Synopsis

Keywords: Data Acquisition, Muscle

Quantitative comparison of novel, rosette trajectory UTE (70 μs) and conventional weighted Cartesian 3D 31P MRSI sequences is performed at 3T in quadriceps muscle. After previous validation, five healthy volunteers were scanned by both sequences without interruption. Fitted metabolite maps and SNR calculations of PCr and ATP signals across selected slices and voxels displayed competitive performance between each acquisition. Retrospective compressed sensing (CS) acceleration results suggest this UTE MRSI may enable faster, higher resolution 31P metabolite mapping in leg muscle and other organs of interest.

Introduction

Plentiful literature has highlighted the value of phosphorus-31 (³¹P) MRS metabolites in providing non-invasive measurements of tissue pH, muscular energy metabolism, and disease prognosis¹. Unfortunately, relatively short T2 relaxation times and low in vivo abundance introduce significant restrictions on SNR and resolution within a “clinically feasible” length scan, especially when working below 7T². Following previous validation³, a rosette UTE 3D sequence was adapted to ³¹P MRSI and observed to produce comparable results to gold-standard conventional 3D CSI given the same amount of scanning time. Due to the rosette k-space trajectory’s relatively incoherent sampling pattern, it is possible that high under-sampling and compressed sensing (CS) reconstruction techniques could greatly accelerate this acquisition⁴. A comparative analysis was performed in five healthy volunteers to assess relative performance between the full conventional scan, full UTE scan, and CS accelerated UTE reconstruction.

Methods

All data acquisition occurred on a 3T Siemens Prisma scanner at the Purdue MRI facility. The same 8-channel, dual-tuned ¹H/³¹P phased array coil (2 parallel plates with 4 ³¹P channels each)⁵ was used in all sessions. Five healthy volunteer subjects (BMI = 26 ± 2 kg/m²; age = 29 ± 5 years; 2 f / 3 m) were positioned feet-first and supine, with the upper thighs tightly surrounded by the coil plates. Figure 1 summarizes the two sets of sequence parameters while Figure 2 illustrates coil setup. The adjustment volume was manually positioned (covering both upper legs) and linewidth manually minimized using a 3D GRE abdomen field map and interactive SIEMENS shimming (VE11c). The total scanning process required approximately 90 minutes for each subject
Raw (TWIX) data files were exported for reconstruction and pre-processing in MATLAB (Mathworks, Natick, USA). Gridding and FFT were completed using the nonuniform FFT (NUFFT) method. Data were Hanning filtered and coil-combined using singular value decomposition. Spectra from both acquisitions were zero-order phased by maximizing the real integral of the PCr peak (~0 ppm), and first-order phased by adding 2.3 ms and 70 μs delays respectively. Spectra were fitted within OXSA using AMARES methods^6,7. SNR was calculated by dividing PCr peak amplitude by the standard deviation of a noisy spectral region (+10 to +14 ppm, left of the PCr peak). Metabolite maps were generated using ratios of fit amplitudes, with voxels disqualified whenever CRLB estimates were exceedingly high.

Results

Figure 3 lists SNR calculations from the full acquisitions for each subject, along with point-spread-function (PSF) simulations. Acquired SNR appears extremely comparable between the methods across all five subjects.
Figure 4 contains example PCr SNR maps calculated from a single slice in one individual. Reported ratio values in healthy quadriceps/hamstrings range from 3.81-5.80², reasonably close to the OXSA AMARES fitting results across all subjects. Although the full metabolite map is closest to the full conventional CSI, the original structure remains clearly visible in the accelerated map. A greater degrees of CS acceleration under-sampling slightly changes the image scaling, however this can be corrected retrospectively. CS acceleration reconstruction also alters the original UTE rosette’s PSF, requiring accurate simulations to calculate the new effective voxel size and compensate SNR calculations.

Discussion and Conclusion

These initial results demonstrated the potential of our novel rosette UTE 31P 3D MRSI sequence for CS acceleration. Although the version shown here had a relatively large FOV with a coarse resolution, acquisition is possible with much smaller FOVs and finer resolution. This acceleration also does not impede the UTE sequence’s intrinsic avoidance of first-order phasing issues nor measurement of extremely fast T2 metabolites. Future work is focused on application to non-alcoholic fatty liver disease (NAFLD) and high resolution 31P brain metabolite mapping. In hepatology clinics, accelerated 3D MRSI with a large FOV might aid in quickly assessing NAFLD prognosis in overweight-to-obese patients. Similarly, a 2x acceleration factor could generate whole-brain, high-resolution metabolite maps in under 15 minutes of scan time.

Acknowledgements

Data acquisition was supported in part by the Indiana CTSI. The authors would also like to thank Dr. Anshuman Panda for his initial work with the 8-channel phased array coil.

References

¹Valkovic L, Chmelik M, and Krssak M. In-vivo ³¹P-MRS of skeletal muscle and liver: A way for non-invasive assessment of their metabolism. Anal Biochem. 2017;529:193-215.

²Meyerspeer M, Boesch C, Cameron D, et al. ³¹ P magnetic resonance spectroscopy in skeletal muscle: Experts' consensus recommendations [published online ahead of print, 2020 Feb 10]. NMR Biomed. 2020;34(5):e4246.

³Shen X, Ozen AC, Sunjar A, et al. Ultra-short T₂ components imaging of the whole brain using 3D dual-echo UTE MRI with rosette k-space pattern [published online ahead of print, 2022 Sep 25.] Magn Reson Med. 2022;10.1002/mrm.29451.

⁴Li Y, Yang R, Zhang C, et al. Analysis of generalized rosette trajectory for compressed sensing MRI. Med Phys. 2015;42:5530-5544.

⁵Panda A, Jones S, Stark H, et al. Phosphorus liver MRSI at 3 T using a novel dual-tuned eight-channel ³¹P/¹H H coil. Magn Reson Med. 2012;68(5):1346-1356.

⁶Purvis LAB, Clarke WT, Biasiolli L, et al. OXSA: An open-source magnetic resonance spectroscopy analysis toolbox in MATLAB. PLoS One. 2017;12:e0185356.

⁷Vanhamme L, van den Boogaart A, and Van Huffel S. Improved Method for Accurate and Efficient Quantification of MRS Data with Use of Prior Knowledge. J Magn Reson. 1997;129:35-43.