Multiplexed edited MRS – more than one edited experiment combined in a single acquisition – involves acquiring subspectra with four or more distinct signal profiles. This technique therefore requires a tailored approach for correcting frequency and phase errors associated with participant head motion and scanner instability. Here, we demonstrate a novel alignment algorithm, termed multi-step frequency-and-phase correction (msFPC), designed to deal with the challenges of aligning individual transients in multiplexed edited data. Testing this method on simulated and in vivo datasets, msFPC was found to outperform other previously demonstrated algorithms (spectral registration and spectral registration with post hoc choline-creatine alignment).
In GABA-/GSH-edited HERMES (4), four sub-experiments are performed (A, B, C and D), in which editing pulses are applied to: both GABA (at 1.9 ppm) and GSH (at 4.56 ppm) (A: ONGABA, ONGSH); GABA only (B: ONGABA, OFFGSH); GSH only (C: OFFGABA, ONGSH); or neither (D: OFFGABA, OFFGSH). The combinations (A + B – C – D) and (A – B + C – D) give GABA- and GSH-edited spectra, respectively.
Simulated in vivo HERMES data were generated using FID-A (5): 320 transients; TE = 80 ms; 2048 data points; 2 kHz spectral width. Gaussian noise was added to each free induction decay (FID) to approximate in vivo signal-to-noise ratios, with known frequency/phase offsets added as previously described (6). In vivo HERMES datasets from three imaging centers were also used (acquisition parameters as above except: TR = 2000 ms; 27- to 46.7-mL voxels).
Several FPC algorithms were compared against msFPC: no correction (NC); spectral registration (SR) (6); and spectral registration with post hoc choline- (Cho-) creatine (Cr) alignment (SR+CC) (4):
NC: No FPC of frequency/phase errors.
SR: Each mth FID Sm(t) is fitted to a reference R(t) by adjustment of its frequency (f) and phase (φ) using nonlinear least-squares minimization:
$$\arg\min\parallel R(t)-G_m(t,f,\phi)\parallel_2$$
where
$$G_m(t,f,\phi)=S_m(t)\exp[2\pi(ft+\phi/360)]$$
R(t) is chosen as the median across the number of FIDs. Global frequency/phase adjustment is also performed in the frequency domain by fitting a two-Lorentzian model to the Cho-Cr signals in the sum of all subspectra.
SR+CC: SR is implemented as above and then the averaged GSH-ON and GSH-OFF subspectra are aligned in the frequency domain. The Cho-Cr signals are shifted in frequency and phase until the standard deviation of the Cho subtraction artifact is minimized.
msFPC: msFPC (Fig. 1) involves segregating each dataset into its four sub-experiment sets Si,m(t), where i $$$\in$$$ {A, B, C, D}. The frequency/phase offsets in each set are estimated using SR as above. The distributions of frequency/phase offsets for each sub-experiment set are then modeled using a Cauchy probability density function. The center values of the Cauchy functions are subtracted from the respective set of estimated frequency/phase offsets, which are then applied to the uncorrected FIDs. The real frequency-domain Cr signal maxima are aligned to remove residual frequency errors. A Cho-Cr model is fitted to the sum spectrum to remove residual phase error.
Alignment quality in the simulated dataset was assessed by the normalized root sum square error Qrsse for each FPC algorithm:
$$Q_{rsse}=1-\sqrt{\frac{\sum(p_{est,m}-p_{true,m})^2}{\sum p_{true,m}^2}}$$
where py,m is either the estimated or true frequency offset, Δf, or phase offset, Δφ, of the mth FID. Qrsse = 1: perfect alignment; Qrsse < 0: worse alignment; Qrsse = 0: no alignment. Alignment quality in the in vivo datasets was assessed by an alignment quality metric Qpre/post based on subtraction artifacts in the pre-/post-alignment edited spectra:
$$Q_{pre}=1-\frac{\sigma(SA_{pre})-\sigma(\epsilon_{pre})}{med[\sigma(SA_{pre})]-\sigma(\epsilon_{pre})}$$
$$Q_{post}=1-\frac{\sigma(SA_{post})-\sigma(\epsilon_{post})}{med[\sigma(SA_{pre})]-\sigma(\epsilon_{post})}$$
where σ(SAy)
is the standard deviation of the Cho subtraction artifact pre- or post-alignment, σ(εy) is the standard deviation
of the noise signal and med denotes the median across the group. Median Qpost > 0: improved
alignment; median Qpost < 0: worse alignment;
Qpost = 1: complete removal of Cho subtraction artifact (noise
only).
1. Harris AD, Saleh MG, Edden RAE. Edited 1H magnetic resonance spectroscopy in vivo: Methods and metabolites. Magn. Reson. Med. 2017;77:1377–1389. doi: 10.1002/mrm.26619.
2. Harris AD, Glaubitz B, Near J, John Evans C, Puts NAJ, Schmidt-Wilcke T, Tegenthoff M, Barker PB, Edden RAE. Impact of frequency drift on gamma-aminobutyric acid-edited MR spectroscopy. Magn. Reson. Med. 2014;72:941–8. doi: 10.1002/mrm.25009.
3. Chan KL, Puts NAJ, Schär M, Barker PB, Edden RAE. HERMES: Hadamard encoding and reconstruction of MEGA-edited spectroscopy. Magn. Reson. Med. 2016;76:11–19. doi: 10.1002/mrm.26233.
4. Saleh MG, Oeltzschner G, Chan KL, Puts NAJ, Mikkelsen M, Schär M, Harris AD, Edden RAE. Simultaneous edited MRS of GABA and glutathione. Neuroimage 2016;142:576–582. doi: 10.1016/j.neuroimage.2016.07.056.
5. Simpson R, Devenyi GA, Jezzard P, Hennessy TJ, Near J. Advanced processing and simulation of MRS data using the FID appliance (FID-A)-An open source, MATLAB-based toolkit. Magn. Reson. Med. 2017;77:23–33. doi: 10.1002/mrm.26091.
6. Near J, Edden R, Evans CJ, Paquin R, Harris A, Jezzard P. Frequency and phase drift correction of magnetic resonance spectroscopy data by spectral registration in the time domain. Magn. Reson. Med. 2015;73:44–50. doi: 10.1002/mrm.25094.