A unified filtering method for estimating asymmetric orientation distribution functions.

Area of Science:

• Neuroimaging
• Diffusion Magnetic Resonance Imaging (dMRI)
• Computational Neuroscience

Background:

• Estimating asymmetric orientation distribution functions (ODFs) in dMRI involves numerous complex filtering methods.
• Existing methods often share similarities, making them difficult to differentiate and integrate.

Purpose of the Study:

• To disentangle and unify existing filtering methods into a single, novel equation for asymmetric ODF estimation.
• To introduce a self-supervised, data-driven approach for calibrating filtering parameters.
• To develop an open-source, GPU-accelerated software for practical application.

Main Methods:

• Developed a unified filtering equation for asymmetric ODFs.
• Implemented a self-supervised, data-driven parameter calibration method.
• Created an open-source, GPU-accelerated Python software for dMRI processing.

Main Results:

• Applied the method to multi-shell and single-shell dMRI datasets, evaluating asymmetry at different resolutions.
• Introduced a novel index, number of fiber directions (NuFiD), to characterize fiber configurations.
• Demonstrated that the unified filtering method identifies asymmetric configurations in over 50% of brain voxels.

Conclusions:

• The unified filtering approach simplifies and enhances asymmetric ODF estimation in dMRI.
• The self-supervised calibration method automates parameter tuning.
• The findings reveal significant widespread brain asymmetries, with implications for understanding neural structure.