Variational inference of the fiber orientation density using diffusion MR imaging.

Area of Science:

• Neuroimaging
• Biophysics
• Computational Neuroscience

Background:

• Diffusion Magnetic Resonance (MR) imaging allows in vivo exploration of human brain connectional architecture.
• It reveals long-range nerve fibers integrating functionally distinct cortical areas.
• A forward model maps nervous tissue microgeometry to diffusion-weighted signals, as fibers are inferred from water diffusion.

Purpose of the Study:

• To propose a generalized spherical deconvolution method for estimating fiber orientation density.
• To solve the inverse problem within a smoothing spline framework, preserving density properties.
• To incorporate a Gaussian process model for confidence bands and hyperparameter selection.

Main Methods:

• Spherical deconvolution of fiber orientation density in a reproducing kernel Hilbert space.
• Smoothing spline framework for solving the inverse problem, ensuring normalization and non-negativity.
• Gaussian process modeling for confidence intervals and hyperparameter selection, relaxing constant diffusivity assumption.

Main Results:

• A generalized approach to spherical deconvolution is presented, improving upon truncated Fourier analysis.
• The method effectively preserves the properties of a density function.
• Demonstrated ability to uncover the fiber orientation field in white matter using high angular resolution diffusion-weighted data.

Conclusions:

• The proposed method offers a robust framework for analyzing white matter architecture from diffusion MRI.
• It provides confidence bands for estimated fiber orientation densities, enhancing result reliability.
• This approach advances the understanding of brain connectivity by accurately mapping fiber orientation fields.