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

To develop a robust and efficient multidimensional MRI (MD-MRI) data processing framework for accurately estimating joint frequency-dependent diffusion-relaxation distributions, while overcoming computational limitations and noise instability inherent to Monte Carlo (MC) inversion.

We introduced an Informed Dictionary-guided Monte Carlo (ID-MC) strategy that incorporates data-driven dictionary matching into the inversion process, followed by targeted local mutation refinement to enhance flexibility and reduce overfitting. This hybrid approach aims to improve the stability, accuracy, and reproducibility of MD-MRI parameter estimation. We evaluated ID-MC through in silico simulations across a range of signal-to-noise ratios and in vivo test-retest experiments in the human brain. Reproducibility was assessed using intraclass correlation coefficients (ICC) and within-subject variability, allowing rigorous comparison with MC.

In simulations, the ID-MC approach consistently achieved lower fitting errors and higher estimation accuracy across a wide range of noise levels, demonstrating its ability to balance local flexibility and global biological plausibility. Compared to MC inversion, ID-MC also reduced computation time by approximately 69%, highlighting its potential for time-efficient large-scale applications. In in vivo test-retest analyses, ID-MC substantially improved reproducibility, doubling the number of MD-MRI parameters with ICC greater than 0.75 relative to MC. Notably, diffusion frequency-dependent parameters, previously poorly reproducible with MC, showed up to 146% higher ICC with ID-MC.

By integrating data-driven dictionary matching with targeted mutation refinement, ID-MC improves the robustness, reproducibility, and computational efficiency of MD-MRI inversion, supporting studies that require highly sensitive detection of subtle brain microstructural changes.