Harmonizing multi-site, multi-scanner fMRI data with ComBat

1

Hello,

I am planning to use ComBat to harmonize fMRI data but I have a question about the batch level. There are seven cohorts (all with different scanning protocols). Some cohorts includes multiple sites, and some sites multiple scanners. It’s my understanding from the literature that I should use scanner as the batch level, however since some sites from different cohorts use the same scanner, should I further categorise my batches to include site and/or cohort variability, e.g. siteA_siemens_triotim, siteB_siemens_triotim?

I am struggling to find guidance on this and would be appreciative of any pointers! Thanks.

2

Hi @nclarke,

I would create a different level for each scanner-site combination. I also recommend looking into Neuroharmonize (paper at https://www.sciencedirect.com/science/article/pii/S1053811919310419), which is built on ComBat, if you suspect sites/scanners might have different demographics.

Best,
Steven

1 Like
3

Hi @Steven, that’s great, thank you very much! I’ll look into Neuroharmonize too.