Free-breathing 2D cine DENSE MRI using localized signal generation, image-based navigators, motion compensation and compressed sensing
Xiaoying Cai1, Xiao Chen2, Yang Yang1, Michael Salerno3, Daniel S. Weller4, Craig H. Meyer1, and Frederick H. Epstein1

1Biomedical Engineering, University of Virginia, Charlottesville, VA, United States, 2Medical Imaging Technologies, Siemens Healthcare, Princeton, NJ, United States, 3University of Virginia, Charlottesville, VA, United States, 4Electrical and Computer Engineering, University of Virginia, Charlottesville, VA, United States


Current cine DENSE protocols require breath-holding, which limits the use of this technique to patients with good breath-holding capabilities and excludes many pediatric and heart failure patients. To accomplish free-breathing scans with high efficiency and quality, we developed a 2D cine DENSE acquisition and reconstruction framework that utilizes localized signal generation, image-based self-navigated motion estimation, k-space motion correction and compressed sensing. Reconstructions and Bland-Altman analysis from 5 volunteers demonstrated that the proposed method recovered high-quality images and strain data from free-breathing data, showing better agreement than conventional reconstructions of the same data with breath-holding scans.


Cine displacement encoding with stimulated echoes (DENSE) is a well-established myocardial MRI strain imaging technique with high accuracy and rapid data analysis1,2. Current cine DENSE protocols require breath-holding, which limits its use in patients who cannot hold their breath such as pediatric patients and heart failure patients. Previous work has applied conventional navigator methods to enable free-breathing DENSE imaging3. However, conventional accept/reject navigator methods have poor efficiency. To accomplish free-breathing scans with high efficiency and quality, we propose a free-breathing 2D DENSE acquisition and reconstruction framework that utilizes localized signal generation4, image-based self-navigated motion estimation (iNAV), k-space motion correction and compressed sensing.


Pulse sequence: A 2D spiral cine DENSE sequence3 was modified to employ localized signal generation, which was implemented by applying orthogonal slice-selective RF pulses in the DENSE preparation so that displacement-encoded stimulated echoes occur only in the heart region where the slice profiles intersect4. In addition, a variable density spiral k-space trajectory was implemented to support the reconstruction of low-resolution intermediate navigator images representing the position of the heart during individual heartbeats.

Self-navigation and motion estimation: To estimate the motion of the heart due to respiration, low-resolution iNAV images were reconstructed for each heartbeat by an intra-heartbeat sliding-window method using the central k-space. Two-dimensional cross-correlation was used to estimate inter-heartbeat respiratory translation. The use of localized signal generation facilitated automatic estimation of heart motion due to respiration, as other tissues such as liver and chest did not generate significant signal.

Motion compensation and Compressed Sensing: Motion compensation was performed in k-space5 using $$$\hat{d}_{i}=d_{i}e^{j2\pi k_{i} t_{i}}$$$ where $$$\hat{d_i}$$$ is the motion-corrected k-space data from the i-th heartbeat, $$$d_{i}$$$ is the acquired k-space data from the i-th heartbeat, $$$k_{i}$$$ denotes the k-space trajectory of $$$d_{i}$$$, and $$$t_{i}$$$ is the 2D translation vector computed from the iNAV images. The motion corrected data were reconstructed using compressed sensing with low-rank constraints6: $$$m=argmin_{m}\parallel F_{u}m-\hat{d} \parallel_{2}+\lambda\parallel m \parallel _* $$$, where $$$m$$$ denotes the dynamic cine DENSE images to be reconstructed, $$$F_{u}$$$ denotes non-uniform fast fourier transform (NUFFT7) operator including the sampling mask and sensitivity encoding, $$$\parallel \parallel _*$$$ denotes the spatiotemporal low-rank constraint and $$$\lambda$$$ is a regularization parameter.

T1-relaxation artifact reduction: By applying localized signal generation, the DENSE stimulated echo signal originates from only the heart region. However, signal due to T1 relaxation still originates from the entire slice. Because subtraction of phase-cycled data to eliminate the artifact-generating T1-relaxation signal is ineffective for free-breathing acquisitions, we applied a circular band-stop k-space filter around the displacement-encoding frequency to reduce artifacts due to T1 relaxation.

Data acquisition: DENSE imaging was performed on 5 healthy volunteers using a 1.5T scanner (Avanto, Siemens) with a 5-channel coil. Mid-ventricular short-axis DENSE datasets were collected with both breath-holding and free-breathing. Imaging parameters were: displacement encoding frequency ke = 0.1 cyc/mm, through-plane dephasing frequency kd = 0.08 cyc/mm, temporal resolution = 28 ms, in-plane resolution = 2.5 x 2.5 mm2, slice thickness = 8 mm, localized excitation width = 60-80 mm, field of view = 160 mm, number of spiral interleaves per image = 4, and number of interleaves acquired per heartbeat = 2. Spiral interleaves were rotated by the golden angle through different cardiac phases. The total imaging time was 14 heartbeats for a full 2D dataset. Images were reconstructed offline using MATLAB (MathWorks, MA). Both breath-holding and free-breathing data were reconstructed using density weighted NUFFT. Free-breathing data were also reconstructed using the proposed method (Figure 1).

Image Analysis: Standard strain analysis2 was performed for 6 segments of the myocardium and Bland-Altman analysis was used to estimate agreement of circumferential strain for free-breathing and breath-holding datasets reconstructed using both NUFFT and the proposed method.


Example reconstructions of a free-breathing dataset are shown in Figure 2. Magnitude and phase images reconstructed with the proposed method (middle row) have image quality and segmental circumferential strain curves in better agreement with the corresponding breath-holding acquisition (bottom row) than data reconstructed with the conventional NUFFT (top row). Bland-Altman plots of strain from all 5 subjects are shown in Figure 3. Strain values using the proposed reconstruction method show better agreement with breath-holding strain values with narrower limits of agreement (-0.123, 0.095) and fewer outliers compared to conventional NUFFT (-0.210, 0.197).


The proposed method recovers high-quality cine DENSE images and strain data. By using all of the acquired data, these methods are more efficient than navigator-based accept/reject methods. The proposed method may extend the use of cine DENSE strain imaging to pediatric and heart failure patients who have difficulty with breath-holding.


NIH R01 EB 001763, R01 HL 115225


1. D. Kim, W. D. Gilson, C. M. Kramer, and F. H. Epstein, “Myocardial Tissue Tracking with Two-dimensional Cine Displacement-encoded MR Imaging: Development and Initial Evaluation 1,” Radiology, vol. 230, no. 3, pp. 862–871, 2004.

2. B. S. Spottiswoode, X. Zhong, a T. Hess, C. M. Kramer, E. M. Meintjes, B. M. Mayosi, and F. H. Epstein, “Tracking Myocardial Motion From Cine DENSE Images Using Spatiotemporal Phase Unwrapping and Temporal Fitting,” IEEE Trans. Med. Imaging, vol. 26, no. 1, pp. 15–30, Jan. 2007.

3. X. Zhong, B. S. Spottiswoode, C. H. Meyer, C. M. Kramer, and F. H. Epstein, “Imaging three-dimensional myocardial mechanics using navigator-gated volumetric spiral cine DENSE MRI.,” Magn. Reson. Med., vol. 64, no. 4, pp. 1089–97, Oct. 2010.

4. L. Pan, M. Stuber, D. L. Kraitchman, N. F. Osman, D. L. Fritzges, W. D. Gilson, and N. F. Osman, “Real-time imaging of regional myocardial function using fast-SENC.,” Magn. Reson. Med., vol. 55, no. 2, pp. 386–95, Feb. 2006.

5. G. Shechter, E. R. Mcveigh, J. A. Derbyshire, L. F. Gutiérrez, and P. Kellman, “MR Motion Correction of 3D Affine Deformations,” Proc. Int. Soc. Mag. Reson. Med, vol. 11, no. C, p. 1054, 2003.

6. S. G. Lingala, Y. Hu, E. DiBella, and M. Jacob, “Accelerated dynamic MRI exploiting sparsity and low-rank structure: kt SLR,” Med. Imaging, IEEE Trans., vol. 30, no. 5, pp. 1042–1054, 2011.

7. J. Fessler, B. P. Sutton, and others, “Nonuniform fast Fourier transforms using min-max interpolation,” Signal Process. IEEE Trans., vol. 51, no. 2, pp. 560–574, 2003.


Figure 1. Flowchart of the proposed reconstruction method for free-breathing 2D cine DENSE imaging with localized signal generation, iNAV motion estimation, motion compensation, and iterative constrained reconstruction.

Figure 2. Images reconstructed from free-breathing (FB) acquisition using NUFFT (top), and the proposed motion-compensated method (middle). Images reconstructed from breath-holding (BH) acquisition using NUFFT (bottom). Magnitude and phase images for y-displacement encoding from systolic phase (A-F) and diastolic phase (G-L) are shown. (M-O) are corresponding segmental strain curves.

Figure 3. Bland-Altman plots of segmental strain demonstrate improved performance of the self-navigated motion compensation and compressed sensing method (MC+CS) compared to NUFFT for reconstruction of free-breathing (FB) cine DENSE. Using data from 5 healthy subjects, the proposed method shows better agreement with breath-holding data than NUFFT.

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