Adaptive Gate-Aware Mamba Networks for Magnetic Resonance Fingerprinting

Magnetic Resonance Fingerprinting (MRF) enables multiparametric mapping but suffers from aliasing artifacts under high undersampling. To address this, we propose GASTMamba, an end-to-end framework for accelerated MRF reconstruction. Our model integrates a dual Mamba-based encoder to capture long-range temporal dependencies in signal evolutions and a Gate-Aware Spatial-Temporal (GAST) processor to adaptively enforce spatial coherence. This design enables robust T1 and T2 mapping under 5-fold acceleration. Simulation results showed GAST-Mamba outperformed state-of-the-art methods by up to 4.5%. In vivo experiments demonstrated effective artifact mitigation and preservation of fine anatomical structures, yielding T1 and T2 values consistent with literature. Our work establishes GAST-Mamba as a robust and scalable solution for substantially accelerating MRF acquisitions without compromising quantitative accuracy. Code available at: https://github.com/dty00/GAST-mamba.