|Home | About | Journals | Submit | Contact Us | Français|
Fast chemical shift imaging techniques are advantageous in metabolic imaging of hyperpolarized compounds due to the limited duration of the signal amplification. At the same time, reducing the acquisition time in hyperpolarized imaging does not necessarily lead to the conventional penalty in signal-to-noise ratio that occurs in imaging at thermal equilibrium polarization levels. Here a high-performance gradient insert was used in combination with undersampled spiral chemical shift imaging to increase either the imaging speed or the spatial resolution of hyperpolarized 13C metabolic imaging on a clinical 3T MR scanner. Both a single-shot sequence with a total acquisition time of 125 ms and a 3-shot sequence with a nominal in-plane resolution of 1.5 mm were implemented. The k-space trajectories were measured and then used during image reconstruction. The technique was applied to metabolic imaging of the rat brain in vivo after the injection of hyperpolarized [1-13C]-pyruvate. Dynamic imaging afforded the measurement of region-of-interest-specific time courses of pyruvate and its metabolic products, while imaging at high spatial resolution was used to better characterize the spatial distribution of the metabolite signals.
Hyperpolarization of metabolically active substrates such as [1-13C]-pyruvate permits new approaches to the investigation of in vivo metabolism using magnetic resonance spectroscopy (MRS) and chemical shift imaging (CSI) (1,2). As the transient nature of the signal amplification necessitates fast data sampling techniques, particularly if the goal is to perform time-resolved measurements of metabolite kinetics, various methods have been developed or adapted for fast metabolic imaging of hyperpolarized compounds (3–11). Among these, spiral CSI (spCSI) (12) is a very efficient acquisition scheme that theoretically permits single-shot CSI by simultaneously encoding 1D spectral and 2D spatial information. However, the spectral bandwidth (SW) is limited by gradient strength and slew rate, which is particularly problematic for hyperpolarized 13C imaging because the lower 13C gyromagnetic ratio corresponds to a four-fold decrease in gradient performance when compared to proton CSI. Taking advantage of the sparse spectra, which occur in metabolic imaging of hyperpolarized [1-13C]-pyruvate, spectral undersampling can be applied to reduce the number of spatial interleaves. A 3-shot version of spCSI has been successfully implemented on a clinical 3T MR scanner for sub-second metabolic imaging in a rat with no penalty in signal-to-noise ratio (SNR) or spatial resolution compared to conventional phase encoded CSI (5,13). The aim of the current work was to further reduce the number of interleaves and achieve single-shot metabolic imaging by using a high-performance gradient insert (14). In an alternative approach, the higher gradient strength and slew rate of the insert were used to increase the spatial resolution of 3-shot spCSI. The technique was applied to dynamic and high-resolution metabolic imaging of the rat brain in vivo after the injection of hyperpolarized [1-13C]-pyruvate.
The experiments were performed on a clinical 3T Signa MR scanner (GE Healthcare, Waukesha, WI) equipped with a high-performance insert gradient coil (peak strength = 600 mT/m; peak slew rate = 3200 T/m/s) (14). The gradient system was operated at a maximum amplitude of 500 mT/m with a slew rate of 1865 mT/m/ms to minimize vibration and acoustic problems. A custom-made quadrature proton birdcage coil (inner diameter: 44 mm) was used for both radiofrequency (RF) excitation and signal reception in phantom experiments measuring the k-space trajectories. A dual-tuned (1H/13C) quadrature coil (inner diameter: 50 mm, length: 70 mm), operating at 127.7 MHz and 32.1 MHz, respectively, was used for the in vivo experiments. The coil was based on a previously published design (15), but with a second half-Helmholtz unit added to provide quadrature operation in the proton mode.
Substrate, polarization procedure using a HyperSense system (Oxford Instruments Molecular Biotools, Oxford, UK), and handling of the healthy male Wistar rats (168 – 246 g body weight) were the same as described in (5) unless stated otherwise. The solvent solution consisted of 40 mM Tris, 80 mM NaOH, 100 mg/L Disodium EDTA and 58 mM NaCl. The final 80-mM [1-13C]-pyruvate solution had a pH of 7.4 and the liquid-state polarization was approximately 20% at dissolution. Targeting a dose of 1 µmol/g, 2.1–3.0 mL were injected manually through a tail vein catheter at a rate of approximately 0.2 mL/s, followed after a 1 to 2 s delay by a flush of 0.5 mL saline with 1% heparin. The time delay between dissolution and start of injection ranged from 19 to 29 s. All procedures were approved by the Institutional Animal Care and Use Committees at Stanford University and SRI International.
For the in vivo experiments, single-shot fast spin-echo (FSE) proton MR images in axial, sagittal, and coronal orientations were acquired as anatomical references for prescribing the 13C-spCSI experiments. In each direction, up to 45 2-mm slices were acquired with 0-mm separation and a nominal in-plane resolution of 0.47 mm (256×192 matrix, echo time (TE)/repetition time (TR) = 39/1492 ms). Additionally, dual-echo FSE images (0.25-mm in-plane resolution, 256×192 matrix, fifteen 1-mm slices, echo train length = 8, TE1/TE2/TR = 11/57/5000 ms) were acquired in axial (animal coronal) direction matching the prescription in the spCSI experiments. The B0 homogeneity over the brain was manually optimized with a point-resolved spectroscopy sequence by minimizing the line width of the unsuppressed water signal with the linear shim currents.
The transmit 13C RF power was calibrated using a reference phantom containing a 8-M solution of 13C-urea in 80:20 w/w water:glycerol and 3 µL/mL Gd-chelate (OmniScan™, GE Healthcare, Oslo), which was placed on top of the animal. With a volume of only 0.8 mL for the urea solution, the effect of the phantom on the coil loading and, hence, the pulse calibration, is negligible. Based on the urea signal, the center frequency was set at approximately 177 ppm. The phantom was also used as a concentration reference to normalize the metabolic images. It was removed prior to the first spCSI measurement because the aliased urea resonance at 163.5 ppm would partially overlap with the pyruvate signal. Although the phantom was next to the animal, quantitation of pyruvate (Pyr) would have been affected due the distorted point spread function of urea (aliased once compared to Pyr).
In the first set of experiments, single-shot spCSI was applied to dynamic metabolic imaging of the rat brain. The spiral gradient waveforms were generated using the design algorithm of Glover (16) for a targeted field-of-view (FOV) of 43.5×43.5 mm2 with a 16×16 matrix corresponding to a nominal in-plane resolution of approximately 2.7 mm. Including the refocusing lobe that returns the trajectory to the center of k-space, the duration of each spiral gradient lobe was 3.568 ms leading to a spectral width of 280 Hz. Therefore, the signals from alanine (Ala) and pyruvate hydrate (Pyh) were within the spectral width, whereas the signals from Pyr and lactate (Lac) were aliased once and the bicarbonate (Bic) signal was aliased twice. The TE was 3 ms and with 32 lobes, the total acquisition time per image was 125 ms. A 10-mm axial (animal coronal) slice through the front and middle of the rat brain was excited with the slice center approximately 1.2 mm anterior of the bregma according to the atlas of Paxinos and Watson (17). Sixteen data sets were acquired at 3-s intervals starting 9 s after the start of injection. A variable-flip-angle scheme (18) was applied for the 16 excitations to preserve the longitudinal magnetization. Although a higher temporal sampling rate of up to eight samples per second would have been possible for the single-shot acquisition, the applied sampling scheme was chosen as a compromise between the number of samples and the SNR for each sample. Each measurement was repeated 3 times per animal (3 animals) to increase SNR and assess reproducibility. The interval between injections was approximately 90 min.
In the second set of experiments, high-resolution metabolic imaging of the rat brain was performed to reduce partial volume effects using 3-shot spCSI with a targeted FOV of 48×48 mm2 and a matrix size of 32×32. Number of echoes and SW were the same as in the dynamic imaging experiments. The slice location was also the same, but with the thickness reduced to 5 mm. Given the lower signal due to the smaller voxel size, the excitation flip angles for the three interleaves were 35°, 45°, and 90°, respectively, to use all the longitudinal magnetization. With a TE of 3.2 ms, the total acquisition time per image was 375 ms and the data from four injections into a single animal were acquired under the same conditions and then averaged.
Currently the gradient driver parameters of the system have not been optimally calibrated to the impedance of the insert gradient, which leads to deviations of the k-space trajectory from the targeted design. Therefore, the trajectories for spCSI were measured using the Fourier transform method suggested by Alley et al. (19). The measurements were performed on a spherical water phantom (38-mm inner diameter) to increase the SNR. Only the trajectory for one spiral lobe was measured and trajectories along the x and y-axes were measured separately. For each gradient waveform, two data sets were acquired with the waveform inverted for the second acquisition. Measurements were performed for a FOV of 50 mm and 256 phase encoding steps.
All data processing was performed using custom software written in Matlab (MathWorks Inc., Natick, MA). For k-space trajectory measurements, the data were processed as described in (19) with the trajectories calculated from the phase of the data after fast Fourier transform (FFT). The corresponding gradient waveforms were obtained by differentiating the measured k-space trajectories.
Unless stated otherwise, the spCSI data were processed using the spectral tomosynthesis reconstruction as described in (5) with the measured k-space trajectories that were corrected for the different gyromagnetic ratios of proton and 13C nucleus. The data were apodized in the time dimension with a 10-Hz Gaussian line broadening. For 3-shot spCSI, the reconstruction included an additional step correcting for phase inconsistencies of the data acquired with the three spatial interleaves (20). Metabolic images were calculated by peak integration in absorption mode with integration intervals of 36 Hz for Pyr, Lac, and Ala. The integration window for Bic was decreased to 18 Hz to reduce contributions from the nearby Pyr peak which was only separated by approximately 40 Hz due to the spectral aliasing. No metabolic images of pyruvate hydrate were calculated because of its partial overlap with pyruvate due to the applied spectral undersampling scheme. However, the partial overlap results in only minor contributions of pyruvate hydrate to the metabolic images calculated for pyruvate because of their small ratio (< 8%) and the blurred pointspread function of pyruvate hydrate (13). The metabolic images were normalized to the signal intensity of the 8-M urea phantom expressed in institutional units (i.u.). In this notation, the urea phantom at thermal equilibrium (polarization of 2.64 ?10−4 %) has an intensity of 2.1 i.u. No correction for flip angle deviations due to B1 inhomogeneity was applied to the data. However, the animals were positioned with the brain approximately at the center of the RF coil to minimize potential flip angle variations. Individual data sets were normalized to a liquid state polarization of 22% and differences in transfer time from dissolution to start of injection were corrected using a T1 for Pyr of 60 s, which was measured in independent phantom experiments.
Figure 1 compares ideal and measured gradient waveform and the corresponding full k-space trajectories for the 3-shot spCSI case. The measured gradient waveform trailed by approximately one sample point and did not reach the full prescribed amplitude. The mean absolute difference for all three interleaves was 11.0 mT/m with a maximum of 23.5 mT/m. As a consequence, the distance between the spiral arms and the maximum value of the k-space trajectory were reduced. This corresponded to the FOV and the nominal voxel size being slightly larger than designed. In first approximation, the scaling factor was 1.036. Similar results were obtained for the single-shot spiral data but the deviations were less pronounced. The mean absolute difference for the gradient waveform was 9.8 mT/m with a maximum of 18.4 mT/m. FOV and voxel size were approximately 1.5% larger than designed.
Results from the time-resolved imaging experiments are shown in Figs. 2 and and3.3. The series of metabolic maps (Fig. 2a to 2d) illustrate the temporal dynamics and spatial distributions of Pyr and its metabolic products. The maps from a single animal, superimposed onto a high-resolution 1H-FSE image, were averaged over data from three injections. Taking into account the spatial apodization and the measured k-space trajectory, the effective voxel size was 130.8 µL compared to the nominal voxel size of 72.9 µL. While Pyr was highest in the vasculature, it was also detected in the brain. Lac signal was detected in the brain already at the first time point and quickly increased over the next 12 s. Although Lac could have also been produced elsewhere in the body, e.g., in the heart, and then have crossed the blood-brain barrier (BBB), the different ratios of brain and vasculature signal, e.g. 0.6 for Pyr and 1.5 for Lac at 21 s, suggest that most Lac was produced in the brain. Ala was high in muscular tissue of the jaw and tongue. Bic was only detected in the brain, in particular in the cortex. The observed Bic was most likely formed through decarboxylation of pyruvate to acetyl coenzyme A. However, another pathway would comprise first the carboxylation of pyruvate to oxaloacetate with subsequent turnover in the tricarboxylic acid cycle. The bicarbonate would later be formed through decarboxylation of isocitrate to alpha-ketoglutarate. The time curves of all four metabolites from a region-of-interest (ROI) predominantly in the cortex are shown in Fig. 3. The curves were calculated from a single animal averaged over three injections and from three animals with three injections for each animal. The error bars given by the corresponding standard deviations demonstrate good reproducibility, both between multiple injections and between animals.
Due to the low spatial resolution, the time course of Pyr was also affected by signal contributions from Pyr in the sagittal sinus. This may explain the relatively large variations of Pyr in the early time points as the vascular Pyr signal was the most sensitive to small variations during the course of the manual bolus injections. Lac and Bic exhibited similar time courses with the maximum for Bic approximately 10 s delayed compared to the Lac. Similar to Pyr, the time course of Ala in the ROI, which showed a steady increase, was dominated by signal contributions from the tissue surrounding the brain due to partial volume effects. The continuing rise of the alanine signal was mainly due to the relatively short observation window (until 52 s after start of injection). Although alanine could still be produced after that time frame, the loss of polarization due to T1 decay ultimately dominates. The intensity increase at the last two time points, particularly noticeable in the Lac dynamics, presumably was due to in-flow, i.e., from spins that have not experienced all excitation pulses. Although the inflow is a cumulative effect and can be present thought the whole acquisition window, its effect is most obvious during the last time points, because of the rapid increase in excitation flip angle at those time points due to the applied variable-flip-angle scheme.
High-resolution metabolic imaging was performed to reduce partial volume effects and to further investigate the spatial origin of the metabolite signals. To compensate for the smaller voxel size, a variable flip angle scheme was used for the three-shot acquisition that excites the full longitudinal magnetization in the imaging slice at a single time point. The nominal voxel size was 11.3 µL and the effective voxel size was 19.2 µL. The metabolic maps (Fig. 4) acquired 27 s after start of injection show a similar metabolite distribution as the dynamic data, but with reduced contamination from signal outside the brain. Compared to the corresponding values for the low-resolution images at the same time point (cf. Fig. 3), the average signal intensities in the cortex ROI were similar for Lac (1.2 i.u.) and even higher for Bic (0.3 i.u.), but they were considerably reduced for both Pyr (1.5 i.u.) and Ala (0.0 i.u.). The fact that no alanine was detected in the brain is consistent with measurements by Erakovic et al. (21) who report activity levels for alanine aminotransferase in multiple regions of the rat brain approximately 50 times lower than for lactate dehydrogenase.The greater heterogeneity of the lactate distribution in the brain in the high-resolution image compared to singe-shot data at the same time point is most likely due to the lower SNR. However, it could also reflect a more accurate depiction of the true metabolite distribution.
The presented data demonstrate the feasibility of single-shot hyperpolarized 13C metabolic imaging in vivo on a clinical MR scanner with a total acquisition time per image of only 125 ms. For comparison, eight spatial interleaves would be necessary to achieve the same spatial resolution and spectral width using the clinical gradient set (40 mT/m, 150 mT/m/ms). Although such short acquisition times might not be necessary for most applications, it allows greater flexibility in trading off temporal resolution against spatial coverage. Using an insert gradient system on a clinical scanner affords the high resolution necessary for small animal imaging while also facilitating the translation of this technology to the clinic. An extension to 3D spCSI is straightforward and should make volumetric metabolic imaging with acquisition times on the order of a few seconds possible. The application of this method to dynamic metabolic imaging of hyperpolarized [1-13C]-pyruvate in the rat brain demonstrates that transport of Pyr through the BBB and subsequent conversion into Lac and Bic is fast enough to be observed during the lifetime of the hyperpolarized signal. The dynamic curves can be used to estimate BBB transport and/or metabolic conversion rates (22). However, the high-resolution data suggest that non-brain tissue contamination due to partial volume effects needs to be taken into account.
We thank Oliver Hsu, B.A., and Evan Nunez, B.A., for assistance in animal handling and monitoring. This study was supported by the Lucas Foundation and National Institutes of Health Grants RR09784, AA05965, AA13521-INIA, AA018681, and EB009070.