Introduction

RS-COKE (Readout-Segmented COnsistent K-t space Eᴘsɪ)¹ is an EPSI (Echo Planar Spectroscopic Imaging) variant offering increased spectral width (SW) and is thus especially promising for spectral imaging of human brain metabolites at 7T. The increased spectral width is achieved both by readout segmentation² — allowing shorter echo-spacing — and by the COKE scheme^3,4 that ensures all same-phase-encode (PE) acquisitions use same-sign readout (RO) gradients — halving the effective dwell time to a single echo-spacing. In practice, COKE switches the N/2 Nyquist ghost artifacts from the temporal domain to the PE domain where they are more benign.
RS-COKE is, however, susceptible to image artifacts from mismatches between readout segments. Artifacts are especially pronounced when lipid suppression is off (to avoid adversely affecting the metabolites signal). In this study, we developed a procedure to measure and reduce the RO-segment mismatches, leading to improved spectral images and spectra. Corrections included i) frequency-dependent phase corrections; ii) a correction of the trajectories used by the reconstruction; and iii) a gradual transition between segments to smooth any residual mismatches. The corrections were verified on a special head-mimicking phantom, and on human volunteers.

Methods

Sequence/Acquisition

Figure 1 shows the RS-COKE pulse sequence, implemented on a 7T MAGNETOM Terra (Siemens Healthcare, Erlangen, Germany). It is readout segmented and employs sine-shaped readout gradients. The COKE scheme adds alternating PE blips between the RO gradients, reversing the sign of both PE blips and RO gradients every excitation, to achieve the trajectories shown. A VAPOR water-suppression module is included, as is a refocusing pulse, but no lipid suppression.
The sequence also includes three reference scans on water. The first, a conventional N/2 Nyquist ghost calibration with no PE gradients. Two additional scans, also with no PE gradients, are acquired for calibrating the trajectories. These are acquired at two adjacent RO segments with an additional prepended RO prephaser shifting both segments to overlap around $$$k_x=0$$$, where the signal is maximal. A separate gradient-echo (GRE) scan on water is used to calibrate the complex coil-combination weights per voxel.

Reconstruction

The basic reconstruction steps were:

• Time apodization (exponential with decay constant of 0.1 s).
• Hann filtering of all k-space dimensions.
• Fourier transform (FT) of all dimensions, using regridding along the non-uniformly sampled RO direction.
• An optimized linear coil combination,⁵ using complex weights derived from the water GRE calibration scan.

Off-Resonance Reconciliation
To resolve off-resonance phase differences along readouts employing opposite gradients, an FT along the time domain was applied (prior to re-gridding) and the linear phase per frequency (and per gradient-sign) was removed. To ensure phase continuity at segment joins, an additional phase — per-segment and per-frequency — was added to each readout.

Segment Trajectory Corrections
The ideal $$$k_x$$$ trajectory of segment $$$s$$$, sampled at time $$$t_n$$$ is given by $$k_x^{(s)}(t_n)=k_{x,\text{ideal}}^{(s=0)}(t_n)+s\cdot\Delta{k}_{\text{seg}},$$ where $$$\Delta{k}_{\text{seg}}$$$ is the $$$k_x$$$ shift between segments. This was replaced by a parametrized trajectory with extra tuning parameters $$$\eta$$$ and $$$\tau$$$ $$k_x^{(s)}(t_n)=\eta\cdot{k}_{x,\text{ideal}}^{(s=0)}(t_n+\tau)+s\cdot\Delta{k}_{\text{seg}}$$ or, for the reference scans (the same with an extra $$$+\Delta{k}_{\text{ref}}$$$ to center the overlap around $$$k_x=0$$$) $$k_x^{(s)}(t_n)=\eta\cdot{k}_{x,\text{ideal}}^{(s=0)}(t_n+\tau)+s\cdot\Delta{k}_{\text{seg}}+\Delta{k}_{\text{ref}}.$$ The parameters $$$\eta$$$ and $$$\tau$$$ were tuned so that odd signals of the trajectory reference scan (acquired during positive gradients) best match at the segments’ overlap — see figure 2. The even trajectories were further updated according to the N/2 Nyquist reference scan. The corrected trajectories were then used in the regridding.
To further reduce artifacts, from residual inter-segment inconsistencies, transitions were also smoothed during the regridding process, effectively averaging the signals with linearly varying weights along the overlap. This was achieved by scaling the regridding sample-weights within the overlap — found for each segment separately — by $$$f$$$ for one segment and $$$(1-f)$$$ for the other, with $$$f$$$ varying along the overlap from 0 to 1; linearly dependent on $$$k_x$$$. See figure 3.

Phantom

A 3D-printed head-shaped phantom designed for 7T was used.⁵ It included an agar “brain” with metabolites and an oil filled “lipid” layer.

Scan parameters

Scan parameters for the phantom were: TR/TE 2000/14.6 ms, FOV (RO×PE) 200×250 mm², slice thickness 20 mm, in-plane resolution 4.17×3.9 mm² (48×64 acquisition matrix), 3 readout segments, SW 2778 Hz, echo spacing 0.360 ms, 294 echoes, and scan duration 6:32 min.
For human imaging the scan parameters were: TR/TE 1600/13 ms, FOV (RO×PE) 260×300 mm², slice thickness 15 mm, in-plane resolution 4.12×4.69 mm² (63×64 acquisition matrix), 3 readout segments, SW 2778 Hz, echo spacing 0.360 ms, 638 echoes, and scan duration 5:14 min.

Results

Figure 4 shows phantom results; showing NAA, Cr, and Cho images as well as spectra at two representing points. Results are shown with and without trajectory corrections. Figure 5, shows the same for the human imaging.

Discussion & Conclusions

Readout segmented acquisition is prone to artifacts from segment inconsistencies, artifacts are exacerbated by the lipids' strong signal, if lipid suppression is off. These artifacts were reduced by using a reference scan based trajectory correction, smoothing of the signal at segment transitions, as well as — per-frequency and per-segment — phase corrections. The in vivo images still occasionally suffer from some artifacts, which are most likely due to motion and breathing. Full analysis still requires phasing of all data and model fitting of the spectra.

Acknowledgements

No acknowledgement found.