J-Score: Joint Distribution Learning With Score-Based Diffusion for Accelerating T1ρ Mapping.

Area of Science:

• Magnetic Resonance Imaging
• Medical Imaging Physics
• Computational Imaging

Background:

• T1rho mapping is crucial for tissue characterization but limited by long scan times.
• Undersampling accelerates T1rho imaging but requires accurate modeling of multi-contrast image correlations.
• Existing methods for joint correlation modeling in T1rho imaging are often inaccurate.

Purpose of the Study:

• To develop an accelerated T1rho imaging technique using accurate joint distribution modeling.
• To address the limitations of lengthy scan times in T1rho mapping for clinical use.
• To improve the reconstruction of undersampled multi-contrast T1rho images.

Main Methods:

• A joint diffusion model was proposed to approximate the joint distribution of multi-contrast T1rho images using score-matching.
• A joint reverse denoising diffusion model was employed to reconstruct undersampled T1rho images guided by the learned joint distribution.
• Bayesian framework principles were leveraged for accurate joint distribution representation.

Main Results:

• The proposed method demonstrated superior performance in accelerating T1rho imaging compared to existing techniques.
• Accurate characterization of the joint distribution of multi-contrast T1rho images was achieved.
• In vivo experiments and patient image reconstruction validated the method's feasibility and effectiveness.

Conclusions:

• The developed joint diffusion model effectively accelerates T1rho parameter imaging while maintaining high image quality.
• This approach offers a promising solution for overcoming the clinical limitations of T1rho mapping due to long acquisition times.
• The method's ability to accurately model joint distributions enhances the reconstruction of undersampled MRI data.