IM-Diff: Implicit Multi-Contrast Diffusion Model for Arbitrary Scale MRI Super-Resolution.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Diffusion models show promise for Magnetic Resonance Imaging (MRI) Super-Resolution (SR).
• Existing methods struggle with integrating multi-contrast information and achieving arbitrary magnification factors.
• Clinical MRI SR requires flexible scaling beyond fixed 2x or 4x, highlighting a need for advanced techniques.

Purpose of the Study:

• To develop an implicit multi-contrast diffusion model for arbitrary-scale MRI SR.
• To enhance the exploitation of complementary information from multi-contrast MRI scans.
• To enable practical, flexible magnification for clinical MRI applications.

Main Methods:

• Introduced IM-Diff, an implicit multi-contrast diffusion model for MRI SR.
• Proposed a hierarchical multi-contrast fusion (HMF) module with reference-aware cross Mamba (RCM) for efficient information integration.
• Integrated multiple wavelet implicit neural representation (INR) magnification (WINRM) modules with wavelet activation for continuous image representation.

Main Results:

• IM-Diff demonstrated superior performance compared to state-of-the-art SR models.
• The method effectively reconstructed texture details and anatomical structures.
• Experiments confirmed the model's effectiveness across various arbitrary magnification factors on public datasets.

Conclusions:

• IM-Diff overcomes limitations of existing diffusion-based SR models by effectively utilizing multi-contrast information and INR.
• The proposed HMF and WINRM modules enable arbitrary-scale MRI SR with improved accuracy and robustness.
• This work offers a practical solution for clinical MRI SR, facilitating detailed analysis at any desired magnification.