Reproduce “Prospective Dynamic 3D MRI Reconstruction via Latent-Space Motion Tracking from Single Measurement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lixuan Chen, Zhongnan Liu, Jesse Hamilton, James M. Balter, Jeong Joon Park, Liyue Shen
Researchers at the University of Michigan developed PDMR, a framework that leverages retrospectively learned non-linear latent motion manifolds for real-time, high-fidelity 3D MRI reconstruction from single, ultra-sparse measurements. The system consistently achieved superior quantitative results, including gains of approximately 2 dB PSNR over existing prospective methods, and demonstrated robust adaptation to unseen motion patterns.
Key insights
- Decomposing dynamic MRI into a static template and time-varying DVFs, then representing these DVFs via a non-linear latent manifold, allows for efficient encoding and rapid adaptation.
- A geometry-aware mapping network based on tri-plane representations efficiently translates low-dimensional latent vectors into high-resolution, structurally coherent 3D deformation fields.
- Learning a compact and generalizable latent space offline enables rapid online adaptation for prospective reconstruction by optimizing only a few parameters, meeting stringent real-time latency requirements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.