3562

Improving robustness of hyperpolarized ¹³C MRSI using k-t spiral acquisitions

Erin Adamson¹, Benjamin Cox^1,2,3, and Sean Fain^1,4,5

¹Medical Physics, University of Wisconsin-Madison, Madison, WI, United States, ²Morgridge Institute for Research, Madison, WI, United States, ³Laboratory for Optical and Computational Instrumentation, University of Wisconsin-Madison, Madison, WI, United States, ⁴Radiology, University of Wisconsin-Madison, Madison, WI, United States, ⁵Biomedical Engineering, University of Wisconsin-Madison, Madison, WI, United States

Synopsis

Hyperpolarized ¹³C MRSI with chemical shift encoding relies on a priori knowledge of the N ¹³C-labeled metabolite resonances and N+1 echoes to fully determine the reconstruction. However, the a priori frequencies may shift due to B₀ inhomogeneity, local susceptibilities, or motion, causing spectral leakage, blurring, and biased quantitative measures. To address these uncertainties, we apply IDEAL-based k-t spiral imaging with field-of-view oversampling to further constrain the signal model-based reconstruction in situations where the echo-spacing may be suboptimal, and we test the results in digital simulations, phantom experiments, and in vivo studies of murine renal metabolism. Results support improved metabolic quantification.

Introduction

Many hyperpolarized ¹³C multi-spectral imaging methods rely on a priori knowledge of the number, N, and resonant frequencies of the ¹³C-labeled metabolites expected in the sample.¹ Commonly used IDEAL-based approaches acquire N+1 echoes spaced by ΔTE, chosen to maximize the metabolites’ number of signal averages (NSA).² However, B₀ inhomogeneity, local susceptibilities, and motion can all shift expected resonances, rendering selected ΔTEs suboptimal and causing spectral leakage and blurring in reconstructed images. This work investigates using k-t spiral MRSI³ to further constrain IDEAL reconstructions by providing additional phase information between more densely spaced spiral arms, thereby reducing artifact severity when the selected ΔTE is suboptimal for the distribution of experimental metabolite resonances.

Methods

k-t spiral imaging with a direct-IDEAL reconstruction³ was evaluated using digital simulations³, phantom, and in vivo experiments. For all k-t spiral acquisitions, gradients were derated to maintain a 30ms readout. For all digital simulations and phantom experiments, k-t spirals (nominal field-of-view (FOV)=32×32mm², nominal matrix=16×16) with oversampling factors η=1-7 were tested for good and poor choices of ΔTE, providing NSAs approximately equaling the number of echoes (NE) and ~0, respectively. Metabolite ratios were calculated for each image. All imaging was performed on a 4.7T small animal scanner (Agilent Technologies, Santa Clara, CA) using a dual-tuned ¹H/¹³C volume coil (Doty Scientific, Columbia, SC) and ¹³C-enriched compounds from Cambridge Isotope Laboratories, Inc. (Tewksbury, MA).

Digital simulations, performed in Matlab (R2015b, The MathWorks, Natick, MA), contained [1-¹³C]pyruvate (0ppm, relative amplitude (RA)=1.00), [1-¹³C]lactate (12.2ppm, RA=0.33), [1-¹³C]alanine (5.4ppm, RA=0.10), [1-¹³C]pyruvate-hydrate (8.6ppm, RA=0.20), and ¹³C-urea (-7.3ppm, RA=0.20) imaged with good (1.871ms) and poor (1.630ms) ΔTEs. Ideal (T₂*=∞, B₀=homogeneous, noise=0%) and realistic (T₂*=15ms, B₀ inhomogeneity=50Hz, noise=10%) conditions were tested.

For phantom experiments, microcentrifuge tubes filled with 9.4M ¹³C-urea (-0.2ppm), 2.1M [1-¹³C]glycine (9.4ppm), and 5.0M [1-¹³C]Na-lactate (25.6ppm) doped with 8.5mM, 5.4mM, and 6.3mM of gadobenate dimeglumine (Bracco Diagnostics, Inc., Princeton, NJ), respectively, were placed in a water-filled centrifuge tube and imaged with k-t spiral acquisitions (TR/TE₁=55/1ms, NE=4, slice thickness=10mm, FA=90°, bandwidth=250kHz, averages=800) with good (0.576ms) and poor (2.371ms) ΔTEs. ¹H spoiled gradient-echo T₁-weighted reference images and a field map, to correct for B₀ inhomogeneities, were acquired.

Dynamic, in vivo imaging of renal metabolism, compliant with our Institutional Animal Care and Use Committee, was performed in one ICR mouse in two separate experiments, each with two intravenous injections of hyperpolarized [1-¹³C]pyruvate. In one experiment, a poor ΔTE (1.630ms) was used. The second experiment (ΔTE=1.871ms) tested a ¹³C volume-excite/surface-receive coil setup (Doty Scientific, Columbia, SC), making ¹H field map estimation difficult due to coil coupling. For all experiments, k-t spiral acquisitions (TR/TE₁=140/1ms, NE=6, nominal FOV=48×48mm², nominal matrix=16×16, slice thickness=5mm, FA=10°, bandwidth=250kHz) were interleaved with slice-selective spectra to identify true metabolite resonances. The first and second injections used η=7 and η=1, respectively. ¹H fast spin-echo, T₂-weighted images were acquired for anatomical reference.

Results

Simulated images (Figure 1) demonstrate reduced spectral leakage for good ΔTE and increasing η for both ideal and realistic cases. Furthermore, the percent error of measured species ratios (Figure 2) trends towards 0% for good ΔTE and increasing η.

Phantom results (Figure 3) also demonstrate reduced spectral leakage for good ΔTE and increasing η. Ratio plots display the glycine/urea ratio stabilizing near the true ¹³C concentration ratio, 0.22, for η≥3 and η≥5 for good and poor ΔTEs, respectively. The Na-lactate/urea ratio also reaches a plateau for η≥3 and η≥5 for good and poor ΔTEs, respectively, but underestimates the true ¹³C concentration ratio of 0.53.

In vivo results for the poor ΔTE experiment demonstrate significant spectral leakage in the lactate images for η=1 (Figure 4A) and markedly improved image quality for η=7 (Figure 4B), with the same trend for the volume-excite/surface-receive coil experiment (Figure 5), although spectral leakage with η=1 was subtler in this more realistic scenario.

Discussion

Digital simulations, phantom, and in vivo experiments all displayed decreased or comparable spectral leakage for higher η acquisitions compared to the conventional IDEAL-spiral acquisitions (η=1). Simulations indicated that oversampling k-t space, even with a suboptimal ΔTE, reduced the error in measured metabolite ratios, particularly for higher SNR acquisitions. This was corroborated by phantom experiments in which species ratios stabilized near true ¹³C concentration ratios for higher η values, especially for glycine/urea, even with suboptimal ΔTEs. The larger bias towards underestimation of the Na-lactate/urea ratio was likely due to partial volume effects resulting from blurring.

Conclusion

The accuracy of hyperpolarized ¹³C metabolic imaging and quantification using IDEAL-spiral MRSI in the presence of unexpected resonance shifts can be improved with the use of oversampled k-t spiral acquisitions. This technique also increases the flexibility of ΔTE selection.

Acknowledgements

This project received support from an AAPM 2014 Graduate Fellowship, the Department of Medical Physics at the author’s institution, NIH/NIDDK R01 DK092454, and GE Healthcare.

References

1. Durst M, Koellisch U, Frank A, et al. Comparison of acquisition schemes for hyperpolarised ¹³C imaging. NMR Biomed. 2015;28(6):715-725.

2. Reeder S, Brittain J, Grist T, Yen Y. Least-squares chemical shift separation for ¹³C metabolic imaging. J Magn Reson Imaging. 2007;26(4):1145-1152.

3. Gordon J, Niles D, Fain B, Johnson K. Joint spatial-spectral reconstruction and k-t spirals for accelerated 2D spatial/1D spectral imaging of ¹³C dynamics. Magn Reson Med. 2014;71(4):1435-1445.

Figures

Figure 1: Simulated images for [1-¹³C]pyruvate, [1-¹³C]lactate, [1-¹³C]alanine, [1-¹³C]pyruvate-hydrate, and ¹³C-urea in both an ideal (A) case and a realistic (B) case (including 10% noise, chemical shift due to B₀ inhomogeneity, and a short T₂*). For each metabolite, images for good and poor ΔTEs are displayed (rows) for k-t spiral acquisitions with η=1-7 (columns). The degree of spectral leakage and blurring in the images is reduced with increasing η for all ΔTEs, and for all species, in both the ideal and realistic cases.

Figure 2: Plots of percent error as a function of spiral oversampling factor, η, in the simulated metabolite ratios compared to the known digital phantom ratios for both ideal (A) and realistic (B) cases. For all metabolite ratios, the good ΔTE, with higher NSA performance, resulted in lower errors in the simulated ratios in the ideal cases, with a similar trend in the realistic cases. All ΔTEs display a general reduction in error for all metabolite ratios with increasing η. The realistic case further demonstrates that higher SNR species ratios (e.g. lactate/pyruvate) also have lower measurement errors.

Figure 3: Phantom images (A) for each species are displayed for good and poor ΔTE acquisitions (rows) with oversampling factors η=1-7 (columns). Spectral leakage and blurring are minimal for good ΔTE and high SNR species (urea), but appear in lower SNR species (glycine and Na-lactate) for poor ΔTE and lower η. Image quality improves as η increases. Measured species ratio plots (B) display ratios plateauing near the true ¹³C concentration ratio (dashed line at 0.22) for glycine/urea, but underestimating truth (dashed line at 0.53) for Na-lactate/urea. Species' SNR plots (C) indicate SNR increasing with η, consistent with increased signal averaging.

Figure 4: Images are displayed from the first in vivo experiment, with intentionally poor ΔTE acquisitions. Twelve time frames of dynamic, hyperpolarized [1-¹³C]pyruvate (A,B) and [1-¹³C]lactate (C,D) axial images (color) are overlaid on anatomical T₂-weighted reference images (grayscale). The time since injection is listed at the top of each time frame. Images acquired with η=1 (A,C) show severe spectral leakage artifacts in the lactate images. Images from a separate injection and acquisition in the same scan session, with η=7 (B,D), display no apparent spectral leakage artifacts and clearly display lactate signal in the kidney bodies.

Figure 5: Images are displayed from the second in vivo experiment, acquired with an expected good ΔTE and a volume-excite/surface-receive coil setup known to contribute to local off-resonance. Two time frames of dynamic hyperpolarized [1-¹³C]pyruvate (A,B) and [1-¹³C]lactate (C,D) axial images (color) are overlaid on anatomical T₂-weighted reference images (grayscale). The time since injection is listed at the top of each frame. Images acquired with η=1 (A,C) show apparent spectral leakage artifacts in the lactate images. Images from a separate injection and acquisition during the same scan session, with η=7 (B,D), display noticeably fewer spectral leakage artifacts.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)

3562