Systematic Comparison of Various Voxelwise Harmonization Techniques for Diffusion MRI

Motivation: Diffusion MRI (dMRI) is crucial for studying white matter (WM) changes related to aging, but reproducibility suffers due to scanner and protocol variability. Even large-scale studies face challenges with inter-scanner differences. Goal(s): This study compares various harmonization techniques including ComBat and RISH to validate reduction in scanner-induced variability in diffusion tensor imaging (DTI)-metrics aiming to improve reproducibility in multisite dMRI research. Approach: Data from four 3T-MRI scanners were harmonized using the established protocols. Statistical analyses tested both site-specific and harmonized correlations to evaluate each method. Results: Though ComBat and RISH showed similar post-harmonization correlations, ComBat had a higher overlap with site as regressors. Impact: Our study suggests voxelwise dMRI harmonization using either ComBat or RISH should be interpreted with caution. Furthermore, when need be, traditional methods of using site as regressors may be as powerful as ComBat or RISH for statistical analysis.