High-resolution in vivo MR-STAT using a matrix-free and parallelized reconstruction algorithm.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Imaging Physics
• Computational Imaging

Background:

• Quantitative parameter mapping in MRI is crucial for accurate tissue characterization.
• Existing methods often require long scan times or complex post-processing.
• The MR-STAT framework enables simultaneous spatial localization and parameter estimation without Fast Fourier Transform (FFT).

Purpose of the Study:

• To develop and validate a high-resolution reconstruction algorithm for the MR-STAT framework.
• To address the computational challenges of large-scale nonlinear optimization in MR-STAT.
• To demonstrate the performance of the proposed algorithm in diverse imaging scenarios.

Main Methods:

• A matrix-free and parallelized inexact Gauss-Newton algorithm was developed for MR-STAT reconstruction.
• The algorithm was implemented on a high-performance computing cluster.
• Reconstruction performance was evaluated using simulations, phantom experiments, and in vivo human brain imaging with short pulse sequences and Cartesian sampling.

Main Results:

• The proposed algorithm successfully generated high-resolution (1mm x 1mm in-plane) quantitative parameter maps.
• Reconstructed T1 and T2 values from gel phantoms showed agreement with gold standard measurements.
• In vivo quantitative values from brain imaging correlated well with existing literature data.
• The MR-STAT reconstruction method proved effective where MR fingerprinting reconstructions failed.

Conclusions:

• The developed matrix-free, parallelized Gauss-Newton algorithm enables efficient, high-resolution quantitative parameter mapping using the MR-STAT framework.
• This approach provides accurate quantitative values and overcomes limitations of existing methods like MR fingerprinting.
• The method shows significant potential for accelerated and more informative MRI.