Effects of Diffusion MRI Data Harmonization on Diffusion Measures, Tractography, and White Matter Tract Parcellation

Evdokiya Knyazhanskaya, BA

Brigham and Women’s Hospital
Effects of Diffusion MRI Data Harmonization on Diffusion Measures, Tractography, and White Matter Tract Parcellation

Scientific Abstract

Background: Diffusion MRI (dMRI) data are sensitive to scanner specific differences, which makes analyzing data across scanners challenging. DMRI harmonization methods have been developed to eliminate scanner differences in multi-site data, while preserving inter-subject biological differences that may be crucial to study outcome. The effect of harmonization on dMRI post-processing analyses, like automated tract labeling, has not been thoroughly explored. This study assesses the effect of harmonization on white matter (WM) tract measures of healthy controls (HCs) in a multi-site imaging study.

Methods: DMRI data was acquired from male HCs across four sites from the DIAGNOSE CTE project. All data underwent several artefact correction steps using a custom pipeline. One site was chosen as reference and the other three sites were used as target sites for harmonization. HCs in target sites had a similar age distribution to HCs in the reference site. After our harmonization method was applied, whole brain tractography was run on all pre- and post-harmonization data of the target sites as well as the original reference data. The White Matter Analysis (WMA) package was run to identify fiber bundles, and WM tract measures were compared between pre- and post- harmonization target site data.

Results: T-tests were run to compare whole-brain fractional anisotropy (FA) between the reference site (BU) (n=18; 59.85 ± 8.70 years) and the target site (NYU) (n=16; 54.13 ± 6.57 years) before and after harmonization. Though mean FA values significantly differed between BU and NYU HCs before harmonization (p=0.0023, t=-3.34), these differences were removed after harmonization (p=0.54, t=-0.63). Further analyses are ongoing on the other two sites.

Conclusion: We showed that dMRI harmonization can robustly remove existing statistical differences across scanners and eventually would allow for higher powered statistical analyses after harmonization. In addition, it is important to ensure that harmonization approaches that directly harmonize the dMRI data preserve fiber orientation, as our method was designed to do. We will present the tractography and WMA results during the meeting.

Live Zoom Session – April 21st

research Areas


Evdokiya Knyazhanskaya, BA, Fan Zhang, Ph.D., Martha E. Shenton, Ph.D., Lauren J. O’Donnell, Ph.D., Sylvain Bouix, Ph.D., Yogesh Rathi, Ph.D., Suheyla Cetin-Karayumak, Ph.D.

Principal Investigator

Suheyla Cetin-Karayumak, Ph.D.

Affiliated Website