Bayesian non-linear regression with spatial priors for noise reduction and error estimation in quantitative MRI with...

Area of Science:

• Medical Imaging
• Computational Biology
• Statistical Modeling

Background:

• Quantitative Magnetic Resonance (MR) imaging provides valuable physiological information but is susceptible to noise and uncertainty.
• Accurate estimation of MR parameters like T1 is crucial for reliable diagnostics and research.
• Current methods often require manual hyperparameter tuning, introducing subjectivity and limiting reproducibility.

Purpose of the Study:

• To develop an automated method for reducing and estimating uncertainty in quantitative MR parameter maps.
• To eliminate the need for manual hyperparameter tuning in MR parameter estimation.
• To improve the precision and reliability of T1 parameter maps.

Main Methods:

• A Bayesian hierarchical non-linear regression model incorporating spatial correlations between voxels.
• Markov chain Monte Carlo (MCMC) sampling to explore the high-dimensional posterior distribution with a spatial prior.
• Automatic hyperparameter search using an information criterion for unbiased model determination.

Main Results:

• Noise-reduced T1 parameter maps with associated error estimates were generated.
• The proposed method demonstrated a decrease in estimation error compared to conventional voxel-wise maximum likelihood estimation.
• Increased bias at tissue interfaces and longer computational time were observed as trade-offs.

Conclusions:

• The developed method offers more precise MR parameter estimates than conventional approaches.
• The method is free from user subjectivity, enhancing objectivity and reproducibility.
• It provides valuable uncertainty estimation alongside parameter maps.