Abstract
In recent years, compressed sensing has developed as a promising solution for accelerating MRI data acquisition. Nevertheless, balancing between reconstructing high-quality images from undersampled MRI data and saving scan time is still a challenge due to violation of the Nyquist-Shannon sampling principle. Further, it is a highly ill-posed problem. To fully explore the deformation insight between temporal frames and further enhance spatial-temporal resolution during cardiac Cine MR reconstruction, in this paper, we propose a joint optimisation mechanism for dynamic CMR reconstruction with motion compensation. In particular, by establishing the integration of the motion estimation from undersampled k-space data with motion-guided compressed sensing problem, the spatio-temporal redundancy can be leveraged in the unrolled reconstruction process. The proposed algorithm achieved the best reconstruction performance across both structural and perceptual similarity in terms of multiple aggressive acceleration rates (over 0.900 for SSIM, less than 0.04 for LPIPS), and most accurate registration output across all the cardiac regions on UKBiobank dataset, which demonstrate that the proposed method outperforms comparable methods in terms of the quality to recovery more reliable and anatomical regions. The strain analysis as downstream task by leveraging the estimated motion fields from undersampled k-space also indicated the significant clinical relevance from multi-strain curves.