A Nonrigid-Motion-Correction Method for Coronary Magnetic Resonance Angiography Using 3D Image-based Navigators
Jieying Luo1, Nii Okai Addy1, R. Reeve Ingle1, Corey A. Baron1, Joseph Y. Cheng1, Bob S. Hu1,2, and Dwight G. Nishimura1

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Palo Alto Medical Foundation, Palo Alto, CA, United States

Synopsis

3D image-based navigators (iNAVs) offer the potential to achieve more complete motion correction for coronary magnetic resonance angiography (CMRA). In this work, we develop a method for 3D-iNAV processing to achieve nonrigid motion correction. Both global and localized motion trajectories are extracted from the 3D iNAVs and used to generate candidate motion-corrected images for an autofocus method. Two sets of localized motion trajectories are obtained from deformation fields between 3D iNAVs and reconstructed binned images respectively. Results with this method on whole-heart 3D cones CMRA scans demonstrated improved vessel sharpness as compared to 3D translational motion correction.

Introduction

We have implemented a whole-heart 3D cones coronary magnetic resonance angiography (CMRA) method that acquires a 3D image-based navigator (iNAV) every heartbeat to monitor respiratory motion in every region of the heart (Fig. 1)1. In this work, we develop a nonrigid-motion-correction method that derives both global and localized motion trajectories from these 3D iNAVs.

Methods

The proposed motion-correction method uses an autofocusing method that assembles the motion-corrected image on a pixel-by-pixel basis by selecting the best-focused pixel from a bank of candidate motion-corrected 3D images2,3. Localized gradient entropy serves as the focusing metric4. 3D iNAVs are processed to extract both global and localized motion trajectories that generate the following three categories of candidate images.

1) Rigid-body-corrected image. A global rigid-body motion trajectory (translation and rotation) is estimated from the 3D iNAVs and used to correct the main scan data.

2) Localized-motion-corrected images based on 3D iNAVs. The 3D iNAVs consist of low spatial resolution (4.4 mm) frames acquired with high temporal resolution (per heartbeat). To estimate residual motion trajectories in localized regions of the heart, the 3D iNAVs are first rigid-body corrected and then processed by i) estimating nonrigid motion deformation fields between frames; ii) clustering pixels into regions with similar deformation-field behavior via k-means clustering (Fig. 2); and iii) averaging the deformation fields within each cluster to derive a 3D translational motion trajectory for that cluster. The number of clusters was empirically set to 32, leading to 32 localized motion trajectories. Correction based on each of these 32 trajectories is applied to the main scan data to augment the rigid-body correction, resulting in 32 candidate images.

3) Localized-motion-corrected images based on binned main data. To derive localized motion trajectories from images with higher spatial resolution than the 3D iNAVs, the main scan data (rigid-body-corrected) is divided into three bins, each representing a different respiratory phase. A similarity matrix analysis of the 3D iNAVs guides the binning of the main scan data. The resultant binned images have low temporal resolution (three frames) but high spatial resolution (1.2 mm), with relatively mild artifacts from k-space undersampling (Fig. 3). These frames are subjected to the same deformation-field clustering procedure described above to obtain localized translational motion trajectories that are each applied to the main scan data to augment the rigid-body correction. In this case, the number of clusters was set to 16, leading to 16 additional candidate images with localized motion correction.

To assess the method's performance, six normal subjects were scanned on a 1.5T GE Signa system using an 8-channel cardiac receiver array. The 3D cones CMRA scan was acquired with ATR-SSFP contrast with 1.2 mm isotropic resolution over a 28x28x14 cm3 FOV. Scan time was 509 heartbeats with 100% acquisition efficiency. A 3D iNAV with 4.4 mm isotropic resolution over the same 28x28x14 cm3 FOV was acquired every heartbeat in 176 ms using a 32-interleave variable-density cones scan reconstructed with ESPIRiT1.

Quantitative vessel sharpness measurements using image edge profile acutance (IEPA)5 was compared between CMRA images reconstructed with no motion correction, basic 3D translational motion correction, and the proposed method. Five segments from each of three vessels--right coronary artery (RCA), left anterior descending (LAD) artery, left circumflex (LCx)--were analyzed, resulting in IEPA scores for a total of 90 vessel segments.

Results

Figure 4 shows an example of images reconstructed with no correction, 3D translational motion correction, and the proposed method. The proposed method exhibits improved vessel sharpness in this subject. A summary of the vessel-sharpness measurements with IEPA is given in Fig. 5. Higher IEPA scores indicate increased sharpness. Compared to no correction, 3D translational motion correction and the proposed method increased the vessel sharpness by 13% and 19%, respectively. Out of the 90 vessel segments that were examined, 75 segments showed improved IEPA scores with the proposed method as compared to 3D translational motion correction.

Discussion/Conclusion

The proposed method extracts global and localized motion information from 3D iNAVs to perform nonrigid motion correction for CMRA. While global rigid-body correction by itself substantially improves vessel sharpnesss, additional correction based on localized motion trajectories leads to further improvement. 3D iNAVs acquired every heartbeat offer the flexibility to do binning and direct beat-to-beat localized motion estimation. The high temporal resolution of 3D iNAVs captures beat-to-beat nonrigid motion, while the binned data complements the iNAV information with high spatial resolution.

Acknowledgements

This work was supported by NIH R01 HL127039, NIH T32 HL007846, and GE Healthcare.

References

1) Addy N, Luo J, Ingle R, Hu B, Nishimura D. Accelerated isotropic resolution 3D image- based navigators for coronary MR angiography. Journal of Cardiovascular Magnetic Resonance 2014; 16 (Suppl 1):380.

2) Cheng JY, Alley MT, Cunningham CH, Vasanawala SS, Pauly JM, Lustig M. Nonrigid motion correction in 3D using autofocusing with localized linear translations. Magnetic Resonance in Medicine 2012; 68:1785–1797.

3) Ingle RR, Wu HH, Addy NO, Cheng JY, Yang PC, Hu BS, Nishimura DG. Nonrigid autofocus motion correction for coronary MR angiography with a 3D cones trajectory. Magnetic Resonance in Medicine 2014; 72:347–361.

4) McGee KP, Manduca A, Felmlee JP, Riederer SJ, Ehman RL. Image metric-based correction (autocorrection) of motion effects: Analysis of image metrics. Journal of Magnetic Resonance Imaging 2000; 11:174–181.

5) Olabarriaga S, Rangayyan R. Subjective and objective evaluation of image sharpness - behavior of the region-based image edge profile acutance measure. In: Proc SPIE, , 1996. pp. 154–162.

Figures

(a) Coronary MRA Timing Diagram. FS = fat saturation, 3D Cones ATR-SSFP = main scan acquisition, 3D iNAV = whole-heart 3D navigator image. (b) 3D iNAV. Cross-sectional (coronal, sagittal, axial) images through the whole-heart volumetric data.

(Top Row) Example of clustered regions in the heart with similar localized motion based on deformation fields. The deformation fields within each cluster are averaged to yield localized motion trajectories. (Bottom Row) Corresponding sections through the 3D iNAV.

Example of coronal sections through each of the three reconstructed binned images. Differences in position (e.g., the LAD (arrow) and superior position of the heart (arrowhead)) are indicative of residual motion present, even after rigid-body correction.

Example of thin-slab maximum-intensity-projection reconstructions of the RCA and LAD. (Left) Without motion correction, (Middle Left) 3D translational motion correction, (Middle Right) Proposed method, (Right) Selected motion map, with different gray levels indicating the selection of different motion trajectories.

Quantitative vessel-sharpness analysis of the reconstructed images. Average IEPA scores of the RCA, LAD and LCx in a total of 90 segments from six subjects. In paired t-tests, the IEPA scores are significantly different between the proposed method and 3D translational motion correction (P < 0.001).



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1842