Retrospective harmonization of multi-site diffusion MRI data acquired with different acquisition parameters

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Data Science

Background:

• Joint analysis of multi-site diffusion MRI (dMRI) data increases statistical power for neuroimaging studies.
• Scanner-specific nonlinear effects and acquisition parameter differences prevent naive pooling of dMRI datasets.
• Harmonization is essential to remove scanner-specific effects for integrated dMRI analysis.

Purpose of the Study:

• To propose and validate a dMRI harmonization method for removing scanner-specific effects.
• To account for minor differences in acquisition parameters like b-value, spatial resolution, and gradient directions.
• To determine the minimum number of well-matched subjects needed for effective harmonization.

Main Methods:

• Developed a dMRI harmonization algorithm to remove scanner-specific effects.
• Validated the method on dMRI data from the Philadelphia Neurodevelopmental Cohort (PNC) and Brigham and Women's Hospital (BWH).
• Assessed the preservation of age and gender-related maturation differences using effect sizes.

Main Results:

• The proposed method successfully removed scanner-specific effects while preserving key neurodevelopmental differences.
• Age and gender-related maturation differences were maintained irrespective of sample size after harmonization.
• At least 16-18 well-matched healthy controls per site are required to reliably capture scanner-related differences.

Conclusions:

• The developed dMRI harmonization method enables retrospective harmonization of multi-site data despite acquisition parameter variations.
• The method preserves inter-subject anatomical variability, crucial for accurate neuroimaging research.
• This approach facilitates more powerful and comparative studies of brain disorders using pooled dMRI data.