Introduction

Cardiac cine MRI is an important clinical tool for assessing cardiac diseases¹. The commonly used bSSFP sequence suffers from artifacts and excessive specific absorption rate (SAR) at high field². Therefore, alternative approaches such as spoiled gradient-echo (GRE) sequences are preferred. Auto-calibrated multiband (MB) imaging has been proposed as a general technique to improve scan efficiency³. Previous work has combined GRE cine and auto-calibrated MB imaging at 5T⁴. However, the acquisition and reconstruction were conducted in a prospective manner, which is unable to capture a complete cardiac cycle.
In this work, we expanded the previous work to retrospective GRE cine and improved the reconstruction method. Explicit k-space interpolation was avoided because the same phase-encoding (PE) lines in adjacent phases may have different inter-slice phase modulation. Instead, implicit phase interpolation was incorporated in the compressed sensing (CS) reconstruction; Moreover, periodic boundary condition (PBC) was used for the temporal total variation (TV) operator to reduce artifacts in the starting/ending frames and to ensure smooth cardiac motion. The proposed technique was demonstrated on a 5T scanner on volunteers.

Theory

Auto-calibration: The autocalibrated MB sequence is demonstrated in Fig.1. Phase modulation is achieved by adjusting the relative phase of RF pulses for different slices. The phase modulation not only produces controlled aliasing for better preserving SNR⁵, but also allows self-calibration, i.e., extracting single-slice reference images from the MB data by appropriate linear combination of the phases.
Implicit phase interpolation: Retrospective cine covers the full cardiac cycle and is preferred over prospective cine for clinical diagnosis. Conventional retrospective cine reconstruction has the following steps: 1), Each heartbeat is divided into a desired number of reconstruction phases. Raw data are subsequently rebinned according to the VSM timestamp; 2), Missing lines are generated by linear interpolation of data acquired in previous and next phases at the same PE. 3), Standard image reconstruction follows. In the proposed application, raw acquisitions are still rebinned into reconstruction phases. However, PE lines with the same k-space position and inter-slice modulation are usually more than 1 phase apart. Therefore, we choose not to fill the missing lines using explicit interpolation. Instead, we rely on sparsity constraints in the CS reconstruction for dealiasing.
Periodic boundary condition: Typical CS reconstruction makes use of sparsity in the spatial and temporal TV transform domain. Intuitively, temporal TV promotes sparsity of the difference between each phase and the ones before and after it. However, for the first/last phase, temporal TV is only calculated with the phase after/before it (the reflecting Neumann boundary condition, or RNBC), leading to stronger artifacts and lower SNR compared with middle phases. Because cardiac motion is periodic, we propose using PBC for temporal TV, i.e., temporal TV of the first phase is calculated with the second and the last phase; temporal TV of the last phase is calculated with the second-to-last and the first phase. Such formulation also ensures smooth motion between the last and first phases.
In summary, Cine images are reconstructed by minimizing the following cost function:
$$\underset{x}{\arg\min}\frac{1}{2}\parallel p_1DF(s_1x_1) + p_2DF(s_2x_2)\parallel_2+\lambda_1\parallel T_sx_1\parallel_1+\lambda_1\parallel T_sx_2\parallel_1+\lambda_2\parallel T_{tPBC}x_1\parallel_1+\lambda_2\parallel T_{tPBC}x_2\parallel_1$$
in which x₁ and x₂ are cine image series of the two slices, s₁ and s₂ are the coil sensitivity maps, p₁ and p₂ are the phase modulations, y is the rebinned multi-slice k-space data, D is the k-space sampling operator, F is the Fourier transform operator, T_s is the spatial TV operator, and T_tPBC represents the temporal TV operator with PBC. λ₁ and λ₂ are adjustable parameters for the regularization strength.

Methods

The proposed auto-calibrated multiband GRE cine was tested on a healthy volunteer on a 5 T scanner (United Imaging Healthcare, Shanghai, China) with an 8-channel RF transmit system. Forty-eight receive channels were used (18 channels from a body coil and 24 channels from a spine coil).. Imaging parameters include: FOV = 380 × 340 mm², matrix = 192 × 137, thickness = 6 mm, 2 slices were imaged with gap = 24 mm, bandwidth = 450 Hz/pixel, FA = 12°, TR = 3.9 ms, TE = 1.7 ms, views per frame=13, in-plane undersampling factor = 2, number of heart beats = 6, reconstruction phase = 20. Images of two slices were collected using MB = 2 in 1 breathhold and were reconstructed with either PBC or RNBC for comparison.

Results

Fig. 3 zoomed in on the heart and shows the first 6 phases reconstructed with PBC and RNBC. It can be observed that PBC reduces image artifacts, especially for the first few phases. Similar effects are also observed for the last few phases.
Fig. 4 shows images from the systole to the diastole phase. Heart structures are well delineated, and the motion is well preserved. The blood pool – myocardium contrast is decent, and no obvious artifacts are observed in the heart region.

Conclusion

We have implemented autocalibrated MB retrospective GRE cine at 5 T. The autocalibrated MB acquisition reduces the scan time and simplifies the MB imaging workflow. A CS reconstruction with implicit phase interpolation and temporal TV PBC further improves image quality. This application was evaluated on volunteers and achieved high-quality cine images, providing a useful alternative to bSSFP cine at high field.

Acknowledgements

No acknowledgement found.