FEFA: Frequency Enhanced Multi-Modal MRI Reconstruction With Deep Feature Alignment.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computer Vision

Background:

• Accurate diagnostic decisions in medical imaging often require integrating information from multiple MRI modalities.
• Varying acquisition speeds across MRI modalities can lead to time-consuming scans and increased patient burden.
• Spatial misalignment between different MRI modalities can compromise the quality of reference-based reconstruction.

Purpose of the Study:

• To develop an accelerated MRI reconstruction method that addresses spatial misalignment between different modalities.
• To improve the accuracy and efficiency of reconstructing under-sampled MRI data using information from faster modalities.
• To present a novel, end-to-end trainable approach for reference-based MRI reconstruction.

Main Methods:

• Propose FEFA, a method utilizing cascading FEFA blocks for MRI reconstruction.
• Each FEFA block aligns and fuses multi-modal MRI data at the feature level.
• Features are filtered in the frequency domain to enhance relevant information and suppress noise, ensuring accurate reconstruction.

Main Results:

• FEFA demonstrates effective MRI reconstruction across various under-sampling patterns and ratios.
• The method shows superior performance compared to existing registration-then-reconstruction and cross-attention-based approaches.
• Cascading FEFA blocks stabilizes the training process and enhances reconstruction quality.

Conclusions:

• FEFA offers an end-to-end trainable solution for accelerated MRI reconstruction without additional supervision or heavy computation.
• The proposed method effectively overcomes the challenge of spatial misalignment in multi-modal MRI.
• FEFA significantly improves the accuracy and efficiency of MRI reconstruction, with potential benefits for clinical diagnostics.