3520

PEC-GRAPPA Reconstruction of Simultaneous Multislice EPI with Slice-Dependent 2D Nyquist Ghost Correction
Zheyuan Yi1,2, Yilong Liu1,2, Mengye Lyu1,2, and Ed X. Wu1,2

1Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong, China, 2Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China

Synopsis

Nyquist ghost correction is challenging for simultaneous multislice (SMS) EPI due to the slice-dependent 2D phase error between positive and negative echoes. For this problem, phase error correction SENSE (PEC-SENSE) has been proposed recently, which incorporates slice-dependent 2D phase error maps into coil sensitivity maps. In this study, we extend the concept of PEC-SENSE to k-space based implementation termed as PEC-GRAPPA. It outperforms 1D LPC based GRAPPA reconstruction and requires less tuning than PEC-SENSE such as excluding background areas.

Introduction

Nyquist ghost correction is challenging for simultaneous multislice (SMS) EPI due to the slice-dependent phase error between positive and negative echoes. Several SMS reconstruction methods implemented slice-dependent 1D linear phase correction (LPC), but actual phase errors may vary in both phase encoding and readout directions. For this problem, phase error correction SENSE (PEC-SENSE) has been proposed recently, which incorporates slice-dependent 2D phase error maps into coil sensitivity maps (1-3). In this study, we extend the concept of PEC-SENSE to k-space based implementation termed as PEC-GRAPPA. It outperforms slice-dependent 1D LPC based GRAPPA with lesser artifacts and higher SNR. Compared with PEC-SENSE, it requires less tuning such as excluding background areas (4).

Method

Let Nc denote the number of coil channels, MB the multiband factor and R the in-plane acceleration factor in SMS acquisition.

Coil Sensitivity and Phase Error Calibration: Single-band EPI data are acquired with R interleaved shots to match the geometric distortion of SMS EPI data. As shown in Fig 1a, self-referenced ghost correction is performed on the single-band data using entropy minimization 1D LPC and virtual coil SAKE, so that ghost-free single-band images can be derived and subsequently used to train GRAPPA weights or coil sensitivity maps. The positive- and negative-echo virtual channels generated by VC-SAKE are used to calculate slice-dependent phase error maps.

PEC-GRAPPA: As illustrated in Fig 1b, ghost-free single-band images are Fourier transformed with/without multiplying the phase error maps, synthesizing the ACS lines for positive- and negative-echo data. The synthesized ACS lines are then used to train GRAPPA kernels (based on 2D readout concatenation SENSE-GRAPPA framework). Therefore, in PEC-GRAPPA, the effective acceleration factor is MB×2R, whereas the effective channel number is 2Nc.

Comparisons: (a) 1D LPC GRAPPA: 1D LPC GRAPPA is implemented as described above for PEC-GRAPPA, except that 1D LPC single-band images and 1D linear phase error maps are used to train GRAPPA weights. It can be regarded as a variation of NGC-SENSE-GRAPPA (5). (b) PEC-SENSE: ESPIRiT (6) is used to calculate coil sensitivity maps from ghost-free single-band images. Positive echoes are set to have the original coil sensitivity maps, while coil sensitivity maps for negative echoes are multiplied by phase error maps. The positive and negative echoes are jointly reconstructed to produce final images. PEC-SENSE are performed with and without masking the object areas.

Experiments:

Rat brain imaging at 7T: rat brain data were collected on a 7T Bruker scanner with a 4-channel surface coil. GE SMS EPI data were acquired with in-house programmed multiband sequence (MB=2 and R=1 without CAIPIRINHA shift). Other acquisition parameters were TR/TE=2000/16ms, flip-angle=90°, FOV=40×40 mm2, echo-spacing=0.24ms, bandwidth=4166Hz/pixel, slice-number=20, slice-thickness=1mm.

Phantom and human brain imaging at 7T: phantom and human brain data were collected on a 7T Siemens scanner with a 32-channel head coil. To clearly show 2D phase errors, the phantom was scanned with oblique slice orientation and linear shimming only. GE SMS EPI data were acquired at MB=2 and R=2. Other acquisition parameters were TR=700/2170ms (human/phantom), TE=30 ms, matrix-size=128×128, echo-spacing=0.7-0.8 ms, bandwidth=1628Hz/pixel, slice-number=20/80 (human/phantom, after reconstruction), slice-thickness=2 mm. Human brain data were also acquired at MB=4 and R=2 with 50 repetitions for t-SNR calculation.

Result

Figure 2 shows the reconstruction of rat brain data. 1D LPC GRAPPA led to obvious residual artifacts, which were mostly removed by PEC-GRAPPA, leading to similar image quality to PEC-SENSE.

Figure 3 shows that for the oblique phantom data, strong residual artifacts existed in 1D LPC GRAPPA results, while they were substantially reduced by PEC-GRAPPA as well as PEC-SENSE. SMS reconstruction of human brain data at MB=2 and R=2 is shown in Figure 4, where PEC-GRAPPA effectively reduced the residual artifacts that arise in 1D LPC GRAPPA results. Figure 5 shows the t-SNR map of human brain data at MB=4 and R=2. The t-SNR of PEC-GRAPPA was higher than that of 1D LPC GRAPPA and PEC-SENSE without mask. The overall t-SNR of PEC-GRAPPA was comparable to PEC-SENSE with mask.

Discussion and Conclusions

We have presented PEC-GRAPPA for slice-dependent 2D Nyquist ghost correction in SMS EPI. It outperforms slice-dependent 1D LPC GRAPPA with lesser artifacts. PEC-GRAPPA is conceptually similar to dual-polarity GRAPPA, yet it employs virtual coils instead of polarity dependent kernels, and does not involve temporally encoded reference scans. Compared with PEC-SENSE, PEC-GRAPPA can achieve similar results without tuning masks, and typically has smoother noise distribution. The main disadvantage of PEC-GRAPPA is increased computation time due to doubled channel numbers. Thus, it is desirable to combine it with coil compression in future studies.

Acknowledgements

This work was supported by the Hong Kong Research Grant Council (Grants C7048-16G and HKU17103015 to E.X.W.).

References

1. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med 1999;42(5):952-962.

2. Lyu M, Barth M, Xie VB, Liu Y, Feng Y, Wu EX. Robust 2D Nyquist Ghost Correction for Simultaneous Multislice (SMS) EPI Using Phase Error Correction SENSE and Virtual Coil SAKE. 2017; Honolulu, Hawaii, USA.

3. Xie VB, Lyu M, Liu Y, Feng Y, Wu EX. Robust EPI Nyquist ghost removal by incorporating phase error correction with sensitivity encoding (PEC-SENSE). Magn Reson Med 2017.

4. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med 2002;47(6):1202-1210.

5. Koopmans PJ. Two-dimensional-NGC-SENSE-GRAPPA for fast, ghosting-robust reconstruction of in-plane and slice-accelerated blipped-CAIPI echo planar imaging. Magn Reson Med 2017;77(3):998-1009.

6. Uecker M, Lai P, Murphy MJ, Virtue P, Elad M, Pauly JM, Vasanawala SS, Lustig M. ESPIRiT-an eigenvalue approach to autocalibrating parallel MRI: Where SENSE meets GRAPPA. Magn Reson Med 2013.

Figures

Fig. 1. (a) VC-SAKE based calibration to obtain ghost-free single-band images and phase error maps. To generate phase error maps, the SAKE recovered virtual channels are slice-wise processed by ESPIRiT with phase subtraction. The virtual channels are then combined to generate ghost-free single-band images for GRAPPA kernel training. (b) PEC-GRAPPA reconstruction. ACS lines for positive and negative echoes were obtained by Fourier transform of single-band images with and without multiplying the phase error maps. The GRAPPA kernel training and interpolation are performed under the framework of 2D readout concatenation SENSE-GRAPPA.

Fig. 2. Reconstruction of rat brain SMS EPI data at MB=2 and R=1. Obvious residual ghost artifacts can be found in 1D LPC GRAPPA results as indicated by the arrows. PEC-GRAPPA removed most of such artifacts, leading to better image quality that is comparable to PEC-SENSE results. Note that all methods here utilized L2-norm regularization factor of 0.001.

Fig. 3. Reconstruction of phantom SMS EPI data at MB=2 and R=2. PEC-GRAPPA effectively reduced the strong residual ghost artifacts in 1D LPC GRAPPA results as indicated by yellow arrows. All methods here utilized L2-norm regularization factor of 0.001.

Fig. 4. Reconstruction of human brain SMS EPI data at MB=2 and R=2. PEC-GRAPPA led to well reconstructed images similar to PEC-SENSE with mask, whereas 1D LPC GRAPPA and PEC-SENSE without mask had obvious artifacts. All methods here utilized L2-norm regularization factor of 0.001.

Fig. 5. Temporal SNR maps of reconstructed human brain SMS EPI images at MB=4 and R=2. PEC-GRAPPA yielded higher SNR than 1D LPC GRAPPA and PEC-SENSE without mask as indicated by black circles. The overall t-SNR of PEC-GRAPPA was comparable to PEC-SENSE with mask. All methods here utilized L2-norm regularization factor of 0.001.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
3520