Outliers in diffusion-weighted MRI: Exploring detection models and mitigation strategies.

Area of Science:

• Neuroimaging
• Medical Physics
• Biomedical Engineering

Background:

• Diffusion-weighted MRI (dMRI) is crucial for studying brain microstructure and connectivity.
• dMRI data processing is complex and susceptible to motion-induced signal dropout artefacts.
• Accurate artefact correction is vital for clinical dMRI research.

Purpose of the Study:

• To compare outlier replacement and downweighting methods for dMRI data processing.
• To guide the dMRI community in selecting optimal data processing tools.
• To evaluate the impact of these methods on motion correction and tensor modeling.

Main Methods:

• Simulated realistic whole-brain dMRI data with varying dropout artefacts.
• Applied Gaussian Process (GP) and Spherical Harmonic (SH) based outlier replacement.
• Implemented outlier downweighting techniques.
• Evaluated motion correction, registration, and single tensor model fitting.

Main Results:

• GP-based outlier replacement yielded tensor fit results comparable to GP-based outlier downweighting.
• Both methods effectively addressed signal dropout artefacts in simulated and infant dMRI data.
• Outlier downweighting may offer improved model precision estimates.

Conclusions:

• Outlier replacement is recommended when the primary interest is the least-squares estimate of the single tensor model.
• Outlier downweighting is potentially more suitable for applications requiring precise model estimation, such as probabilistic tractography.
• The choice between methods depends on the specific analytical goals in dMRI research.