From Coarse to Continuous: Progressive Refinement Implicit Neural Representation for Motion-Robust Anisotropic MRI...

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Artificial Intelligence

Background:

• Slice-to-volume reconstruction in motion-robust MRI is crucial for accurate 3D brain imaging, especially with accelerated acquisitions or patient motion.
• Challenges include local detail loss, global structural aliasing from motion, and volumetric anisotropy, hindering high-quality reconstruction.
• Existing methods struggle to address these hierarchical structural disruptions effectively.

Purpose of the Study:

• To propose a novel progressive refinement implicit neural representation (PR-INR) framework for robust 3D brain MRI reconstruction.
• To unify motion correction, structural refinement, and volumetric synthesis within a geometry-aware coordinate space.
• To improve the quality and consistency of 3D brain volumes reconstructed from undersampled and motion-corrupted 2D MRI slices.

Main Methods:

• Developed a PR-INR framework integrating a motion-aware diffusion module for initial volumetric reconstruction and artifact suppression.
• Introduced an implicit detail restoration module for residual refinement, correcting local structures and enhancing boundary precision.
• Employed a voxel continuous-aware representation module for accurate inter-slice completion and high-frequency detail recovery.

Main Results:

• PR-INR demonstrated superior performance over state-of-the-art methods on five public MRI datasets under various motion and undersampling conditions.
• Achieved significant improvements in quantitative reconstruction metrics and visual quality compared to existing techniques.
• Showcased strong generalization and robustness across diverse, unseen MRI domains and motion scenarios.

Conclusions:

• The proposed PR-INR framework effectively addresses hierarchical structural disruptions in motion-robust MRI reconstruction.
• PR-INR offers a unified approach for motion correction, detail enhancement, and volumetric synthesis, leading to high-fidelity 3D brain volumes.
• This method holds promise for improving diagnostic accuracy and efficiency in clinical MRI applications.