Product-of-Gaussian-mixture diffusion models for joint nonlinear MRI reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• Diffusion models show promise for high-quality magnetic resonance image (MRI) reconstruction.
• Existing methods often use large, opaque networks and require separate coil sensitivity estimation, limiting interpretability and flexibility.
• These limitations hinder real-world application and adaptability to diverse acquisition setups.

Purpose of the Study:

• To develop a more interpretable and flexible MRI reconstruction method.
• To address the limitations of existing diffusion model-based approaches.
• To improve the robustness and efficiency of MRI reconstruction.

Main Methods:

• Jointly reconstructing images and coil sensitivities using a parameter-efficient product-of-Gaussian-mixture diffusion model as an image prior.
• Incorporating a classical smoothness prior for coil sensitivities.
• Proposing a novel, more expressive parameterization for the image prior.

Main Results:

• The proposed method achieves fast and robust MRI reconstruction.
• It demonstrates resilience to shifts in contrast, anatomical distribution, and varying k-space trajectories.
• The enhanced image prior parameterization improves denoising and reconstruction performance.

Conclusions:

• The developed method offers a significant advancement in MRI reconstruction by enhancing interpretability and flexibility.
• It provides a robust and efficient alternative to existing techniques.
• The findings pave the way for broader adoption of diffusion models in medical imaging.