MMR-Mamba: Multi-modal MRI reconstruction with Mamba and spatial-frequency information fusion

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Multi-modal MRI provides complementary diagnostic information but is limited by long scan times.
• Reconstructing target modality images from under-sampled data using a reference modality is a promising acceleration strategy.
• Current methods struggle to efficiently integrate multi-modal information, with convolutional models lacking long-range dependency capture and transformers having high computational costs.

Purpose of the Study:

• To develop a novel framework, MMR-Mamba, for efficient and high-quality multi-modal MRI reconstruction.
• To address the limitations of existing convolution-based and transformer-based models in integrating complementary multi-modal MRI data.
• To leverage Mamba's linear complexity for long-range dependency modeling and Fourier domain properties for enhanced reconstruction.

Main Methods:

• Proposed MMR-Mamba framework utilizing Mamba for efficient long-range dependency modeling.
• Designed Target modality-guided Cross Mamba (TCM) module for spatial domain feature integration.
• Introduced Selective Frequency Fusion (SFF) module for Fourier domain information integration and high-frequency signal recovery.
• Developed Adaptive Spatial-Frequency Fusion (ASFF) module for mutual enhancement of spatial and frequency domains.

Main Results:

• MMR-Mamba demonstrated superior performance compared to state-of-the-art methods on BraTS and fastMRI knee datasets.
• The framework effectively integrates complementary multi-modal MRI information for high-quality image reconstruction.
• Achieved efficient reconstruction by capturing long-range dependencies with linear computational complexity.

Conclusions:

• MMR-Mamba offers a significant advancement in multi-modal MRI reconstruction, addressing key challenges in data integration and computational efficiency.
• The proposed framework enhances the clinical utility of multi-modal MRI by enabling faster acquisition times without compromising image quality.
• MMR-Mamba provides a robust and efficient solution for accelerating MRI acquisition through intelligent multi-modal data fusion.