SHARD: Spherical Harmonic-based Robust Outlier Detection for HARDI Methods.

Area of Science:

• Neuroimaging
• Medical Physics
• Biomedical Engineering

Background:

• High Angular Resolution Diffusion Imaging (HARDI) models capture complex intra-voxel microarchitectures.
• Diffusion MRI sequences are susceptible to artifacts from motion and physiological noise.
• Existing robust statistical methods for Diffusion Tensor Imaging (DTI) reduce artifacts but require HARDI-specific variants.

Purpose of the Study:

• To introduce a general robust outlier imputation model for HARDI analysis.
• To mitigate artifacts prior to traditional HARDI analysis, avoiding the need for bespoke robust HARDI variants.
• To improve the accuracy and reliability of HARDI data processing.

Main Methods:

• A weighted spherical harmonic fit of diffusion-weighted MRI scans was employed.
• Spherical harmonics of 6th order generated basis functions weighted by diffusion signal for outlier detection.
• The model estimates and restores corrupted signal values caused by outliers.

Main Results:

• The imputation model demonstrated a reduction in root mean squared error of raw signal intensities.
• Significant improvement was observed for the HARDI Q-ball method using the Angular Correlation Coefficient.
• Quantitative and qualitative improvements in HARDI data were revealed.

Conclusions:

• The proposed robust outlier imputation model effectively mitigates artifacts in diffusion MRI data.
• This general pre-processing model enhances various HARDI techniques by restoring outlier diffusion signals.
• The method offers a computationally efficient solution for improving HARDI analysis robustness.