Inter-site and inter-scanner diffusion MRI data harmonization.

Area of Science:

• Neuroimaging
• Medical Physics
• Biomedical Engineering

Background:

• Diffusion MRI (dMRI) data acquired across different sites and scanners exhibit significant variability.
• This variability hinders joint analysis, limiting sample size and statistical power in neuroimaging studies.
• Existing harmonization methods often rely on compartmental modeling or fail to account for spatial signal variations.

Purpose of the Study:

• To develop and validate a novel method for harmonizing dMRI data acquired from multiple sites and scanners.
• To enable robust joint analysis of dMRI data by correcting for scanner-dependent spatial signal variability.
• To maintain inter-subject variability and avoid reliance on diffusion compartmental models.

Main Methods:

• The method uses spherical harmonics to represent dMRI signals and computes rotation-invariant features.
• Region- and tissue-specific linear mappings are estimated to correct for scanner-related differences.
• A feature-based refinement of brain parcellation (e.g., Freesurfer) is proposed to improve harmonization accuracy.

Main Results:

• The proposed method effectively removes scanner-specific differences in dMRI data acquired from seven different sites and scanner types (GE, Philips, Siemens).
• Statistical comparisons of diffusion measures (FA, MD, GFA) show significant reduction in site-specific variability post-harmonization.
• Tract-based spatial statistics (TBSS) analysis confirms the efficacy of the harmonization method.

Conclusions:

• The novel dMRI harmonization method successfully corrects for scanner-induced biases while preserving biological variability.
• This approach significantly enhances the reliability and power of multi-site dMRI data for large-scale neuroimaging research.
• The method offers a robust solution for harmonizing dMRI data, independent of specific diffusion models.