Whole‐heart coronary MR angiography using a 3D cones phyllotaxis trajectory

Self Cite

Purpose:To rapidly reconstruct undersampled 3D non-Cartesian image-based navigators (iNAVs) using an unrolled deep learning (DL) model, enabling nonrigid motion correction in coronary magnetic resonance angiography (CMRA). Methods: An end-to-end unrolled network is trained to reconstruct beat-to-beat 3D iNAVs acquired during a CMRA sequence. The unrolled model incorporates a nonuniform FFT operator in TensorFlow to perform the data-consistency operation, and the regularization term is learned by a convolutional neural network (CNN) based on the proximal gradient descent algorithm. The training set includes 6,000 3D iNAVs acquired from 7 different subjects and 11 scans using a variable-density (VD) cones trajectory. For testing, 3D iNAVs from 4 additional subjects are reconstructed using the unrolled model. To validate reconstruction accuracy, global and localized motion estimates from DL model-based 3D iNAVs are compared with those extracted from 3D iNAVs reconstructed with l 1 -ESPIRiT. Then, the high-resolution coronary MRA images motion corrected with autofocusing using the l 1 -ESPIRiT and DL modelbased 3D iNAVs are assessed for differences. Results: 3D iNAVs reconstructed using the DL model-based approach and conventional l 1 -ESPIRiT generate similar global and localized motion estimates and provide equivalent coronary image quality. Reconstruction with the unrolled network completes in a fraction of the time compared to CPU and GPU implementations of l 1 -ESPIRiT (20× and 3× speed increases, respectively). Conclusions:We have developed a deep neural network architecture to reconstruct undersampled 3D non-Cartesian VD cones iNAVs. Our approach decreases reconstruction time for 3D iNAVs, while preserving the accuracy of nonrigid motion information offered by them for correction. K E Y W O R D S3D cones trajectory, convolutional neural networks, coronary MRA, non-Cartesian ORCID Mario O. Malavéhttp://orcid.org/0000-0003-0063-564X Corey A. Baron http://orcid.org/0000-0001-7343-5580 Srivathsan P. Koundinyan http://orcid. org/0000-0002-2977-6166 Christopher M. Sandino

{cm}^{3}

Section: Methodsmentioning

confidence: 99%

“…Also, oblique reformatted maximum intensity projection (MIP) images of the RCA and LCA are shown with cross‐sectional views of the vessels before and after motion correction when using the

l_{1}

Section: Methodsmentioning

confidence: 99%

Reconstruction of undersampled 3D non‐Cartesian image‐based navigators for coronary MRA using an unrolled deep learning model

Malavé

Baron

Koundinyan

et al. 2020

Self Cite

“…For each heartbeat, 18 readouts were acquired, for a total of 10,980 readouts over 610 heartbeats. The cones were acquired in a phyllotaxis order, which improves robustness to motion . The SSFP was catalyzed by a cosine ramp of 10 pulses for a total catalyzation time of 56 ms.…”

Section: Methodsmentioning

confidence: 99%

“…The cones were acquired in a phyllotaxis order, which improves robustness to motion. 36 The SSFP was catalyzed by a cosine ramp of 10 pulses for a total catalyzation time of 56 ms. The T 2 -prepared OVS sequence was played once every R-R interval immediately before the SSFP catalyzation.…”

Section: Phantom Experimentsmentioning

confidence: 99%

Combined T₂‐preparation and multidimensional outer volume suppression for coronary artery imaging with 3D cones trajectories

Zeng

Baron

Malavé

et al. 2019

Self Cite

Purpose To develop a modular magnetization preparation sequence for combined T2‐preparation and multidimensional outer volume suppression (OVS) for coronary artery imaging. Methods A combined T2‐prepared 1D OVS sequence with fat saturation was defined to contain a 90°−60180°60 composite nonselective tip‐down pulse, two 180°Y hard pulses for refocusing, and a −90° spectral‐spatial sinc tip‐up pulse. For 2D OVS, 2 modules were concatenated, selective in X and then Y. Bloch simulations predicted robustness of the sequence to B0 and B1 inhomogeneities. The proposed sequence was compared with a T2‐prepared 2D OVS sequence proposed by Luo et al, which uses a spatially selective 2D spiral tip‐up. The 2 sequences were compared in phantom studies and in vivo coronary artery imaging studies with a 3D cones trajectory. Results Phantom results demonstrated superior OVS for the proposed sequence compared with the Luo sequence. In studies on 15 healthy volunteers, the proposed sequence had superior image edge profile acutance values compared with the Luo sequence for the right (P < .05) and left (P < .05) coronary arteries, suggesting superior vessel sharpness. The proposed sequence also had superior signal‐to‐noise ratio (P < .05) and passband‐to‐stopband ratio (P < .05). Reader scores and reader preference indicated superior coronary image quality of the proposed sequence for both the right (P < .05) and left (P < .05) coronary arteries. Conclusion The proposed sequence with concatenated 1D spatially selective tip‐ups and integrated fat saturation has superior image quality and suppression compared with the Luo sequence with 2D spatially selective tip‐up.

“…Given its ultrashort TE capability and much greater sampling efficiency than 3D radial acquisition, 3D Cones has been used to measure proton density, T 1 ,

T_{2}^{*}

, susceptibility variation, and magnetization transfer in cortical bone 3‐15 as well as the short

T_{2}^{*}

tissues of the knee including tendons, ligaments, and meniscus 16‐22 . Given its robustness to motion and flow artifacts, 3D Cones has also been considered for whole‐heart coronary MRA, 23,24 together with motion correction 25,26 and off‐resonance artifact correction, for pediatric body 27 . In addition, the efficiency of 3D Cones over standard radial acquisition has made it a candidate for both free‐breathing pediatric lung 28 and abdominal 29 imaging.…”

Section: Introductionmentioning

confidence: 99%

Three‐dimensional Yarnball k‐space acquisition for accelerated MRI

Stobbe

Beaulieu

2020

To introduce an efficient sampling technique named Yarnball, which may serve as a direct alternative to 3D Cones. Methods: Yarnball evolves through 3D k-space with increasing loop size, and the differential equations defining this flexible trajectory are presented in detail. The sampling efficiencies of Yarnball and 3D Cones were compared through point spread function analysis and simulated imaging (which highlights undersampling in the absence of other scanning effects). The feasibility of Yarnball implementation was demonstrated for fully sampled T 1-weighted images of the human head at 3 T. Results: The mostly large 3D loops of the Yarnball trajectory facilitate rapid sampling under peripheral nerve stimulation constraint, an advantage that increases with readout duration (T RO). Point spread function analysis yielded 89% (T RO = 2 ms) and 77% (T RO = 10 ms) of Yarnball voxels with magnitude less than 0.01% of the point spread function peak. For 3D Cones, these values were only 52% and 29%. The 3D-Cones technique required 1.4 times (T RO = 2 ms) and 1.8 times (T RO = 10 ms) more trajectories than Yarnball to produce simulated images of a sphere free from undersampling artifact. For a prolate spheroidal (head-like) object, 1.75 times and 2.6 times more trajectories were required for 3D Cones. Yarnball produced 0.72 mm (1/2k max) isotropic T 1-weighted human brain images free from undersampling artifact in only 98 seconds at 3 T. Conclusion: Yarnball demonstrated greater k-space sampling efficiency than directly comparable 3D Cones, and may have value wherever 3D Cones has been considered. Yarnball may also have value in the context of rapid T 1-weighted brain imaging.