In Vivo ¹H NMR Spectroscopy of the Human Brain at 9.4 Tesla: Initial Results

Abstract

In vivo proton NMR spectroscopy allows non-invasive detection and quantification of a wide range of biochemical compounds in the brain. Higher field strength is generally considered advantageous for spectroscopy due to increased signal-to-noise and increased spectral dispersion. So far ¹H NMR spectra have been reported in the human brain up to 7 Tesla. In this study we show that excellent quality short echo time STEAM and LASER ¹H NMR spectra can be measured in the human brain at 9.4 Tesla. The information content of the human brain spectra appears very similar to that measured in the past decade in rodent brains at the same field strength, in spite of broader linewidth in human brain. Compared to lower fields, the T₁ relaxation times of metabolites were slightly longer while T₂ relaxation values of metabolites were shorter (< 100 ms) at 9.4 Tesla. The linewidth of the total creatine (tCr) resonance at 3.03 ppm increased linearly with magnetic field (1.35 Hz/Tesla from 1.5 T to 9.4 T), with a minimum achievable tCr linewidth of around 12.5 Hz at 9.4 Tesla. At very high-field, B₀ microsusceptibility effects are the main contributor to the minimum achievable linewidth.

Introduction

Higher signal-to-noise ratio (SNR) and spectral resolution at high magnetic fields have enabled significant gains in the quantification precision for a wide range of metabolites that can be measured using in vivo ¹H magnetic resonance spectroscopy (MRS). In rodents at 9.4 T, close to 20 metabolites can be reliably quantified in specific regions of the brain (“neurochemical profiles”) [1]. ¹H NMR spectra were also recently reported in rat brain at even higher fields, 14.1 T [2] and 16.4 T [3]. Numerous studies performed in the past decade have shown how such ¹H MRS measurements provide unique insights into brain biochemistry in normal brain and in animal models of brain disorders [4].

In the human brain, several studies reported gains in quantification precision at 3 T or 4 T compared to 1.5 T [5; 6] and at 7 T compared to 3 T [7]. A more recent study found close to a 4-fold improvement in quantification precision at 7 T compared to 4 T [8]. At 7 T, up to 15 metabolites can be quantified reliably [9]. Gains in sensitivity have also been reported for ¹H spectroscopic imaging at 7 T compared to 1.5 T [10].

Such gains at high magnetic field have been achieved in spite of several limiting factors encountered at high fields. These include: longer T₁ and shorter T₂ relaxation times of metabolites, higher RF power deposition (SAR), lower available transmit B₁ RF field, stronger B₀ and B₁ inhomogeneity and increased spectral linewidth. In addition, the exact gain in quantification precision at high field is dependent on the experimental conditions (such as preamplifiers, RF coils and pulse sequences) and therefore difficult to determine precisely. Nonetheless, the fact that the quantification precision increases 4-fold from 4 T to 7 T [8] clearly shows the benefits of high fields for ¹H MRS up to 7 T in humans and suggests that further gains may be possible beyond 7 T.

Magnetic fields of 9.4 Tesla recently became available for human studies and initial results of ¹H and ²³Na MRI of the human brain have been reported at such ultra high field [11; 12]. However, to the best of our knowledge, no in vivo ¹H NMR spectroscopy has been reported in human brain above 7 T.

In the present study we report the first ¹H NMR spectra acquired in the human brain at 9.4 Tesla, obtained with two single-voxel methods, STEAM (TE = 8 ms) and LASER (TE = 41 ms), in the occipital lobe. T₁ and T₂ relaxation times of brain metabolites are also reported. The B₀ field dependence of the spectral linewidth and of the relaxation times of metabolites is also examined.

Materials and Methods

Subjects and MR system

Six healthy subjects (20 to 55 years of age) participated in this study. Informed consent was obtained according to procedures approved by the Institutional Review Board of the University of Minnesota. All experiments were performed using a 9.4 Tesla / 65 cm [11] horizontal bore magnet (Magnex Scientific Ltd, Oxford, UK) interfaced to a Varian DirectDrive console (Varian, Palo Alto, CA, USA) with 8 independent transmit channels. The magnet system was equipped with an asymmetric self-shielded head gradient (40 cm inner diameter, maximum strength of 28.5 mT/m^† with a rise time of 150 µs) powered by a Siemens gradient amplifier (Siemens AG, Medical Engineering, Erlangen, Germany). The subjects lay in a supine position with their head placed above an RF coil (described below) and wore earplugs throughout the study to attenuate the gradient acoustic noise.

A home-built half-volume RF probe covering the posterior half of the head and composed of eight ¹H microstrip elements (13 cm in length with a 12 mm wide copper strip conductor and a 4.2 cm wide RF ground plane, separated by a Teflon dielectric layer with a thickness of 12 mm) was used for both RF transmission and reception [13; 14]. These elements were powered by one 1 kW and seven 500 W RF amplifiers (CPC, New York, USA). The 1 kW amplifier was connected to the coil element which was contributing the maximum transmit B₁ (B₁⁺) in the region-of-interest based on the relative B₁ maps of the individual coil element. The applied RF power delivered to each coil element was continuously monitored with a 10 sec and a 10 min moving average using an in-house built monitoring system. The latter consisted of directional couplers attached to each RF amplifiers. The power envelope of the RF waveform was continuously sampled by a calibrated ADC board. If the power of any coil were to exceed a predefined threshold, calculated based on FDA guidelines, then all RF amplifiers would be immediately shut down. To ensure safety, the threshold for the monitoring system was conservatively defined assuming that electric fields from all RF coils were to add constructively in the sample. Electromagnetic simulations were also performed to evaluate SAR. For this purpose, complex B₁ and E fields were generated for the eight-element microstrip RF coil loaded with a human head model using XFDTD software (Remcom Inc., PA, USA). SAR values were computed through the whole brain, taking into account the relative RF input power, B₁ shimming, RF pulse shapes, RF pulse durations and repetition time used for in vivo experiments. The resulting local and global SAR levels were found to be well below the FDA limits (data not shown).

NMR Spectroscopy Measurements

In order to minimize destructive B₁⁺ interferences [15] in the single voxel used for spectroscopy (in this case located in the visual cortex), the relative phase of the transmit B₁⁺ field for each coil element was optimized using a fast local B₁⁺ shimming technique as recently described [16; 17]. A series of low flip angle gradient-echo images were acquired while transmitting RF power through one coil element at a time and receiving with all eight elements simultaneously. The relative transmit phase for each coil element was optimized in order to obtain maximal B₁ in the region of interest using Matlab (The MathWorks Inc., Natick, MA, USA). Signals from all eight channels were acquired using a digital receiver system that was developed in-house; this receiver used an Echotek (Huntsville, AL, USA) ECDR-814 board to oversample the 20-MHz intermediate frequency (IF) at 64 MHz and 14-bit resolution, with digital band-pass filtering.

Using transverse and sagittal gradient-echo images obtained after B₁⁺ shimming, a volume-of-interest (VOI) of 8 ml (2×2×2 cm³) was positioned in the visual cortex. All first- and second-order shim terms were automatically adjusted using a multi-transmit version of FAST(EST)MAP [18] resulting in a water linewidth of 15.7 ± 1.4 Hz (full width at half height). The water linewidth did not improve when the voxel size was decreased from 8 ml to 1 ml indicating that macroscopic B₀ shimming was optimal.

Single-voxel localization was achieved using STEAM with TE of 8 ms (1.5 ms asymmetric RF pulses, 4.5 kHz bandwidth) transmitting on all eight elements with the same RF power and the phases determined from the B₁ shimming procedure. Water suppression was performed with VAPOR using eight RF pulses with variable pulse power and optimized timing [19]. To suppress unwanted signal outside the volume-of-interest, outer volume suppression (OVS) pulses were also applied as described in [19]. The OVS scheme was made up of four blocks, each consisting of at least six slice-selective hyperbolic-secant RF pulses (6 ms duration, 7 kHz bandwidth) followed by gradient spoilers. These OVS modules were interleaved with the water suppression pulses. Spectra were acquired with a repetition time of 6 s, 8 k complex data points and spectral width of 8 kHz. Each free induction decay (FID) was individually saved for subsequent scan-to-scan frequency correction. A water reference spectrum was also recorded in order to perform eddy current correction and to serve as a concentration reference for absolute quantification. Macromolecule spectra were acquired using the inversion-recovery technique with the inversion time optimized to null the signals from metabolites.

¹H NMR spectra were also acquired using LASER (Localization by Adiabatic SElective Refocusing, [20]) with TE = 41 ms. The sequence consisted of a 3 ms nonselective adiabatic half-passage pulse followed by three pairs of slice-selective adiabatic full-passage pulses (2.5 ms duration, 4.8 kHz bandwidth, HS4 modulation) for 3D-localization. Water suppression was also achieved with VAPOR. Spectra were acquired with a repetition time of 6 s with 8 k of data points and spectral width of 8 kHz.

In both pulse sequences, the transmitter frequency was set at 2.65 ppm (close to the NAA multiplets). This resulted in a ~13% displacement artifact for resonances at the edges of the detected spectrum (4.2 ppm and 1.1 ppm) compared to the nominal voxel selection.

Measurements of Relaxation Times

T₁ relaxation times of metabolites were measured using STEAM preceded by an adiabatic inversion pulse. An echo time of 80 ms was chosen in order to minimize signals from macromolecules, which have a much shorter T₂ than metabolites. Six different inversion times t_IR were used (0.006, 0.22, 0.45, 1.75, 3 and 5 s). Spectra were acquired under nearly fully relaxed conditions with a delay of 6 seconds between acquisition of each fid and the next inversion pulse and with 32 repetitions. Metabolite T₁ values were obtained by fitting the relative concentrations obtained after LCModel analysis of the inversion recovery ¹H spectra using a single exponential function with 3 fitted parameters. T₁ values for J-coupled metabolites were not estimated due to the relatively long TE used in these measurements.

For T₂ relaxation times measurements, spectra were collected at seven different echo times: 70, 80, 90, 100, 120, 150 and 200 ms with a repetition time of 6 s. For each echo time, 32 averages were acquired except for TE ≥ 150 ms where 64 averages were recorded in order to increase the signal-to-noise ratio. Signals from macromolecule resonances were negligible at these long TEs. Metabolite T₂ relaxation times were determined by fitting the relative concentrations obtained after LCModel analysis of ¹H spectra using a single exponential delay function with 2 fitted parameters.

Processing and Quantitation of In Vivo Spectra

All spectral pre-processing was performed in Matlab. Water suppressed FIDs acquired from each coil element were corrected for eddy current by using their corresponding water reference scan. After eddy current correction, all FIDs had similar phase. These FIDs were individually corrected for any B₀ frequency shift then summed together after applying a weighting factor for each channel in order to maximize the SNR of the summed spectrum. The weighting factor was based on the SNR of each individual channel determined from their N-acetyl aspartate (NAA) peak [21; 22]. All ¹H NMR spectra were analyzed using LCModel (Stephen Provencher Inc., Oakville, ON, Canada). The basis spectra for each detectable brain metabolite were simulated using home-written programs based on density matrix formalism in Matlab with measured and published chemical shifts and J-coupling values [23]. No baseline correction, zero-filling or apodization functions were applied to the in vivo weighted summed data prior to the analysis.

Results and Discussion

Local B₁⁺ Phase Shimming

At high magnetic field, destructive interferences can severely reduce the available B₁⁺. High peak B₁⁺ is highly desirable for NMR spectroscopy in order to minimize the duration of the RF pulses, thereby reducing the echo-time and chemical-shift displacement artifacts. Maximizing B₁⁺ efficiency is also critical to minimize SAR, particularly at high field where SAR deposition is more likely to become a limiting factor.

One way to maximize the available B₁⁺ over a region-of-interest at high field is to perform local B₁⁺ shimming [16; 17]. Figure 1 demonstrates the gain in transmit B₁ obtained using B₁⁺ shimming with our setup at 9.4 Tesla. The maximum B₁⁺ was 205 Hz across the defined region-of-interest in the occipital lobe after B₁ phase shimming versus only 60 Hz before B₁⁺ shimming, corresponding to a 10 dB gain in transmit power efficiency. Note that these B₁⁺ maps were acquired with an RF power well below the maximum power available in order to remain within SAR limits. Based on the measured B₁⁺ maps at low power, we extrapolated that the maximum B₁⁺ available for this particular experimental setup was ~1.5 kHz in the occipital lobe when using the full transmit power available with our RF amplifiers (1 kW RF amplifier output on the channel with the highest B₁⁺ contribution in the VOI and 500 W amplifier output on the remaining seven channels). This maximal B₁⁺ value was also confirmed directly from the calibration of RF pulses in STEAM at full power. Note that only about half of the transmitted power was actually available at the coil ports due to losses in the cables.

Figure 1.

Top left. The two color pictures show the absolute B₁⁺ maps in Hz before and after local B₁⁺ phase shimming, optimized for the region-of-interest shown as a pink circle in the occipital lobe on the left picture (|B₁⁺| maps were obtained using AFI [50] with: TR1/TR2/TE=25/125/5 ms, FOV=20×20×12 mm, data matrix=256×64×24). Note the very substantial increase in |B₁⁺| after B₁ shimming. Bottom left. Gradient-echo axial image after B₁⁺ shimming. The eight yellow bars (labeled 1 to 8) indicate the position of the eight coil elements. Right. ¹H NMR spectra from each coil element (labeled 1 to 8 on the left vertical axis), acquired in an 8 mL VOI (denoted by the red square on the bottom left image) with STEAM (TE = 8ms, NT = 32). The respective SNR is shown on the right end of each spectrum. Note that the spectra were sampled with the eight coil elements simultaneously. Each spectrum was scaled to have the same NAA peak intensity.

In the particular case shown in Figure 1, the relative transmit phase of each element (from 1 to 8) after B₁⁺ shimming was: 0, 33, 20, 319, 284, 292, 297 and 322 degrees where element #1 was arbitrarily chosen as the reference phase with 0 degree. This optimal phase for each element after B₁⁺ phase shimming was generally reproducible from one subject to the next, with the biggest subject-to-subject variations (up to 40 degrees) for element #7. STEAM ¹H spectra acquired with each of the eight individual coil elements are also shown in Figure 1. The highest receive sensitivity for the VOI illustrated in Figure 1 was obtained with elements #4 to 7. Summing spectra from all eight elements (after weighting each spectrum with their SNR [21; 22]) further increased the SNR by ~20% compared to the case when only the four spectra with the highest SNR (elements #4 to 7) were used.

In Vivo Proton NMR Spectra

We assessed the feasibility of ¹H NMR spectroscopy with two single-voxel sequences widely used in our and other laboratories, namely STEAM and LASER.

STEAM Spectra

¹H NMR STEAM spectra obtained in the human brain at 9.4 T at very short echo-time (TE = 8 ms, NT = 384) showed excellent spectral quality (Figure 2, top). There was no noticeable baseline distortion or contamination by signals from outside the voxel (such as lipids), indicating satisfactory outer volume suppression (OVS). Excellent water suppression was also consistently achieved, as illustrated by the small water residual observed in all subjects. The spectrum was dominated by strong singlet resonances of NAA, Cho, Cr and PCr as expected. The glutamate multiplet at 2.35 ppm was clearly resolved from neighboring resonances such as glutamine (2.45 ppm). Contributions from smaller resonances such as aspartate and GABA were also noticeable.

Figure 2.

Comparison between a STEAM ¹H NMR spectrum acquired in vivo at 9.4 T from the human visual cortex (top, VOI = 8 ml, TE = 8 ms, TM = 35 ms, lb (line broadening) = −3, gf (Gaussian factor) = 0.15, 384 scans) and a similar spectrum acquired in rat brain (bottom, VOI = 65 µl, TE = 2 ms, lb = −1, gf = 0.25, 640 scans, courtesy of Dr. I. Tkáč). Human data represents the weighted sum from all eight coil elements using the spectral processing steps described in the Methods section.

The human brain STEAM proton spectrum had an overall spectral pattern very similar to that obtained in the rat brain at the same field strength (Figure 2, bottom). However the spectral linewidth was noticeably smaller in rat brain (8–10 Hz, [1]) compared to human brain (~13 Hz) due to longer T₂ and smaller microsusceptibility effects. In addition, several metabolites showed different intensities in human brain compared to rat brain. For instance, taurine (3.2 – 3.4 ppm) and lactate (1.3 ppm) levels were lower in human than in the rat brain. Note that the concentration of taurine is known to be different across species [24] and that the higher lactate concentration observed in the rat is explained by the use of isoflurane anesthesia in this particular animal.

LASER Spectra

In addition to STEAM spectra, LASER ¹H NMR spectra were also acquired at a minimum TE of 41 ms (NT = 168, Figure 3). Even at this relatively longer TE, signals from strongly coupled resonances such as glutamate or the multiplet of NAA were still being observed as expected, since the Carr-Purcell pulse train employed in LASER [20] minimizes J-evolution. Macromolecule resonances were strongly reduced compared to the STEAM spectrum (Figure 2) due to their short T₂ relaxation times.

Figure 3.

Weighted summed in vivo ¹H LASER spectrum (TE = 41 ms, VOI = 6 ml, 168 scans, gf = 0.09, lb = −2) acquired from the human visual cortex at 9.4 T.

Absolute Quantification of Short Echo Time STEAM Spectrum

LCModel analysis of in vivo ¹H STEAM spectrum (TE = 8 ms, TM = 35 ms, 32 averages, occipital lobe) has allowed the absolute concentration of at least 15 metabolites to be reliably determined (Table 1) using the water peak of the unsuppressed reference scan as an internal concentration reference. These concentrations were in excellent agreement with previously published data [6; 9]. Cramer-Rao Lower Bounds (CRLBs) were also below 20% (Table 1) for most metabolites, except those at a very low concentration (< 0.5 µmol/g).

Table 1.

Metabolites	Concentration (µmol/g)	CRLB (%)
Creatine	3.2 ± 0. 5	7.4 ± 1.7
Phosphocreatine	4.5 ± 0.4	5.3 ± 1.0
Creatine + Phosphocreatine	7.7 ± 0.4	2.0 ± 0.0
Total Choline (GPC +PCho)	0.9 ± 0.2	5.7 ± 1.4
NAA	13.5 ± 1.6	1.9 ± 0.4
Glutamate	9.3 ± 0.9	2.0 ± 0.0
Glutamine	2.2 ± 0.2	6.9 ± 0.9
Myo-inositol	5.3 ± 0.4	3.0 ± 0.6
Glutathione	1.1 ± 0.3	8.9 ± 3.4
GABA	1.3 ± 0.4	12.4 ± 3.6
NAAG	1.1 ± 0.5	18.7 ± 7.5
Taurine	1.3 ± 0.2	11.4 ± 2.3
Phosphethanolamine	1.6 ± 0.4	10.7 ± 2.3
Aspartate	2.1 ± 0.5	14.4 ± 5.2
Lactate	0.5 ± 0.1	23.4 ± 10.5
Alanine	0.3 ± 0.3	35.4 ± 29.0
Scyllo-inositol	0.3 ± 0.2	39.7 ± 42.8

Concentrations of cerebral metabolites in the human brain measured at 9.4 T after LCModel analysis of six STEAM spectra (TE = 8 ms, 32 averages each) from 6 subjects. Data represent mean ± standard deviation.

The CRLBs of metabolites obtained in this study were comparable to those obtained at 7 T using STEAM [9] and slightly lower than those reported in another 7 T study with SPECIAL pulse sequence [7]. We would like to emphasize that comparing sensitivity at 9.4 Tesla with that at lower fields was not our objective in this initial study. Such comparisons require careful matching of RF coil designs and acquisitions conditions on different MR systems. The 8-element microstrip RF coil used in this study may not provide optimal sensitivity for spectroscopy in the occipital lobe. We expect that further developments in RF coil design will lead to improvements in sensitivity.

T₁ and T₂ Relaxation Times of Brain Metabolites

Measured T₁ and apparent T₂ relaxation times for the singlet resonances of NAA, total creatine and total choline are reported in Table 2.

Table 2.

Metabolites	T1 (ms)	T2 (ms)
NAA singlet	1777 ± 82	98.0 ± 8.4
Total Creatine (CH3 group)	1746 ± 133	71.9 ± 5.0
Total Creatine (CH2 group)	1030 ± 270	68.3 ± 6.4
Total Choline	1513 ± 153	70.7 ± 7.1

T₁ and apparent T₂ relaxation times (mean ± SD) of metabolites measured in the brain of six healthy subjects at 9.4 T using STEAM sequence. T₁ values were fitted with R² ≥ 0.96 and T₂ with R² ≥ 0. 97.

The T₁ relaxation times of metabolites were measured using STEAM preceded by inversion-recovery. The longest T₁ values were found for the methyl groups of NAA and total creatine. The T₁ relaxation time for the CH₂ resonance of total creatine was significantly shorter than that for the CH₃ resonance as expected from previous studies.

T₂ relaxation times were measured using STEAM at multiple echo times. All apparent T₂ relaxation times were below 100 ms (Table 2). The CH₃ resonance of NAA had the longest apparent T₂ (98 ms) and the CH₂ and CH₃ resonances of total creatine had almost identical apparent T₂ values; 68.3 ms and 71.9 ms respectively.

B₀ Field Dependence of Relaxation Times

T₁ and T₂ relaxation times of metabolites measured at 9.4 T were compared to values from literatures at lower fields (T₁: [25; 26; 27; 28; 29; 30; 31; 32; 33; 34]; T₂: [9; 25; 26; 27; 28; 29; 30; 31; 32; 33; 34; 35; 36; 37; 38; 39; 40; 41; 42; 43]; Figure 4). Only values from studies using STEAM or PRESS were retained for this comparison.

Figure 4.

T₁ and T₂ relaxation times as a function of static B₀ field strength for the methyl resonance of NAA and total creatine measured in the human brain using STEAM and/or PRESS. Data points with error bars represent the mean and standard deviation of published relaxation times at each field strength.

T₁ relaxation times of metabolites in human brain at 9.4 T were slightly longer than at lower fields, i.e. 1.5, 4 and 7 T (Figure 4, left) although no clear B₀ field dependence could be established due to the relatively large dispersion in the reported T₁ values at lower fields.

In contrast, T₂ relaxations of metabolites were significantly shorter in humans at 9.4 T (Figure 4, right) than at lower fields. On average, the apparent T₂ relaxation times in human at 9.4 T were ~30% shorter than at 7 T [9]. This is consistent with the decrease in T₂ relaxation times with B₀ demonstrated in a number of previous studies in both humans and animals [25; 40; 42; 43]. Based on the T₂ values at various field strengths (from 1.5 to 9.4 T, Figure 4), relaxation rates R₂ (i.e. 1/T₂) as a function of B₀ were fitted with a quadratic function in the form (A · B₀ⁿ + C). The term C can be attributed to dipole-dipole relaxation (R_2,dd) which is virtually independent of B₀ and was found to have values of several seconds⁻¹ (C = 2.6 ± 0.6 s⁻¹ and C = 3.8 ± 0.7 s⁻¹ for methyl group of NAA and tCr respectively) consistent with previous studies [44]. The fitted parameter n corresponding to the best fit was n = 1.7 ± 0.4 s⁻¹ and n = 1.3 ± 0.2 s⁻¹ for the CH₃ group of NAA and tCr respectively. The B₀-dependent term (proportional to B₀ⁿ) can be attributed to a large extent to increased dynamic dephasing due to increased local susceptibility gradients as recently suggested by Michaeli et al. [40; 44]. These susceptibility fields result in a loss of phase coherence between spins thereby shortening the T₂. The effect of this signal loss can be minimized considerably by placing a Carr-Purcell pulse train between excitation and signal acquisition. Examples of such pulse sequence are CP-PRESS [35] and CP-LASER [40] where the apparent T₂ times of metabolites is prolonged. The observation that n < 2 (statistically significant only for tCr and not for NAA) must be taken with caution, as the fitted data come from multiple studies published by different groups with different techniques and potentially variable diffusion weighting with increasing TE.

Comparison of Relaxation Times in Human and Rat Brain

T₁ relaxation times of metabolites were generally longer in human compared to rat brain at 9.4 T (Table 3). For example, T₁ for the methyl group of NAA was 1.78 s in human versus 1.41 s in rat brain at 9.4 T [1].

Table 3.

	T1 (s)		T2 (ms)
Metabolites	Human (current study)	Animal [1]	Human (current study)	Animal [1; 45; 46]
NAA	1.78	1.41	98	190
tCr	1.75	1.34	72	117.7

Comparison of T₁ and T₂ relaxation times of methyl group of NAA and tCr between human and animal (rat) brain at 9.4 T.

In contrast, the apparent T₂ relaxation values were shorter in human compared to rat brain at the same field strength (Table 3). For example, the mean apparent T₂ of methyl group of NAA was 98 ms in human versus 190 ± 15 ms in rat brain at 9.4 T [1; 45; 46].

Differences in T₁ and T₂ relaxation times between human and rat brain at 9.4 Tesla may be explained in part by differences in the structural organization of the brain. For example, in the rat brain local susceptibility gradients are likely smaller than in humans, and therefore spin dephasing would be less pronounced than in human brain which would cause longer T₂.

Comparison of Linewidths at Different B₀ Values

In the present study, the minimum in vivo linewidth measured on the methyl resonance of total creatine was ~12.5 Hz after FAST(EST)MAP was used to adjust all first- and second-order shim terms. Reducing the voxel size did not yield a narrower linewidth, indicating that B₀ macroscopic shimming was optimal. This minimum achievable linewidth was markedly larger than at lower fields [9] (Figure 5) with a linear relationship between the minimum achievable linewidth and B₀, corresponding to a linewidth increase of 1.35 ± 0.02 Hz/Tesla. The linewidth measured in humans at 9.4 T was also broader compared to that reported in adult rat brain (~9 Hz [1]) at 9.4 T, consistent with the shorter T₂ values observed in human brain.

Figure 5.

Microscopic susceptibility effect (Δν*) as a function of field strengths based on the experimentally determined linewidth (Δν1/2) of the methyl group of total creatine in human brain. Δν* is defined as (Δν_1/2 − (πT₂)⁻¹ − Δδ) where T₂ is the transverse relaxation of the CH₃ group of total creatine. The chemical shift difference between Cr and PCr resonances (Δδ) was set to 0.0033 ppm as determined on high-resolution phantom data. Data points at 4 and 7 T were taken from [9].

Line broadening due to microscopic susceptibility effect (Δν^*), T₂ relaxation effects ((πT₂)⁻¹) and chemical shift difference (Δδ) were calculated based on the in vivo measured linewidth of the CH₃ group of total creatine for 3 different fields strengths (Figure 5) using the following equation: Δν^* = (Δν_1/2 − (πT₂) ⁻¹ − Δν). This result clearly shows that microscopic susceptibility effects are the main factor contributing to the line broadening as the field strength increases since the contribution of the relaxation pathways via dipolar mechanisms to T₂ are relatively small compared to the relaxations induced by local magnetic susceptibilities [44].

Future Perspectives

The main objective of this paper was to show the feasibility of acquiring high quality ¹H MRS in the human brain at 9.4 T and to provide reference data for the minimum linewidth and for relaxations times in human brain at 9.4 T. Further work is needed to assess potential sensitivity gains at 9.4 T compared to 7 T. The sensitivity of the 8-channel microstrip RF coil used in the present study may not be optimal for MRS in the occipital lobe. Comparing sensitivity at 9.4 T and 7 T will require careful matching of RF coils and other acquisition conditions on different systems, similar to what was done recently when comparing 7 T with 4 T [8]. Given the significant gains at 7 T compared to 4 T, one can reasonably expects sensitivity gains at 9.4 T compared to 7 T. This, however, remains to be demonstrated, and the potential gains may vary depending on brain location. Sensitivity gains would allow measurement in deep brain structures, such as hippocampus, and smaller volumes.

Other spectroscopic measurements may also benefit from 9.4 Tesla in spite of the added challenges. For example, spectral editing would benefit from ultrahigh field since signals separation arising from larger chemical shift dispersion increases with field strength [47], although some of the gain in selectivity would be offset by signal loss due to shorter T₂ relaxation times. Improved sensitivity in spectroscopic imaging (i.e. smaller voxel with the ability to distinguish different tissues types) is also expected, especially in the view of recent result reported at 7 T [48].

Conclusion

In this study, we demonstrated high quality ¹H NMR spectra measured in vivo in the human brain at 9.4 T and these spectra were very similar to that obtained in the past decade in rodent brains at the same field strength [1; 19; 45; 49]. We also observed that T₂ relaxation times of metabolites were shorter at 9.4 T than at 7 T, although signal loss due to T₂ relaxation was still largely negligible at an echo time of 8 ms. The minimum achievable linewidth increased with B₀ field strength (1.35 Hz/Tesla from 1.5 T to 9.4 T) and our results indicate that, at very high field of 9.4 T, B₀ microsusceptibility effects are the main contributor to the minimum linewidth that can be achieved in vivo.