2065

High Resolution, Respiratory Motion-Resolved Coronary MRA Using a Reordered Variable-Density 3D Cones Trajectory
Srivathsan P. Koundinyan1, Frank Ong2, Mario O. Malavé1, Bob S. Hu3, and Dwight G. Nishimura1

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Electrical Engineering and Computer Science, University of California, Berkeley, Berkeley, CA, United States, 3Cardiology, Palo Alto Medical Foundation, Palo Alto, CA, United States

Synopsis

High resolution MR angiography is important to properly visualize fine lesions in coronary arteries. In this work, we employ a time-efficient and motion-robust variable-density 3D cones trajectory for sub-millimeter data acquisition. Although the imaging data is severely undersampled, we capitalize on the diffuse aliasing properties of the non-Cartesian cones trajectory to separate the data into several respiratory phases for motion compensation. For improved image quality, we present and analyze a modified cones reordering strategy, mindful of balanced SSFP imaging and unwanted eddy current effects, for distributed k-space coverage irrespective of how the readouts are retrospectively sorted.

Introduction

To achieve a reasonable scan time for sub-millimeter coronary MR angiography, we undersample the imaging data with an efficient variable-density (VD) 3D cones trajectory1,2. Accurate respiratory motion correction must be performed to fully benefit from the spatial resolution. We implement and compare two different strategies in this regard: (1) an autofocusing technique utilizing localized motion information from 3D image-based navigators (3D iNAVs)3,4,5 and (2) a parallel imaging and compressed sensing (PICS) framework to resolve imaging data along the respiratory dimension6,7. To enable the second approach, we improve the temporal flexibility of the VD cones acquisition scheme with a modified phyllotaxis readout ordering scheme. The proposed acquisition, reordering, and motion correction techniques are examined in clinical patients with suspected coronary artery disease.

Methods

Sequence and Acquisition: Free-breathing, cardiac-gated 3D CMRA data and beat-to-beat 3D iNAVs are collected on a 1.5T GE scanner with a balanced SSFP (bSSFP) sequence (Figure 1). We use a VD cones trajectory designed for an isotropic spatial resolution of 0.98 mm. Specifically, we fully sample k-space up to 102 m-1, and undersample higher frequency information. The cones reordering technique determines the specific undersampling factor.

Cones Reordering: We adapt a phyllotaxis readout ordering scheme to maximize k-space coverage within each heartbeat, while avoiding eddy current effects in bSSFP imaging. The total number of readouts must be the product of a Fibonacci number and an integer specifying the number of cones per phyllotaxis segment. Previous work utilized a Fibonacci number of 610 and 18 cones per phyllotaxis segment for an acquisition with spatial and temporal resolution of 1.2 mm and 18 cones per heartbeat, respectively8. However, such parameters for the VD 0.98 mm trajectory results in eddy current artifacts (Figure 2(a)), as the mean l2 distance between adjacent readouts in each heartbeat is 14% larger than the case of a 1.2 mm acquisition (Figure 2(b)). Therefore, to reduce the jumping in k-space in the higher resolution regime, we implement reordering with a Fibonacci number of 377 and 30 readouts per phyllotaxis segment. A temporal resolution of 18 cones per heartbeat allows us to traverse 60% of the full k-space extent within each cardiac cycle. Note in this case that the number of cones per phyllotaxis segment does not equal the number of readouts per heartbeat. Thus, once readouts in one phyllotaxis segment are completed, the cone closest (in an l2 sense) to the last readout is selected as the starting point in the next phyllotaxis segment.

The modified cones ordering strategy for the VD 0.98 mm trajectory mitigates eddy current artifacts and provides uniform k-space coverage regardless of how the data is retrospectively binned (Figure 2(c) and Figure 3). For the number of readouts (11,310 = 377*30) associated with this reordering scheme, we utilize a sampling density in k-space periphery of 0.85. A fully sampled 0.98 mm cones trajectory requires 16,661 readouts.

Motion Correction: We compare two reconstruction pipelines for sub-millimeter VD coronary scans.

Our standard approach for correction has been autofocusing, in which 32 3D translational motion estimates are extracted from different ROIs on the 3D iNAVs, and independently applied to reconstruct a bank of motion-compensated images. Minimization of a gradient entropy metric across the 32 translationally corrected images yields the final image.

Alternatively, because of our novel reordering scheme for VD cones imaging, we are able to retrospectively bin the data and achieve pseudo-random undersampling in each bin, which facilitates PICS. Note also that PICS benefits from the favorable noise-like aliasing in an undersampled cones trajectory. Leveraging these advantages, a superior-inferior translational motion estimate of the heart is derived from the 3D iNAVs, and subsequently used to group the VD data into five bins from end-inspiration to end-expiration. Intrabin 3D translational correction is then performed, and a temporal total variation (TV) regularized PICS reconstruction framework is applied to resolve the different motion states.

Results

Figure 4 and Figure 5 present the coronary angiograms following the two different motion correction approaches for clinical patients with suspected coronary artery disease. Qualitative inspection suggests the end-expiration respiratory phase from the temporal TV regularized reconstruction combined with intrabin 3D translation correction presents the sharpest vessels. Note that a gridding reconstruction of the end-expiration phase exhibits notable aliasing and blurring artifacts, which is effectively suppressed by the PICS framework.

Conclusion

Here, we have: (1) showcased a technique for sub-millimeter coronary scans with a variable-density cones trajectory, (2) proposed and analyzed a cones reordering strategy for high resolution bSSFP imaging, and (3) presented two different motion correction techniques for high resolution coronary imaging in clinical studies.

Acknowledgements

We gratefully acknowledge the support of NIH grants R01HL127039, T32HL007846, and T32EB009653. This work was also supported by the Hsi-Fong Ho Stanford Graduate Fellowship, the Ruth L. Kirschstein National Research Award, and the National Science Foundation Graduate Research Fellowship under Grant No. DGE-114747.

References

[1] Gurney et al. Magnetic Resonance in Medicine 55.3 (2006): 575-582.

[2] Addy et al. Magnetic Resonance in Medicine 74.3 (2015): 1874-1883.

[3] Addy et al. Magnetic Resonance in Medicine 77.5 (2017): 1874-1883.

[4] Cheng et al. Magnetic Resonance in Medicine 68.6 (2012): 1785-1797.

[5] Luo et al. Magnetic Resonance in Medicine 77.5 (2017): 1884-1893.

[6] Feng et al. Magnetic Resonance in Medicine 75.2 (2016): 775-788.

[7] Piccini et al. Magnetic Resonance in Medicine 77.4 (2017): 1473-1484.

[8] Malavé et al. Magnetic Resonance in Medicine 00.0 (2018): 1-12.

Figures

Figure 1: A cardiac-gated, alternating TR balanced SSFP (ATR-bSSFP) sequence with a variable-density cones trajectory designed for 0.98 mm isotropic resolution and 28x28x14 cm3 FOV is leveraged for coronary imaging. Beat-to-beat 3D iNAVs enable motion tracking. Because an image’s spectral content is concentrated near k-space origin, reducing the sample density in the periphery creates subtle aliasing artifacts. For our variable-density sampling scheme, we fully sample up to 20% of the max k-space extent and undersample thereafter by a factor of 0.85. This decreases the number of readouts required for 0.98 mm isotropic resolution by 33%.

Figure 2: (a) Reordering of the variable-density 0.98 mm cones trajectory with a Fibonacci number of 610, 18 cones per phyllotaxis segment, and temporal resolution of 18 cones per heartbeat results in eddy current artifacts (arrows). (b) The same set of parameters cause no artifacts for a 1.2 mm acquisition. But, at a resolution of 0.98 mm, the mean l2 distance between adjacent readouts in a heartbeat is 14% larger compared to a 1.2 mm trajectory. (c) An alternate set of phyllotaxis parameters that minimize k-space jumping, while retaining distributed coverage of k-space within each heartbeat, eliminates eddy current artifacts.

Figure 3: With the proposed cones readout ordering scheme, a uniform coverage of k-space is achieved when further acceleration is applied to the undersampled variable-density dataset by sorting it into five respiratory phases from end-expiration to end-inspiration. Note that a sequential acquisition strategy results in notable gaps in the k-space of the different bins. The modified phyllotaxis ordering scheme combined with the diffuse aliasing properties of the variable-density cones trajectory, despite significant undersampling associated with binning into respiratory phases, enables effective application of parallel imaging and compressed sensing reconstruction techniques.

Figure 4: The right coronary artery (RCA) and left coronary artery (LCA) (acquired with the proposed reordered variable-density cones trajectory) of a 50-year-old patient with suspected coronary artery disease. The nonrigid autofocusing scheme improves the depiction of vessels compared to no motion correction. The end-expiration phase from a temporal TV reconstruction with intrabin translational correction also offers details not present without any correction. Note that the gridding reconstruction of the end-expiration phase presents significant aliasing and blurring artifacts that are effectively mitigated by the TV reconstruction scheme. Arrows indicate regions of differences.

Figure 5: As with the other patient study, the right coronary artery (RCA) and left coronary artery (LCA) of this 75-year-old subject with suspected disease are visualized in a sharp manner with temporal TV reconstruction. In particular, for the LCA, the nonrigid autofocusing correction technique fails to depict the medial segment (arrow). A gridding reconstruction alone of the end-expiration respiratory phase also exhibits this shortcoming, but after the application of temporal TV regularized reconstruction, image quality improves significantly. Arrows indicate regions of differences.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
2065