (Table 2).

The coil g-factor allowed 3x acceleration in the A-P direction for the diffusion measurement and R=4 for the MEMPRAGE. The FOV was not perfectly tight to the subject neck or brain, so the g-factors might have been a little lower for a given R. High acceleration factors will help speed up conventional T1- or T2-imaging and reduce susceptibility distortions in EPI-based techniques. For DW imaging, the 32-channel coil showed lower uncertainty of the DTI first eigenvector at high in-plane resolution (0.6×0.6 mm²) and enabled full brain and cervical cord tractography at 1.7 mm isotropic resolution. For fMRI, the overall 2-fold gain in tSNR in the cervical spine is likely to improve the sensitivity of activation detection.

The higher SNR performance of the 32-channel coil notably arose from the closer fitting of the coil to the head/neck of each individual, as well as from the smaller size of each individual coil elements.

Coil layout

Although a further coverage down the spinal cord would have been beneficial (e.g., to cover the whole thoraco-lumbar cord), this would have required more elements to keep the same coil density around the brain and c-spine. Since our goal was to limit the number of channels to what is commonly available in clinical scanners (32-channel), distributing elements in the thoraco-lumbar region would have reduced the number of elements in the brain/c-spine region, therefore lowering SNR and acceleration capabilities. Instead we chose to focus our development on the brain and cervical cord only to maximally facilitate the large number of existing applications in basic science and clinical research in that region.

The developed coil consisted of a shaped former covering the posterior head and neck that housed 30 elements with the possibility to add a paddle (2-channel) for either the frontal brain region or the anterior part of the neck. Although SNR in the spinal cord was higher when the neck paddle was used, the gain was somewhat marginal. In all subjects we managed to position the anterior neck paddle as close as possible to the top of the neck, however to avoid subject discomfort, the paddle never touched the neck. We also allowed some space for swallowing. The distance between the paddle and the neck was approximately 1 cm. As an alternative design, a single CP style channel could have been used for both paddles yielding a total of 32 channels, without having to switch.

DW-MRI of the spinal cord

Acquisition of 0.6×0.6×5 mm³ resolution DW images was feasible with few distortions thanks to the R=2 accelerated acquisition. Higher sensitivity for the 32-channel coil yielded higher reproducibility of DW measurements and therefore more consistent DTI metrics, as assessed by the angular confidence interval of the first eigenvector. The high in-plane resolution offers the possibility to quantify diffusion metrics in various sub-quadrants of the spinal cord with limited partial volume effects (38). This method improves the delineation of white matter pathways and could potentially increase correlations between clinical parameters and DTI metrics, e.g., between somatosensory evoked potentials and DTI metrics measured in the dorsal columns.

Full brain and spinal cord tractography was feasible at 1.7 mm isotropic resolution with R=3 acceleration factor. Such acquisition scheme is valuable for conducting tractography-based spatial statistics throughout the whole brain and cervical spinal cord. One possible application would be to study the spatial distribution of FA in the full corticospinal tract to obtain more insights in some neurodegenerative diseases, like in amyotrophy lateral sclerosis, where the origin of central and peripheral motor-neuron degeneration is still under investigation (39). Here, we chose to image in the sagittal plane to cover the full brain and cervical cord with minimum number of slices (74 slices were used here). However, the TR was still relatively large (>14s) therefore imposing long acquisition time (>7min). Recent developments in simultaneous multi-slice imaging techniques (40) will enable further reduction of scan time by taking advantage of the acceleration capabilities of the 32-channel coil.

FMRI of the spinal cord

Here we demonstrated increase in tSNR from 13.9 +/− 5.8 for the standard coil to 24.7 +/− 5.9 for the 32-channel coil. In a previous study conducted in nine subjects using the same standard coil as here (6), tSNR was 14 +/− 2 in the cervical spinal cord, in good agreement with our measurement. Note that the tSNR was estimated from the raw time series without regressing out any component using a general linear model. This implies that the tSNR could be further improved by careful modeling of physiological source variations, arising from respiratory, cardiac and cerebrospinal fluid fluctuations (41).

Applications

Having demonstrated the benefits of an optimized array coil to image the brain and c-spine region, this technical development opens the door to numerous investigations involving the whole central nervous system. Being able to image the full brain and spinal cord will be of major interest for improving differential clinical diagnosis towards other neurodegenerative diseases, such as in multiple sclerosis where the spinal cord is heavily involved (42). On a more fundamental aspect, improvements in brain, brainstem and spinal cord imaging could allow combined functional mapping of these regions with large-scale tractography, advancing imaging protocols for studying pain, motor learning or anatomo-functional plasticity of the central nervous system following trauma. The 2-fold increase in SNR in the cerebellum also motivates further investigations of DW-MRI in this region such as the structural connectivity of the principle circuits in this region (43).

Conclusion

We developed and validated a new 3T 32-channel coil for MR imaging of the brain and cervical cord region, which is compatible on clinical imagers without additional hardware/software. The improved sensitivity and parallel acquisition capabilities enable faster and higher spatial resolution anatomical and diffusion imaging. The two-fold increase in temporal SNR could potentially improve sensitivity in spinal cord fMRI studies. Improvements of this nature potentially enhance numerous basic and clinical research studies focused on spinal and supraspinal regions.