Estimation of white matter fiber parameters from compressed multiresolution diffusion MRI using sparse Bayesian...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biophysics

Background:

• Diffusion MRI (dMRI) is crucial for mapping white matter architecture.
• Estimating fiber parameters from compressed dMRI data remains challenging.
• Existing methods struggle with accuracy and uncertainty in under-sampled datasets.

Purpose of the Study:

• To introduce BusineX, a novel sparse Bayesian unmixing algorithm.
• To enhance the estimation of white matter fiber parameters from compressed dMRI data.
• To improve reconstruction accuracy and reduce uncertainty, particularly with limited diffusion gradients.

Main Methods:

• BusineX combines compressive sensing with linear unmixing and sparse Bayesian learning.
• It formulates fiber parameter estimation as a sparse signal recovery problem.
• The algorithm utilizes a parametric spherical deconvolution model and dictionary representation with data-dependent hyperparameter learning.

Main Results:

• BusineX demonstrates improved accuracy and lower uncertainty in fiber parameter estimates on synthetic and in-vivo data.
• The algorithm successfully resolves a higher number of complex fiber crossings (second and third order).
• BusineX provides more reliable estimates from under-sampled dMRI data.

Conclusions:

• BusineX offers a robust solution for estimating white matter fiber parameters from compressed dMRI.
• The algorithm's sparse Bayesian approach enhances reconstruction and reduces uncertainty.
• BusineX advances the analysis of complex white matter architecture using under-sampled dMRI data.