Model-based multi-parameter mapping.

Area of Science:

• Medical Imaging
• Computational Biology
• Biophysics

Background:

• Quantitative MR imaging offers rich, standardized data but parameter map computation is complex.
• Current methods often neglect noise propagation, leading to inaccurate quantitative parameter maps.
• Non-linear inversion for parameters like R1, R2*, and MTsat is susceptible to noise.

Purpose of the Study:

• To develop a probabilistic generative model for quantitative MR imaging data.
• To improve the accuracy and reduce noise in quantitative parameter maps (R1, R2*, MTsat).
• To provide a probabilistic interpretation and uncertainty estimation for recovered parameter maps.

Main Methods:

• Formulation and inversion of a probabilistic generative (forward) model for the entire MR dataset.
• Utilizing second-order optimization with a novel approximate Hessian for rapid and stable model fitting.
• Incorporating a joint total variation prior to further reduce map noise.
• Implementation using PyTorch with GPU acceleration.

Main Results:

• Demonstrated improved accuracy in quantitative parameter maps derived from multi-parameter mapping data.
• Successfully reduced noise in parameter maps through the proposed probabilistic modeling and priors.
• Enabled estimation of uncertainty on recovered parameter maps due to the probabilistic formulation.

Conclusions:

• The proposed probabilistic framework offers a flexible and accurate method for quantitative MR parameter mapping.
• The approach effectively mitigates noise and provides reliable uncertainty quantification.
• The open-source implementation facilitates broader adoption and further research in quantitative MRI.