Click here to


Are you sure ?

Yes, do it No, cancel

Motion Compensated Dynamic MRI Reconstruction in Compressed Sensing Framework Using Local Affine Optical Flow Constraint

N Zhao*, D O'Connor , D Ruan , K Sheng , UCLA School of Medicine, Los Angeles, CA


(Tuesday, 7/31/2018) 1:45 PM - 3:45 PM

Room: Karl Dean Ballroom C

Purpose: Due to the presence of anatomical motion in dynamic magnetic resonance imaging (DMRI), combining the motion estimation with the compressed sensing (CS)-based DMRI reconstruction from the accelerated undersampled k-space data has attracted research interest. In this work, motion compensated CS-based DMRI reconstruction is explored. Specifically, we reconstruct the DMRI sequences and estimate motion vectors simultaneously. The estimated motion vectors are then employed to compensate the DMRI reconstruction.

Methods: The proposed framework includes two sub-problems. One is to jointly reconstruct the DMRI sequences and estimate the inter-frame motion vectors. This problem was formulated by combining the traditional CS-based image reconstruction scheme and the optical flow constraint, which was solved by a primal-dual algorithm. Moreover, the affine model for the motion vectors was employed to model complex motion patterns. The other sub-problem is to refine the DMRI reconstruction by motion compensation (MC) using the estimated motion vectors. This optimization problem was also addressed using the primal-dual algorithm. Since motion estimation using optical flow can be conducted with a multi-resolution coarse-to-fine scheme, we updated the variables including temporal image sequences and motion vectors, and image reconstruction alternately by progressively increasing the resolution for motion estimation.

Results: Simulations were conducted on various datasets with various acceleration ratios, including i) coronal lung MRI (down-sampling ratio is of 7), ii) cardiac cine MRI (down-sampling ratio is of 12). The image reconstruction performance of the proposed algorithm was superior to three state-of-the-art reconstruction algorithms including kt-SLR, L+S and MaSTER in terms of RMSE and SSIM.

Conclusion: We proposed a novel framework to jointly reconstruct the DMRI and estimate the inter-frame motion vectors. The estimated motion vectors can then be employed for MC of the reconstructed DMRI. We demonstrated that the proposed algorithm is superior to three representative state-of-the-art algorithms in terms of the image reconstruction quality.


Optimization, Motion Artifacts, Reconstruction


IM- MRI : General (Most aspects)

Contact Email