Hello!
I am looking to use ComBat to harmonize some new structural MRI data with the RBC data. I’m aware that the original, unharmonized RBC data is unavailable, but is the original covfam model available? Looking to use the original parameters to extend the harmonization to this new data with ComBat’s predict function. Would also be open to any suggestions for a different approach!
Hi @jkropf and welcome to neurostars!
Yes, the code is available here:rbc_data_analysis/code/covbat.R at main · gshafiei/rbc_data_analysis · GitHub
You might have mistyped, but it is the harmonized data that are unavailable. The original, unharmonized data are publicly available.
Best,
Steven
Hello!
Yes, an example script on how to use covbat to harmonize RBC MRI data is available here: rbc_data_analysis/code/covbat.R at main · gshafiei/rbc_data_analysis · GitHub
More information on ComBatFamily package is available here:
thanks @Steven and @gshafiei for your quick replies!
I see that the outputs from that script are the harmonized data, however my understanding is that to harmonize new data with the same parameters, I’d need the actual covfam model/parameters (e.g., from data.harmonized in line 80 of that script you linked) used to produce those harmonized data. I think I’d then use that model with predict.covfam to extend that harmonization model to my new data. Is this the best way to go about this? New to MRI harmonization, so if there is a better way to harmonize new data with a pre-harmonized dataset, I’m happy to hear it!
Hi @jkropf,
That would not work, as you would need to train a harmonization model on all data, including yours. Unless of course your new data are from the same original datasets, which seems unlikely.
Best,
Steven
MANAGED BY INCF