Nonnegative definite EAP and ODF estimation via a unified multi-shell HARDI reconstruction

Area of Science:

• Neuroimaging
• Diffusion MRI
• Computational Neuroscience

Background:

• High Angular Resolution Diffusion Imaging (HARDI) uses Orientation Distribution Functions (ODFs) and Ensemble Average Propagators (EAPs) to map water diffusion and neural pathways.
• Current methods like Spherical Polar Fourier Imaging (SPFI) often fail to enforce the physical non-negativity constraint on ODFs and EAPs.
• Existing Riemannian frameworks for ODFs/EAPs rely on pre-estimated values and do not directly estimate the underlying wavefunction.

Purpose of the Study:

• To propose a unified model-free multi-shell HARDI method for simultaneous estimation of EAP wavefunctions, ODFs, and EAPs.
• To incorporate the non-negativity constraint directly into the estimation process.
• To improve the robustness and accuracy of HARDI parameter estimation, particularly in noisy conditions.

Main Methods:

• Developed the Square Root Parameterized Estimation (SRPE) method, integrating Riemannian geometry and Spherical Polar Fourier (SPF) basis representation.
• Directly estimated the wavefunction of EAPs, enabling simultaneous estimation of non-negative definite ODFs and EAPs from diffusion signals.
• Utilized multi-shell HARDI data for model-free estimation.

Main Results:

• SRPE demonstrated superior robustness to noise compared to existing methods like SPFI.
• The new method achieved better Ensemble Average Propagator (EAP) reconstruction, especially for profiles at larger radii.
• Experiments on both synthetic and real diffusion imaging data validated the performance of SRPE.

Conclusions:

• SRPE offers a principled approach to HARDI parameter estimation by directly enforcing physical constraints.
• The method provides more accurate and robust estimations of ODFs and EAPs, advancing diffusion MRI analysis.
• SRPE shows significant potential for improving the interpretation of complex neural tissue microstructures from diffusion MRI data.