Multimodal MRI Reconstruction Assisted With Spatial Alignment Network.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Computational Imaging

Background:

• Multi-modal MRI is crucial for comprehensive tissue assessment in clinical practice.
• Accelerated MRI acquisition using k-space under-sampling requires efficient reconstruction methods.
• Existing multi-modal reconstruction techniques can be degraded by spatial misalignment between MRI contrasts.

Purpose of the Study:

• To enhance the quality of accelerated multi-modal MRI reconstruction.
• To address the negative impact of spatial misalignment on image reconstruction.
• To develop a robust method for improving MRI scan efficiency and diagnostic accuracy.

Main Methods:

• A novel spatial alignment network was developed to estimate and correct spatial displacements between MRI modalities.
• The network warps a fully-sampled reference MRI to align with an under-sampled target image.
• A cross-modality-synthesis-based registration loss was combined with reconstruction loss for joint network training.

Main Results:

• The proposed spatial alignment network significantly improves the quality of multi-modal MRI reconstruction.
• The method demonstrates superior and robust performance on clinical MRI and multi-coil k-space data.
• Accurate spatial alignment enhances the efficiency of reconstructing under-sampled MRI modalities.

Conclusions:

• Spatial alignment is critical for effective multi-modal MRI reconstruction.
• The developed spatial alignment network offers a robust solution for improving accelerated MRI.
• This technique has the potential to advance clinical MRI by enabling faster, higher-quality scans.