Bayesian Magnetic Resonance Joint Image Reconstruction and Uncertainty Quantification using Sparsity Prior Models and Markov Chain Monte Carlo Sampling

2026-06-15 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

A novel Bayesian framework is introduced for magnetic resonance (MR) image reconstruction and uncertainty quantification, leveraging compressed sensing. This approach formulates the problem as a linear inverse problem, assigning prior distributions to unknown model parameters, specifically assuming image sparsity in bases like spatial gradients (using a total variation prior) or wavelet transforms. A Markov chain Monte Carlo (MCMC) method, employing a split-and-augmented Gibbs sampler and proximal MCMC for non-differentiable conditional distributions, samples from the posterior. Validated on both single-coil and multi-coil datasets with various k-space sub-sampling patterns and ratios, the framework demonstrates superior image reconstruction performance over optimization-based methods. Crucially, it effectively quantifies uncertainty, showing a strong correlation between estimated uncertainty maps and error maps, surpassing existing deep learning-based methods.

Key takeaway

For Research Scientists developing advanced MR imaging techniques, consider integrating Bayesian MCMC with sparsity priors. This framework offers superior image reconstruction and robust uncertainty quantification compared to traditional optimization or deep learning methods, particularly for compressed sensing applications. You should explore its application to diverse k-space sub-sampling patterns and multi-coil datasets to enhance diagnostic confidence and improve image quality in challenging scenarios.

Key insights

Bayesian MCMC with sparsity priors enables superior MR image reconstruction and uncertainty quantification, outperforming optimization and deep learning methods.

Principles

• Bayesian formulation improves inverse problem solutions.
• Sparsity priors enhance image reconstruction quality.
• MCMC sampling quantifies parameter uncertainty.

Method

Formulates MR reconstruction as a Bayesian linear inverse problem with sparsity priors (total variation or wavelet). Employs a split-and-augmented Gibbs sampler with proximal MCMC for posterior sampling and uncertainty quantification.

In practice

• Apply Bayesian MCMC for MR image reconstruction.
• Use sparsity priors in spatial gradients or wavelets.
• Validate on single-coil and multi-coil MR data.

Topics

• Bayesian Inference
• Magnetic Resonance Imaging
• Image Reconstruction
• Uncertainty Quantification
• Compressed Sensing
• Markov Chain Monte Carlo

Best for: AI Scientist, Research Scientist

Related on AIssential

• Bridging data-driven priors via the score function for posterior sampling -- Comparative review and experimental study — LwSGS unifies diverse data-driven priors for Bayesian inverse problems by leveraging their score functions for efficient posterior sampling.
• Markov Chain Monte Carlo (MCMC) — Markov Chain Monte Carlo (MCMC) is a class of algorithms fundamental to Bayesian statistics and machine learning, designed to sample from complex probability distributions by constructing a Markov cha…
• Bayesian 3D Steerable CNNs: Enabling Equivariance and Uncertainty Quantification Simultaneously — Bayesian Steerable-CNNs unify SE(3)-equivariance with uncertainty quantification via stochastic kernels and variational inference.
• Amortized mean-shift interacting particles — Amortized mean-shift interacting particles efficiently integrate posteriors from samples, avoiding costly density or score evaluations.
• Realistic noise synthesis reduces bias and improves tissue microstructure estimation with supervised machine learning — Realistic noise synthesis in simulated data is crucial for unbiased dMRI microstructure estimation, especially at low SNR.
• Optimizing Diffusion Priors with a Single Observation — Optimize diffusion priors from a single observation by combining them into a product-of-experts and maximizing Bayesian evidence.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.