INTRODUCTION

3D cardiac cine MRI techniques have been developed to overcome limitations of the current standard breath-hold 2D cine imaging, either during a single breath-hold or free breathing. The cardiac image quality heavily relies on the subject’s breathholding capability or reliable compensation of respiratory motion during free breathing imaging, both can vary substantially among subjects. In this study, we tested a 3D cine imaging method which only requires one minute scan time during free breathing and allows robust and reliable motion compensation by using a data-adaptive scheme to determine respiratory self-gating efficiency for individual subjects.

MATERIALS AND METHODS

Accelerated free-breathing 3D cardiac cine imaging has been developed using a pseudo-random variable-density undersampling strategy, CIRcular Cartesian UnderSampling (CIRCUS) [1]. Similar to many other self-gating methods, respiratory gating efficiency was manually chosen (such as an empirical range between 25%-50% can generate reasonable motion compensation). We aim to improve the motion compensation strategy while further reducing scan time. One minute free breathing 3D cine MRI was applied to cover the left ventricle (LV) in a short axis view on 3 healthy volunteers at a 3.0T MR scanner (GE Medical Systems, Milwaukee, WI) with an 8-channel cardiac coil. The imaging parameters were: FOV = 34.0×25.5 cm2, TR/TE = 3.5/1.4ms, flip angle = 45°, readout bandwidth = ± 125kHz, slice thickness = 8 mm, image matrix = 192×144, and number of slices = 16. Prospective ECG triggering was applied. 20 cardiac phases were reconstructed using a combined compressed sensing and parallel imaging reconstruction method (k-t SPARSE-SENSE method) [2,3]. The variable-density k-space data acquisition using CIRCUS repeatedly acquires a center k-space line along kx and the calculated cross-correlation of those center lines can provide respiratory motion tracking [1], by exploiting their principle correlation using principle component analysis (PCA). Here we applied k-means clustering on the derived gating signal to four clusters, which can represent the respiratory motion from end-expiration to end-inspiration (and the two transition phases in between), and then the data corresponding to the cluster with minimal average distance (minimal intra-cluster motion) was chosen for image reconstruction. Conventional 2D cine and single breath-hold 3D cine MRI sequences were also acquired for comparison. The 2D cine MRI was acquired with a lower in-plane resolution (2.4×2.2mm). The single breath-hold 3D cine had the same scan settings as the free breathing sequence, but with a conventional segmented view ordering instead of CIRCUS patterns. The subject-adaptive self-gating scheme has also been tested in 4D flow cardiac MRI, with scan parameters: FOV = 34.0×25.5 cm2, TR/TE = 6.3/3.1ms, flip angle = 8°, readout bandwidth = ± 125kHz, slice thickness = 8 mm, image matrix = 180×136, number of slices = 16, and scan time of ~ 5 mins.

RESULTS AND DISCUSSION

One minute free breathing 3D cine MRI has been successfully applied on all three healthy subjects (shown in Figure 1). The subject-adaptive respiratory gating efficiencies were calculated using k-means clustering (Table 1 shows the data distribution for the clusters). Overall, free-breathing 3D cardiac cine images with CIRCUS sampling strategy have inherent motion robustness due to use of spiral-like sampling trajectory and oversampling at central k-space. For respiratory pattern that contains a longer period of time at end-expiration (case #1), the self-gated free-breathing method could provide almost identical image quality to those acquired with the single breath-hold 3D MRI (first row in Figure 1 and Figure 2). For the subject who barely could hold his/her breath (case #3), both 2D and 3D breath-hold MRI suffer from severe motion artifacts, however, the free-breathing method with the subject-adaptive respiratory gating provides diagnosable image quality (third row in Figure 3). Figure 3 demonstrated the improvements in the images of self-gated 4D flow cardiac MRI using the subject-adaptive respiratory gating.

CONCLUSIONS

We have developed a one-minute free-breathing 3D cine MRI technique with subject-adaptive respiratory gating efficiency, which provides reliable image quality regardless of the respiratory motion pattern of the subject.

Acknowledgements

NIH K25 EB014914 (JL), NIH R56HL133663 (JL), GE Healthcare Research Grant (JL).