Enhance Cardiac MRI Reconstruction via Unfolding Network with Hybrid Mamba-LSTM

Cardiac Magnetic Resonance Imaging is an important clinical imaging modality for cardiovascular disease diagnosis. However, long scan times and high costs make cardiac imaging inaccessible, especially as it often requires both high spatial and temporal resolution in a limited time. While existing methods have achieved promising performance, they still lack the ability to capture long spatiotemporal relationships. In this paper, we introduce a novel dynamic reconstruction method to accelerate cardiac MRI reconstruction. Our method adopts a deep unrolling structure and introduces a novel spatiotemporal learning model for high fidelity reconstruction. Specifically, inspired by the recently proposed selective structure state space models, which have shown promising performance due to their efficiency and long-range modelling capability, we propose a hybrid Mamba-LSTM (MLSTM) module as the basic cell to fully exploit the spatiotemporal complementary information hidden in scan sequences. Extensive experiments show that the proposed method can reconstruct high quality images from the highly undersampled data and achieves superior performance compared to state-of-the-arts.