Uncertainty mapping and probabilistic tractography using Simulation-based Inference in diffusion MRI: A comparison...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Physics

Background:

• Diffusion MRI (dMRI) enables non-invasive brain white matter imaging.
• Parametric spherical deconvolution requires robust Bayesian inference for accurate fiber orientation estimation and uncertainty quantification.
• Established methods like Markov-Chain Monte-Carlo (MCMC) are computationally intensive.

Purpose of the Study:

• To introduce and evaluate a Simulation-Based Inference (SBI) framework for parametric spherical deconvolution in dMRI.
• To compare the performance of SBI against MCMC for estimating white matter fibre orientations and their uncertainties.
• To assess the utility of SBI for probabilistic tractography.

Main Methods:

• Developed an SBI framework utilizing neural networks trained on simulated dMRI data.
• Applied the framework to estimate fibre orientations and uncertainty in voxel-based analyses.
• Performed probabilistic tractography by propagating orientation uncertainty.
• Conducted extensive comparisons with MCMC-based Bayesian methods.

Main Results:

• In-silico trained SBI networks achieved calibrated, accurate parameter estimates and uncertainty mapping for single- and multi-shell dMRI.
• SBI provided amortised inference, enabling posterior distribution estimation for unseen data orders of magnitude faster than MCMC.
• SBI-based tractography showed high agreement with MCMC results, matching or exceeding scan-rescan reproducibility.

Conclusions:

• SBI presents a powerful, fast, and accurate alternative to classical Bayesian inference for dMRI model estimation and uncertainty mapping.
• The framework facilitates efficient probabilistic tractography with reliable uncertainty quantification.
• Optimal SBI design practices were identified, enhancing its applicability in neuroimaging research.