Yan Li1
1University of California, San Francisco, San Francisco, CA, United States
Synopsis
The accuracy of measuring brain metabolite levels highly relies on spectral quality. Assurance of quality control is required for translating proton MR spectroscopy (1H MRS) technologies into clinical studies. This lecture focuses on the basics of MRS methodology, artifacts with their causes and possible solutions, and quality control for single and multi-voxel 1H MRS.
Introduction
Proton magnetic resonance spectroscopy (1H MRS) is a powerful non-invasive tool for characterizing biochemical changes within tissue and has been widely used in psychiatric and neurologic diseases to estimate the relationship between metabolites and disease onset or progression. The accuracy of measuring brain metabolite levels highly relies on spectral quality [1-2]. Artifacts could interfere with the spectra and make the interpretation challenging [3]. Assurance of quality control is required for translating MRS technologies into clinical studies. The basics of 1H MRS methodology, artifacts, and quality control are discussed below. Basics of MRS methodology
A number of different strategies for measuring brain metabolites have been implemented. The basic elements required to obtain 1H MRS data comprise water suppression, lipid suppression, and spatial localization. The use of the semi-LASER sequence is currently recommended for single-voxel acquisitions [4]. Multi-voxel 1H MRS, MR spectroscopic imaging (MRSI), initially used standard phase encoding procedures to generate the spatial information, and now full slice or whole brain with fast acquisition techniques have been implemented. Long echo time methods (TE>130ms) provide more reliable estimates of the higher signal to noise metabolites. In the spectrum with short TEs (<40ms), it is possible to detect additional metabolites. Spectral editing separates signals from the metabolites of interest from partially overlapping resonances by taking advantage of the coupling patterns of protons, requiring two cycles of editing-on and editing-off. The presentation will discuss artifacts related to these acquisitions. Quality metrics for spectra data include signal to noise ratio (SNR), linewidth (full width at half maximum, FWHM, inversely proportional to T2*), and quantitative error estimates, such as Cramer-Rao lower bounds (CRLB) [2,3,4,5]. Artifacts
Examples of 1H MRS artifacts include 1) Artifacts related to motion [6]: Small changes in phase and frequency due to minor motions can be corrected in retrospective post-processing [7]. 2) Chemical shift displacement errors [2]: The variation in slice excitation between resonances can be minimized by using RF pulses with high bandwidths and/or increasing gradient strength. 3) Artifacts related to magnetic field imperfections [8]: 1H MRS is highly susceptible to magnetic field inhomogeneity. This results in the broadening of peak linewidths, inversely proportional to T2*. The non-uniform RF excitation can be reduced by using B1 insensitive pulses. 4) Artifacts from water signal [9]: Insufficient suppressed water can be removed in processing, while the sidebands of the water require improved water suppression or additional acquisitions. 5) Artifacts from lipid contamination [9]: This can be minimized for single-voxel 1H MRS data by careful choice of the selected volume and placement of outer volume suppression bands. The effect becomes more significant for 1H MRSI because of the point spread function for phase encoding. 6) Artifacts related to eddy currents: This effect can be eliminated by using unsuppressed water signals. Spurious echo, spectral aliasing, spatial aliasing, and subtract artifacts in spectral editing will also be discussed. Unlike MR images, most 1H MRS artifacts are not eye-catching, requiring specialized expertise.Quality control
A minimum SNR of 3 for normal singlets and a spectral resolution ≤ of 0.1ppm are recommended for single voxel spectroscopy data, along with efficient water and lipid suppression and artifact-free [2,4]. No specific criteria have not been established for multi-voxel 1H MRSI [1]. Similar single-voxel criteria can be applied but with the caution of using CRLB criteria [10]. The confirmation of spectral artifacts for 3D datasets by manual review could be time-consuming. Several recent studies have applied machine learning methods to filter and remove artifacts automatically [11-14]. Acknowledgements
No acknowledgement found.References
1. Maudsley et al. Advanced magnetic resonance spectroscopic neuroimaging: Experts' consensus recommendations. NMR in Biomed. 2019. e4309.
2. Oz et al. Clinical Proton MR Spectroscopy in Central Nervous System Disorders. Radiology. 2014.
3. Kreis et al. Issues of spectral quality in clinical 1H-magnetic resonance spectroscopy and a gallery of artifacts. NMR in Biomed. 2004. e891.
4. Oz et al. Advanced single voxel 1H magnetic resonance spectroscopy techniques in humans: Experts' consensus recommendations. NMR in Biomed. 2019. e4236.
5. Wilson et al. Methodological consensus on clinical proton MRS of the brain: Review and recommendations. Magn Reson Med. 2019.
6. Andronesi et al. Motion correction methods for MRS: experts' consensus Recommendations. NMR in Biomed 2020. e4364.
7. Near et al. Preprocessing, analysis and quantification in single-voxel magnetic resonance spectroscopy: experts' consensus recommendations. NMR in Biomed. 2019. e4257.
8. Juchem et al. B0 shimming for in vivo magnetic resonance spectroscopy:Experts' consensus recommendations. NMR in Biomed 2019. e4350.
9. Trac et al. Water and lipid suppression techniques for advanced 1H MRSand MRSI of the human brain: Experts' consensusrecommendations. NMR in Biomed 2020. E4459.
10. Kreis. The trouble with quality filtering based relative Cramer-Rao lower bounds. Magn Reson Med. 2016. 75:15-18.
11. Kyathanahally et al. Quality of clinical brain tumor MR spectra judged by humans and machine learning tools. Magn Reson Med. 2018;79(5):2500-2510
12. Menze et al. Mimicking the human expert: pattern recognition for an automated assessment of dataquality in MR spectroscopic images. Magn Reson Med. 2008;59(6):1457-1466.
13. Kyathanahally et al. Deep learning approaches for detection and removal of ghosting artifacts in MR spectroscopy. Magn Reson Med.2018;80(3):851-863.
14. Gurbani, et al. A convolutional neural network to filter artifacts in spectroscopic MRI. Magn Reson Med.2018;80(5):1765-1775.