Bayesian MRI reconstruction with joint uncertainty estimation using diffusion models.

Area of Science:

• Medical Imaging
• Computational Science
• Machine Learning

Background:

• Deep learning methods are common for MRI reconstruction.
• Conventional methods often lack uncertainty quantification.

Purpose of the Study:

• Introduce a novel framework for efficient MRI reconstruction.
• Enable sampling from learned probability distributions for improved image reconstruction.

Main Methods:

• Utilize Markov chain Monte Carlo (MCMC) sampling from posterior distributions.
• Employ score-based generative models for data-driven Markov chains.
• Framework is independent of the k-space measurement forward operator.

Main Results:

• Framework enables computation of maximum a posteriori and minimum mean square error estimates.
• Quantifies uncertainty maps for reconstructed MRI images.
• Investigated framework's performance regarding uncertainty interpretation, generative model training, MCMC efficiency, and transferability.

Conclusions:

• The framework connects diffusion processes, generative models, and Markov chains.
• Demonstrates flexibility in MRI contrasts and sampling patterns.
• Highlights benefits of pixel-wise uncertainty quantification in MRI reconstruction.