Identifying microstructural changes in diffusion MRI; How to circumvent parameter degeneracy.

Area of Science:

• Biophysics
• Computational Neuroscience
• Medical Imaging

Background:

• Biophysical models often suffer from over-parameterization, leading to degenerate parameter estimation.
• Identifying changes in model parameters between conditions is crucial for many applications, but challenging with ill-posed inversions.

Purpose of the Study:

• To develop a Bayesian framework, Bayesian Estimation of Noise in Change (BENCH), for inferring parameter changes in biophysical models despite degeneracies.
• To enable robust estimation of parameter changes even when direct model inversion is ill-posed.

Main Methods:

• BENCH employs a two-step Bayesian approach: training models to predict measurement changes from parameter change directions using simulations.
• Pre-trained models are then used to assess the probability that observed data differences are explained by specific models of change.

Main Results:

• The framework is applicable to diverse data and models, particularly beneficial for biophysical models with sparse parameter degeneracies.
• Simulations demonstrate BENCH's ability to identify microstructural parameter changes in white matter using diffusion MRI data.
• Application to UK-Biobank data successfully identified dominant parameter changes in white matter hyperintensities.

Conclusions:

• BENCH provides a robust method for inferring parameter changes in over-parameterized biophysical models, overcoming limitations of traditional inversion techniques.
• The approach is particularly effective for diffusion MRI microstructural modeling and has potential for clinical applications like analyzing white matter hyperintensities.