Simulation-Based Inference at the Theoretical Limit: Fast, Accurate Microstructural MRI with Minimal diffusion MRI...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Diffusion-weighted magnetic resonance imaging (dMRI) is crucial for non-invasive brain microstructure analysis.
• Current dMRI protocols require lengthy acquisition times due to parameter space oversampling, limiting patient access and data quality.

Purpose of the Study:

• To develop and validate a novel simulation-based inference (SBI) approach using neural networks for accelerated dMRI parameter estimation.
• To demonstrate that SBI can accurately approximate posterior distributions of diffusion parameters from experimental measurements without real-data training.

Main Methods:

• Leveraging neural networks for simulation-based inference (SBI) to directly approximate posterior distributions of diffusion parameters.
• Comparing SBI performance against standard nonlinear least squares under various sampling conditions, including noisy and sparse data.
• Applying SBI to diffusion tensor imaging, diffusion kurtosis imaging, and biophysical models of axonal density and calibre.

Main Results:

• SBI achieved accurate parameter estimation with up to 90% fewer acquisitions compared to conventional methods.
• SBI outperformed standard nonlinear least squares, particularly under noisy and sparse sampling conditions.
• The approach demonstrated robustness and generalizability across simulated and real datasets from healthy and pathological brains.

Conclusions:

• SBI offers a significant acceleration of dMRI acquisition, potentially expanding access for time-sensitive patients like children.
• This method enhances the accuracy of microstructure estimation and can improve the utility of legacy or lower-quality dMRI data.
• SBI-integrated dMRI workflows promise faster, more comfortable patient examinations and have the potential to revolutionize radiology by enabling MRI-based virtual tissue biopsy.