Multi-modal MRI cascaded incremental reconstruction with coarse-to-fine spatial registration

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Magnetic resonance imaging (MRI) uses multiple contrasts for tissue assessment, but long scans risk motion artifacts.
• Compressive sensing MRI (CS-MRI) accelerates imaging via computational reconstruction, often using full-sampled auxiliary images to aid under-sampled target image reconstruction.
• Cross-modal fusion of auxiliary and target MRI data is challenging due to spatial offsets and differing imaging parameters.

Purpose of the Study:

• To propose an end-to-end network for multi-modal CS-MRI that addresses cross-modal fusion challenges.
• To integrate spatial registration and cascaded incremental reconstruction for improved CS-MRI.
• To enhance the quality and reduce artifacts in accelerated MRI scans.

Main Methods:

• An end-to-end network featuring a coarse-to-fine spatial registration sub-network to align under-sampled target and full-sampled auxiliary MR images.
• A cascaded incremental reconstruction sub-network utilizing a separated criss-cross window Transformer in a dual-path architecture.
• Fusion of inter-modal and intra-modal features for iterative refinement of target MR images.

Main Results:

• The proposed network effectively mitigates spatial offsets between multi-modal MR images.
• The cascaded reconstruction successfully recovers incremental details and refines target images.
• Experimental validation on the IXI brain dataset demonstrated superior performance compared to existing CS-MRI methods.

Conclusions:

• The developed network provides an effective solution for cross-modal fusion in multi-modal CS-MRI.
• The integration of spatial registration and cascaded reconstruction significantly improves accelerated MRI.
• This approach offers a promising direction for enhancing the efficiency and quality of MRI acquisition.