Stacked UNet-Assisted Joint Estimation for Robust 3D Motion Correction
Brian Nghiem, Zhe Wu, Lars Kasper, Kâmil Uludaǧ · Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2024
We showed that accurate 3D retrospective motion correction of T1w MPRAGE data can be achieved with a UNet-assisted joint estimation algorithm. We compared the proposed method to using the UNet on its own and the standard joint estimation algorithm. Joint estimation (with and without the UNet) outperformed using the stand-alone UNet. The UNet-assisted joint estimation algorithm converged faster than its UNet-free counterpart. We demonstrated the importance of adapting to the changing levels of artifacts over the course of the joint estimation algorithm by sequentially employing different UNets trained for correcting different levels of motion corruption.