A Deep Learning Approach to Multi-Fiber Parameter Estimation and Uncertainty Quantification in Diffusion MRI.

Area of Science:

• Neuroimaging
• Biophysics
• Computational Neuroscience

Background:

• Diffusion MRI (dMRI) is crucial for in vivo brain microstructure analysis.
• Accurate parameter inference in dMRI is challenging due to complex models, noise, and unknown fiber populations.
• Existing methods often oversimplify models, leading to biologically implausible assumptions.

Purpose of the Study:

• To develop a novel, computationally efficient method for multi-fiber parameter inference in dMRI.
• To address the challenges of variable dimensionality, low signal-to-noise ratios, and non-linear models in dMRI analysis.
• To enable scalable parameter estimation and uncertainty quantification for dMRI biophysical models.

Main Methods:

• A sequential inference method decomposing the problem into subproblems.
• Deep neural networks tailored to specific problem structures and symmetries.
• Training via simulation for amortized inference and uncertainty quantification.

Main Results:

• Demonstrated advantages over standard methods using simulations and Human Connectome Project (HCP) data.
• Showed high uncertainty in extracellular parallel diffusivity estimates under HCP-like acquisitions.
• Indicated high precision in estimating intracellular volume fraction.

Conclusions:

• The novel deep learning approach offers a scalable and accurate solution for multi-fiber dMRI parameter inference.
• The method provides valuable insights into the precision of specific microstructural parameter estimations.
• This work advances the analysis of brain white matter microstructure using diffusion MRI.