Methods for Uncertainty Quantification in Dictionary Matching to Advance Reliability of Quantitative MRI.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging
• Biophysics

Background:

• Quantitative MRI (qMRI) relies on dictionary matching, a technique often lacking uncertainty quantification (UQ).
• Advanced MRI reconstructions introduce complex noise, violating standard assumptions and necessitating robust UQ methods.
• Accurate UQ is crucial for reliable parameter mapping in clinical applications.

Purpose of the Study:

• To introduce and validate two novel voxel-wise UQ methods for dictionary-matched qMRI.
• To address the challenge of non-independent and identically distributed (non-iid) noise in modern qMRI reconstructions.
• To provide statistically interpretable UQ for assessing the reliability of qMRI parameter maps.

Main Methods:

• Developed a frequentist Likelihood Ratio Test (LRT) and a Bayesian marginal posterior approach for UQ.
• Modeled noise as spatially varying and temporally correlated using covariance estimated from background regions.
• Validated methods using simulations, phantom experiments (T2 and T1 mapping), and in vivo studies with varying acceleration factors.

Main Results:

• Both LRT and Bayesian methods achieved nominal coverage rates in simulations where standard assumptions failed.
• Phantom experiments demonstrated excellent agreement with gold-standard spin-echo references.
• In vivo studies showed increased uncertainty intervals for T1 and T2 mapping with higher acceleration factors; LRT was computationally efficient.

Conclusions:

• Presented a robust framework for UQ in dictionary-matched qMRI.
• The proposed methods effectively model non-iid noise from advanced reconstructions.
• The UQ framework enhances the statistical interpretability and clinical reliability of qMRI parameter maps.