Real-time Registration In K-space For Cardiac Cine MR Of The Fetal Heart

Motivation: Fetal cardiac MRI demands high spatial and temporal resolution but is hindered by complex motion. Conventional image-based registration approaches, commonly used for motion correction, remain slow and susceptible to artifacts. Goal(s): We aim to estimate bulk motion from real-time radial fetal cardiac MR to facilitate cardiac synchronization and enable motion-robust imaging of the fetal heart. Approach: We perform in-plane registration from highly undersampled k-space data achieving a temporal resolution of up to 45 ms using our developed LAPANet. Results: Our framework provides accurate motion estimates despite accelerated data, enhances structural delineation, and outperforms image-based registrations in real-time motion correction while being 3750× faster. Impact: Our method reliably estimates planar bulk fetal movement and periodic maternal respiration from highly undersampled radial k-spaces in real-time acquisitions. This eliminates prior reconstruction for registration purposes and dealing with residual artifacts, enabling more efficient and motion-robust fetal heart imaging.