Robust Self-Gated Free-Breathing 3D Cardiac MRI Using DC Signals and Virtual Coils
Xinwei Shi1,2, Joseph Y Cheng1,2, Michael Lustig3, John M Pauly2, and Shreyas S Vasanawala1

1Radiology, Stanford University, Stanford, CA, United States, 2Electrical Engineering, Stanford University, Stanford, CA, United States, 3Electrical Engineering and Computer Science, UC Berkeley, Berkeley, CA, United States


In cardiac MRI, self-gating using the DC signal provides a promising alternative to EKG gating. However, the DC signal is often affected by other moving structures in the FOV, such as the liver, which degrades the accuracy of the extracted cardiac triggers. In this work, we demonstrate the use of virtual coils to focus the DC signal on cardiac motion and to provide a robust and generalized self-gating approach for 3D cardiac MRI. In free-breathing 4D-Flow scans of pediatric subjects, the proposed method improved the accuracy of self-gating triggers, and the self-gated images showed comparable quality with EKG gated reference.


In cardiac MRI, the EKG signal is commonly used for synchronizing data acquisition to the cardiac cycle. However, the EKG suffers from multiple sources of interferences in MRI[1], increases the complexity of scan setup, and the leads can be intimidating for children. Self-gating using the DC signal (center of k-space)[1-3] provides a promising alternative. This DC signal can be repeatedly acquired with minimal scan time penalty for most sequences.

Especially in 3D imaging, the cardiac motion extracted from DC signal is mixed with motion of other structures (such as the liver) in the FOV, which may degrade the accuracy of self-gating triggers. Band-pass filtering[4] can reduce the effect of respiratory motion, but may be unreliable during arrhythmias or irregular breathing. Here, we demonstrate the use of virtual coils (VC’s)[5] to focus the DC signal on cardiac motion and to provide a robust, simple and generalized approach for self-gated 3D cardiac MRI.


Fitting for Virtual Coils The virtual coil is a linear combination of array coil elements, which has high sensitivity only inside the chosen ROI. The weights of the combination $$$w(c)$$$ are fitted by $$$\mathrm{minimizing}_{w}\|Sw-\mathrm{diag}(m)b\|_2$$$, where $$$S(\textbf{r},c)$$$ and $$$b(\textbf{r})$$$ are the multi-coil images and the root-sum-of-squares (RSOS) image, 3D spatial locations and coils are indexed with $$$\textbf{r}$$$ and $$$c$$$. $$$m(\textbf{r})$$$ is a mask selecting voxels inside the ROI, which has smooth transitions at the edge of the ROI. Since changes in the DC signal of the VC correspond to changes of the total signal in its sensitivity region, the ROI of the VC is placed around the edge of the left ventricle (Fig.1), so that the DC signal over time reflects cardiac motion.

Extraction of Self-Gating Signals For each TR, N samples of the DC signal are collected for each coil element. The multi-coil DC data are first linearly combined using the computed weights $$$w(c)$$$ in the coil dimension. The magnitude sum of the N virtual DC samples of each TR is taken as the DC signal. A bandpass filter with a 0.5-2.5Hz passband is applied to remove the baseline drift and high-frequency noise. A template matching algorithm[3] is used to extract the cardiac triggers.

Experiments Free-breathing 4D-Flow scans were performed on 4 pediatric subjects in a 3T GE MR750 scanner using a 32-channel cardiac coil. Ferumoxytol contrast-enhancement and a 3D SPGR sequence with fat saturation were used with the parameters in Table 1. Both the DC signal and Butterfly navigators were acquired without modifying the gradient waveforms[6,7] (Fig.2). The original array coil element with the highest DC energy in the frequency range of cardiac motion was used for comparison with the VC. The same filtering procedure was applied to the DC waveforms of the VC and the selected original coil element. The cardiac triggers extracted from the VC and this original coil element were compared with the EKG triggers by evaluating the trigger variability $$$\mathrm{TV}=\sqrt{\Sigma_{n=1}^{N}{\frac{(S_n-R_n)^2}{N-1}}}$$$, where N is the number of cardiac cycles, $$$S_n$$$ and $$$R_n$$$ represent the self-gating (mean trigger delay corrected) and EKG trigger points. A soft-gated ESPIRiT reconstruction[8] with local translational motion correction[6] was used to reconstruct the end-diastolic images, based on retrospective self-gating using the VC and EKG gating.


The virtual coil had superior spatial selectivity compared to the original coil (Fig.1) and suppressed most of the unwanted influence from liver in the DC waveform (Fig.3). The self-gating triggers from the VC had lower TVs of 10.3±2.2ms (mean ± standard deviation), compared with the results of the original coil (14.2±4.7ms). The self-gating TVs of VC were similar to the results in breath-hold 2D cine scans reported in a previous study[3]. The EKG missed 1-3 trigger points in 3 studies, but there were no misses or false positives in self-gating triggers. Self-gated images had comparable quality to the EKG-gated reference (Fig.4).

The self-gating method is applicable to general cardiac imaging. It is especially useful for 4D-Flow, since the EKG gating is more likely to fail in the middle of a long scan, particularly in small children. The DC signal has a high temporal resolution of one TR for extracting accurate trigger times; the operator has more flexibility and control for tuning how the self-gating signal is extracted.


With virtual coils, the influence of respiratory motion on the DC signal is suppressed and the accuracy of self-gating triggers is improved. It permits robust self-gating in free-breathing 4D-Flow scans and general 3D cardiac scans, with minimal modification to most pulse sequences.


NIH R01-EB009690, NIH P41-EB015891, research support from GE Healthcare.


[1] AC Larson et al, MRM 2004, 51:93-102; [2] AC Brau et al, MRM 2006, 55:263-270; [3] GM Nijm et al, JMRI 2008, 28:767-772; [4] J Liu et al, MRM 2010, 63: 1230-1237; [5] X Shi et al, ISMRM 2015, p811; [6] JY Cheng et al, MRM 2012, 68: 1785-1797; [7] JY Cheng et al, ISMRM 2015, p451; [8] JY Cheng et al, JMRI 2015, 42: 407-420.


Fig. 1. The RSOS images and the profiles of the VC and the optimal selected original coil. The ROI of the VC is shown by the red box in the RSOS images. The VC sensitivity region is localized to the ROI, indicating that the DC signal of VC is less affected by the liver (pointed by arrows), compared with the original coil.

Table 1. Scan parameters and subject heart rate of the 4D-Flow studies.

Fig. 2. Sequence diagram of one TR in 4D-Flow. DC samples (orange) are acquired right after the RF excitation and before the velocity-encoding gradients. Intrinsic Butterfly navigators (green) are acquired during the velocity-encoding gradients. The Butterfly navigators of a VC localized on the whole heart provide 3D motion measurements for translational motion correction.

Fig. 3. Filtered DC waveforms of some array coils (a) and the virtual coil (b).Although the waveform of the selected original coil (green waveform in a) shows strong cardiac motion, the superimposed respiratory motion is obvious. The respiratory motion is suppressed in the results of VC, and the estimated cardiac trigger points (blue triangles) are in good agreement with the EKG trigger points (red circles).

Fig. 4. Comparison of EKG gated and VC self-gated images (cropped to enlarge the heart). The self-gated images show no additional artifacts and comparable sharpness in details, including the right coronary artery (solid arrows) and right ventricular trabeculation (dashed arrows in the bottom images).

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)