Informed Dictionary-Guided Monte Carlo Inversion for Robust and Reproducible Multidimensional MRI.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Biophysics

Background:

• Multidimensional MRI (MD-MRI) enables detailed characterization of tissue microstructure.
• Traditional Monte Carlo (MC) inversion methods for MD-MRI data are computationally intensive and prone to noise instability.
• Accurate estimation of frequency-dependent diffusion-relaxation distributions is crucial for understanding brain tissue properties.

Purpose of the Study:

• To develop a robust and efficient processing framework for MD-MRI data.
• To accurately estimate joint frequency-dependent diffusion-relaxation distributions.
• To overcome computational limitations and noise instability in MC inversion.

Main Methods:

• Introduced an Informed Dictionary-guided Monte Carlo (ID-MC) strategy.
• Incorporated data-driven dictionary matching and targeted local mutation refinement.
• Evaluated ID-MC using in silico simulations and in vivo human brain test-retest experiments.
• Assessed reproducibility using intraclass correlation coefficients (ICC) and within-subject variability.

Main Results:

• ID-MC achieved lower fitting errors and higher estimation accuracy across various noise levels in simulations.
• ID-MC reduced computation time by approximately 69% compared to standard MC inversion.
• ID-MC substantially improved reproducibility in vivo, doubling parameters with ICC > 0.75 and significantly increasing ICC for diffusion frequency-dependent parameters.

Conclusions:

• ID-MC offers a robust, reproducible, and computationally efficient approach for MD-MRI data processing.
• The method enhances the detection of subtle brain microstructural changes.
• ID-MC supports time-efficient, large-scale applications in neuroimaging research.