Simulation-based inference at the theoretical limit for fast, robust microstructural MRI with minimal diffusion data.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biophysics

Background:

• Diffusion-weighted magnetic resonance imaging (dMRI) non-invasively probes brain microstructure.
• Current dMRI methods require long scan times due to extensive data sampling.
• Limited scan times hinder clinical feasibility and accessibility of dMRI.

Purpose of the Study:

• To evaluate simulation-based inference for reducing dMRI data requirements.
• To assess if reduced data preserves estimation fidelity across diffusion models.
• To enhance the speed and accessibility of brain microstructure imaging.

Main Methods:

• Applied simulation-based inference with neural posterior estimation.
• Tested on diffusion tensor imaging, diffusion kurtosis imaging, and biophysical models.
• Trained models on simulated data and validated with experimental brain data.

Main Results:

• Simulation-based inference achieved reliable parameter estimates with up to 90% fewer measurements.
• Outperformed standard fitting under noisy and sparse data conditions.
• Demonstrated robustness across various models, sampling schemes, and patient data.

Conclusions:

• Simulation-based inference enables fast and robust microstructural imaging.
• Substantially reduced scan times improve clinical feasibility.
• Expands dMRI access, supports privacy, and enhances data quality for diverse applications.