Mamba2SVN: a Mamba2 and reconstruction-cooperative sensitivity refinement-based variational network for parallel MRI...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Magnetic Resonance Imaging (MRI) offers superior soft-tissue contrast and safety but is limited by long scan durations.
• Accelerating MRI acquisition is critical for improving patient comfort and throughput in clinical settings.

Purpose of the Study:

• To introduce Mamba2SVN, a novel variational network designed to significantly reduce MRI acquisition times.
• To improve the efficiency and robustness of MRI reconstruction using deep learning techniques.

Main Methods:

• Developed Mamba2SVN, a network coupling a Mamba2-based reconstruction branch with sensitivity refinement.
• Incorporated a Mamba2-based Cross-iteration Feature Reconstruction (MCFR) module with cross-iteration feature fusion.
• Implemented a reconstruction-cooperative sensitivity refinement (RCSR) module and a dual data-consistency (DDC) mechanism.

Main Results:

• Mamba2SVN demonstrated superior performance compared to state-of-the-art deep learning methods across multi-coil knee, brain, and cardiac datasets.
• The method consistently achieved high-quality reconstructions even with various undersampling patterns.

Conclusions:

• Mamba2SVN offers a promising solution for accelerating MRI scans, particularly in scenarios with limited autocalibration signal (ACS) regions.
• The approach holds potential for enabling highly accelerated clinical MRI protocols, enhancing diagnostic efficiency.