Introduction

Simultaneous Multi-slice (SMS) imaging has gained interest in cardiac MRI (CMR) for improving coverage with minimal SNR loss¹. However, due to unfavorable coil geometries, acceleration potential of SMS CMR remains limited compared to neuroimaging. Image quality with high acceleration rates is compromised due to leakage artifacts from extra-cardiac tissues¹. To enable higher acceleration factors in SMS CMR, outer volume suppression (OVS) techniques have been proposed to suppress unwanted signals from surrounding tissues^2,3. Even with OVS, the reconstruction method remains important for artifact-free SMS CMR. Numerous techniques have been proposed to reconstruct SMS accelerated images, providing different trade-offs between noise reduction and leakage-blocking^4-6. In this work, we sought to develop a new SMS reconstruction technique called readout-concatenated k-space (ROCK)-SPIRiT to reduce leakage artifacts and noise amplification in highly-accelerated SMS imaging. ROCK-SPIRiT was evaluated on both perfusion and cine imaging and compared to SENSE/GRAPPA⁴, split slice(ss)-GRAPPA⁵ and SMS-SPIRiT⁶. It was shown to reduce leakage artifacts compared to SMS-SPIRiT and noise amplification compared to SENSE/GRAPPA and ss-GRAPPA.

Methods

Reconstruction: ROCK-SPIRiT utilizes the idea of readout concatenation in image domain as in SENSE/GRAPPA⁴ to encode SMS acceleration. SMS-accelerated slices are inverse Fourier transformed along readout and concatenated, forming an extended k-space. SMS acceleration then corresponds to undersampling in this concatenated direction (Fig. 1). ROCK-SPIRiT solves the following objective function in the extended-kspace:
$$arg\ \underset{\bf{\kappa_{ext}}}{min} ||\bf{P_{\Omega}}\bf{\kappa_{ext}}-\bf{\kappa_{SMS}}||_2^2 + \lambda||\bf{G_{ROCK}}\bf{\kappa_{ext}}-\bf{\kappa_{ext}}||^2_2 + \sigma\bf{\Psi}(\bf{E}\bf{\kappa_{ext}})$$
where $$$\bf{P_{\Omega}}$$$ is a subsampling operator in k-space, $$$\bf{\kappa_{ext}}$$$ is the extended k-space in readout direction, $$$\bf{\kappa_{SMS}}$$$ is the acquired SMS data across all coils in k-space, $$$\bf{G_{ROCK}}$$$ is the ROCK-SPIRiT self-consistency operator calibrated on low resolution extended k-space, $$$\bf{E}$$$ is SENSE-1 operator, $$$\bf{\Psi}$$$ is a regularizer, $$$\lambda$$$ and $$$\sigma$$$ are weight terms. Fig. 1 shows a schematic of data consistency (first term) and SPIRiT⁸ self-consistency (second term). The objective function is solved using ADMM.

Images were reconstructed with SENSE/GRAPPA⁴, ss-GRAPPA⁵, SMS-SPIRiT⁶ and ROCK-SPIRiT. SENSE/GRAPPA utilizes standard GRAPPA on readout-extended k-space, ss-GRAPPA works on regular k-space providing leakage-blocking as an extension to slice-GRAPPA, and SMS-SPIRiT employs individual SPIRiT kernels on SMS-accelerated k-space. For all methods, interpolation kernels were generated from calibration data.

ROCK-SPIRiT for cine imaging was implemented with $$$\lambda = 10^{-4}$$$ and $$$\sigma=1.2 \times 10^{-7}$$$ and the regularizer was based on locally low rank constraint⁹. For perfusion imaging, three-set of SMS-accelerated slices (for a total of 9 slices) were processed together for each dynamic. Thus, no temporal information was shared across dynamics. ROCK-SPIRiT for perfusion imaging was implemented with $$$\lambda = 10^{-4}$$$ and $$$\sigma=2.5 \times 10^{-8}$$$ and regularizer was based on low dimensional-structure self-learning and thresholding¹⁰.

In Vivo Imaging: Imaging was performed at 3T (Siemens Prisma) with institutional review board approval. Written informed consent was obtained. A multiband combination of two slab-selective single band saturation pulse along phase-encoding was utilized to perform OVS² (Fig. 2), suppressing signals from chest and back simultaneously. First-pass perfusion imaging was performed on a healthy subject with injection of 0.05 mmol/kg gadabutrol at 4 mL/s followed by a 10-mL saline flush, followed by cine imaging². Perfusion parameters were: SMS-factor=3, in-plane acceleration=4, partial Fourier=6/8, FOV/resolution=360×360/1.7x1.7mm², slice-thickness=8mm, temporal resolution=110ms, saturation time=150ms. Cine parameters were: SMS-factor=5, FOV/resolution=320×320/1.7x1.7mm², slice-thickness=6mm, temporal resolution=41ms. Cine data was further retrospectively undersampled to SMS=5 × in-plane acceleration=2 with 24 ACS lines. Slices were simultaneously excited with CAIPIRINHA shifts⁷. Both sequences used the aforementioned OVS module, which was interleaved within the imaging window between every 9 imaging pulses for perfusion and 8 for cine. OVS parameters were²: saturation slab=150mm (each side), 3.8ms asymmetric sinc, BWT=8 and RF peak shift = 15%. Calibration scans were acquired prior to imaging with FOV/resolution=360×360/1.7×5.6mm² and FOV/resolution=320×320/1.7×4mm² for perfusion and cine respectively.

Results

Fig. 3a shows cine results without and with OVS at SMS=5. Non-OVS images suffer from substantial artifacts. SENSE/GRAPPA with OVS shows leakage artifacts and suffers from noise amplification. SMS-SPIRiT with OVS shows less noise amplification, but residual leakage artifacts remain visible. ss-GRAPPA eliminates leakage artifacts albeit at increased noise amplification. ROCK-SPIRiT shows improved image quality compared to other methods with less noise and no leakage artifacts.
Fig. 3b shows OVS-based cine with additional 2-fold in-plane acceleration. In this scenario, SENSE/GRAPPA has significant artifacts. SMS-SPIRiT shows more leakage artifacts with the higher acceleration rate. ss-GRAPPA eliminates the leakage artifacts but suffers from high noise amplification. ROCK-SPIRiT improves reconstruction by both eliminating leakage and reducing noise amplification, although some artifacts remain.
Fig. 4 shows OVS-based perfusion CMR for a highly-accelerated scan (SMS=3×in-plane acceleration=4×partial Fourier=6/8). Noise amplification is visible in SENSE/GRAPPA and ss-GRAPPA reconstructed images. SMS-SPIRiT shows leakage artifacts (yellow arrows). ROCK-SPIRiT reduces noise and improves image quality compared to other methods.

Discussion

In this study, we proposed a new regularized SMS reconstruction technique called ROCK-SPIRiT. This allows further denoising than SENSE/GRAPPA and ss-GRAPPA by enabling regularization. ROCK-SPIRiT also reduces leakage compared to SMS-SPIRiT.

Conclusion

ROCK-SPIRiT provides improved image quality for highly accelerated SMS CMR, reducing leakage and noise amplification compared to existing methods.

Acknowledgements

Funding: Grant support: NIH R00HL111410, NIH P41EB027061, NIH 1S10OD017974-01 and NSF CAREER CCF-1651825.