Using COMBAT to compare two separately acquired datasets

, ,
1

Hi All,

I’m wondering if anyone’s tried to compare datasets that use the same task (e.g. Stroop task), but were acquired with fundamentally different sequences (e.g. one uses 20channel and another using 64 channel; one using multi-band multi-echo, while another only uses multi-band)?

If so, did you use any software (e.g. ComBat) to get rid of the fundamental scanner differences before running any downstream analysis?

Thanks in advance!

2

Yeah, you can totally use ComBat for this—just treat the different hardware and sequence setups as batch effects. A quick heads-up on a few practical things, though: definitely apply it to your first-level contrast maps (like your beta or t-maps) rather than the raw time series, since you’ll want to run sequence-specific prep like TEDANA for your ME data first anyway. Also, remember ComBat isn’t magic. If your standard MB sequence has bad signal dropout in certain regions, harmonization won’t bring that signal back. Lastly, just make sure your conditions of interest aren’t completely confounded with the datasets (e.g., all patients in one dataset, all controls in the other), or the algorithm will wipe out your actual effects right along with the scanner noise. Good luck with the analysis!