Evaluating Transferability of ComBat Harmonization of Diffusion Tensor Magnetic Resonance Imaging Data

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Data Science

Background:

• Multisite MRI data harmonization is crucial for large-scale studies.
• Traditional ComBat requires recomputation for new subjects, limiting scalability.
• Diffusion tensor MRI (DT-MRI) data presents unique harmonization challenges.

Purpose of the Study:

• To evaluate the efficacy of a transferable ComBat (T-ComBat) algorithm for harmonizing DT-MRI data.
• To determine the optimal size of a training data pool for T-ComBat.
• To assess if T-ComBat can harmonize new subject data without re-analyzing existing data.

Main Methods:

• Applied T-ComBat to harmonize fractional anisotropy (FA) and mean diffusivity (MD) maps from 314 adolescents across two MRI sites.
• Varied the size of the training data pool to assess T-ComBat performance.
• Assessed harmonization effectiveness by testing for significant voxel-wise and region-of-interest (ROI) differences across sites.

Main Results:

• T-ComBat demonstrated improved cross-site harmonization compared to no harmonization.
• T-ComBat performance was slightly lower than full ComBat.
• Sufficient harmonization was achieved with approximately 25% of subjects for FA and 10% for MD in the training set.

Conclusions:

• T-ComBat offers a viable solution for harmonizing DT-MRI data from new subjects without re-analyzing previously harmonized data.
• The method is effective for data from previously encountered scanners.
• This approach enhances the efficiency of multisite DT-MRI data analysis.