MRS ArtifactsWhat MRS artifacts should we know about?
|
|
As with other specialized MR techniques a variety of artifacts may affect spectral quality. Some of the most important are briefly described below.
|
Poor Shimming
Beneficial effect of shimming |
Shimming is an iterative adjustment process to improve magnetic field homogeneity within a spectroscopic volume. First order shimming is now entirely automated on clinical scanners, but occasionally second order (manual) shimming may be needed. As shown left, poor shimming can be recognized by spectral lines that are too wide, short, and poorly separated.
|
|
Incomplete water suppression
Because metabolites of interest exist in such low concentrations in tissues, suppression of water signal is an essential component of nearly all ¹H spectroscopic examination. When water suppression fails or proves inadequate, the metabolite spectra may be inapparent or lost in background noise.
|
|
Absorption and dispersion modes of an MRS spectrum
Phasing Errors
The time domain NMR signal in a spectroscopy experiment is typically considered to be an exponentially decaying sine wave or FID (Free Induction Decay). After Fourier transformation, the FID becomes a complex-valued function in the frequency domain known as a Lorentzian. Lorentzian functions have both "real" and "imaginary" components in a fixed 90° (quadrature) relationship to one another. The relative values of these two components, however, depend on the initial (and generally unknown) phase offset (φ) of the FID. In the ideal case where φ = 0, the real component of the Lorentzian has the desired shape of a narrow, upright spectral peak known as the absorption mode lineshape. The imaginary component becomes an generally undesirable biphasic waveform known as the dispersion mode lineshape.
The time domain NMR signal in a spectroscopy experiment is typically considered to be an exponentially decaying sine wave or FID (Free Induction Decay). After Fourier transformation, the FID becomes a complex-valued function in the frequency domain known as a Lorentzian. Lorentzian functions have both "real" and "imaginary" components in a fixed 90° (quadrature) relationship to one another. The relative values of these two components, however, depend on the initial (and generally unknown) phase offset (φ) of the FID. In the ideal case where φ = 0, the real component of the Lorentzian has the desired shape of a narrow, upright spectral peak known as the absorption mode lineshape. The imaginary component becomes an generally undesirable biphasic waveform known as the dispersion mode lineshape.
Overly wide and short peaks in a spectrum commonly result when the absorption and dispersion modes are mixed. Through a procedure known as phasing, the spectrum is adjusted to display the optimal absorption mode waveform. The initial step (frequency-independent or zero-order phasing) adjusts the phase offset to optimize a single spectral line. But because the phase offsets of different peaks may vary as a function of frequency, a second step (linear or first-order phasing) is also often used. The process may be repeated in an iterative fashion until the best possible spectrum is obtained. Phasing is largely automated on today's scanners, but sometimes requires manual intervention.
Mismapping of lactate (yellow) and myo-Inositol (blue) due to chemical shift displacement.
Chemical Shift Displacement
Chemical shift displacement artifacts between metabolites in MRS are identical in nature to familiar fat-water artifacts seen on conventional MRI. Consider, for example, a brain voxel containing lactate (Lac) and myo-inositol (mI). Their chemical shift difference is 2.3 ppm, which translates into a frequency difference of about 300 Hz at 3.0T. If the transmit RF-bandwidth defining the voxel were 1500 Hz, then the two metabolites would be spatially mismapped about 20% (300/1500) relative to one another. Because this displacement would occur in all 3 directions, only (1−0.20)³ = 51% overlap would occur in the entire voxel, even though the two resonances should coincide exactly. Chemical shift displacement artifacts are most problematic with single voxel spectra at higher magnetic fields. The general solution is to employ larger bandwidth RF pulses and stronger imaging gradients.
Chemical shift displacement artifacts between metabolites in MRS are identical in nature to familiar fat-water artifacts seen on conventional MRI. Consider, for example, a brain voxel containing lactate (Lac) and myo-inositol (mI). Their chemical shift difference is 2.3 ppm, which translates into a frequency difference of about 300 Hz at 3.0T. If the transmit RF-bandwidth defining the voxel were 1500 Hz, then the two metabolites would be spatially mismapped about 20% (300/1500) relative to one another. Because this displacement would occur in all 3 directions, only (1−0.20)³ = 51% overlap would occur in the entire voxel, even though the two resonances should coincide exactly. Chemical shift displacement artifacts are most problematic with single voxel spectra at higher magnetic fields. The general solution is to employ larger bandwidth RF pulses and stronger imaging gradients.
Multi-voxel Spectral Contamination
Spectral contamination refers to the erroneous assignment of metabolite signals to the wrong voxel. As described above, chemical shift displacement is one source of localized spectral contamination prevalent in single-voxel spectroscopy. In multi-voxel spectroscopy a more insidious form of spectral contamination occurs that may be extend many voxels away from its source. This is a type of aliasing similar to the more familiar Gibbs (ringing) artifact seen on conventional MRI.
Point spread function (psf)
Because typical spectroscopic grids are in the range of 8x8 to 32x32, a significant truncation of the digitized signal must take place. After Fourier transformation the spectral peaks arising from a voxel spread out in ripples to its neighbors, described mathematically as the point spread function (psf). This effect can be reduced (but not completely eliminated) through digital filtering techniques and increasing the number of voxels.
References
Bottomley PA. The trouble with spectroscopy papers. Radiology 1991; 181:344-350. (An old but classic reference by one of the fathers of clinical MRS. Many of the points he makes are still valid 25 years later.)
Cianfoni A, Law M, Re TJ, et al. Clinical pitfalls related to short and long echo times in cerebral MR spectroscopy. J Neuroradiol 2011; 38:69-75. (pitfalls in interpretation of brain spectra, not really artifacts, but worth a look)
Juchem C, de Graaf R. Bo magnetic field homogeneity and shimming for in vivo magnetic resonance spectroscopy. Anal Biochem 2016; 1-13.
Kreis R. Issues of spectral quality in clinical 1H-magnetic resonance spectroscopy and a gallery of artifacts. NMR Biomed 2004;17:361-381. [DOI LINK] (excellent review of artifacts, also discussion of SNR)
Murali N. Fourier Transform. Notes for Lecture 3, Chem 542, NMR spectroscopy: Principles and applications. Spring, 2010. Available at this link. (an excellent description of absorption and dispersion spectra and mathematical derivation of the Lorentzian).
Bottomley PA. The trouble with spectroscopy papers. Radiology 1991; 181:344-350. (An old but classic reference by one of the fathers of clinical MRS. Many of the points he makes are still valid 25 years later.)
Cianfoni A, Law M, Re TJ, et al. Clinical pitfalls related to short and long echo times in cerebral MR spectroscopy. J Neuroradiol 2011; 38:69-75. (pitfalls in interpretation of brain spectra, not really artifacts, but worth a look)
Juchem C, de Graaf R. Bo magnetic field homogeneity and shimming for in vivo magnetic resonance spectroscopy. Anal Biochem 2016; 1-13.
Kreis R. Issues of spectral quality in clinical 1H-magnetic resonance spectroscopy and a gallery of artifacts. NMR Biomed 2004;17:361-381. [DOI LINK] (excellent review of artifacts, also discussion of SNR)
Murali N. Fourier Transform. Notes for Lecture 3, Chem 542, NMR spectroscopy: Principles and applications. Spring, 2010. Available at this link. (an excellent description of absorption and dispersion spectra and mathematical derivation of the Lorentzian).
Related Questions
What is a Gibbs artifact?
What is a chemical-shift artifact?
Why and how do you suppress water signal in MRS?
What is shimming and why is it needed?
What is a Gibbs artifact?
What is a chemical-shift artifact?
Why and how do you suppress water signal in MRS?
What is shimming and why is it needed?