Reproduce “SO-Mamba: State-Ownership Mamba for Unrolled MRI Reconstruction” 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
Pengcheng Fang, Hongli Chen, Fangfang Tang, Feng Liu, Xiaohao Cai, Shanshan Shan
SO-Mamba introduces a state-ownership principle within Mamba-based unrolled MRI reconstruction, effectively routing different reconstruction evidence types. The method achieves superior reconstruction quality across diverse MRI datasets while maintaining computational efficiency.
Key insights
- Explicitly defining "state ownership" for different types of reconstruction evidence within Mamba's recurrent computation is crucial for effective unrolled MRI reconstruction.
- Disentangling a "resident carrier" (smooth anatomical layout) from "non-resident evidence" (local details, artifacts) within the Mamba's internal paths prevents entanglement and improves information flow.
- Non-resident evidence can effectively modulate Mamba's state interfaces and contribute to output via a non-recurrent path, enhancing performance without directly entering the recurrent content.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.