Comparing Harmonization Approaches for Protocol-Related Variability in Multisite Diffusion MRI Data

Area of Science:

• Neuroimaging
• Radiology
• Neurology

Background:

• Diffusion MRI (dMRI) reveals white matter changes in Alzheimer's Disease (AD).
• Multisite dMRI data present challenges due to varying acquisition protocols.
• The Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset utilizes over 10 dMRI protocols.

Purpose of the Study:

• To compare three data harmonization methods for dMRI data.
• To assess the reduction of protocol-related variability in diffusion tensor imaging metrics.
• To evaluate the preservation of associations with cognitive impairment and AD biomarkers.

Main Methods:

• Comparison of three harmonization techniques: mixed-effects models, ComBat-GAM, and eHarmonize.
• Analysis of diffusion tensor imaging fractional anisotropy (FA) and mean diffusivity (MD) data.
• Evaluation in 1,086 ADNI3/4 participants with cognitive and PET imaging data.

Main Results:

• All three methods successfully aligned FA and MD distributions across different dMRI protocols.
• Associations between dMRI metrics and cognitive impairment were consistent across harmonization approaches.
• Associations with amyloid-beta and tau PET burden showed more variability.

Conclusions:

• Multiple harmonization strategies effectively address protocol-related variability in multisite dMRI data.
• These methods preserve crucial associations between dMRI metrics and cognitive status in AD.
• Careful consideration of harmonization is essential for robust analysis of pooled dMRI datasets.