search

rpomponio / neuroHarmonize

Harmonization tools for multi-site neuroimaging analysis. Implemented as a python package. Harmonization of MRI, sMRI, dMRI, fMRI variables with support for NIFTI images. Complements the work in Neuroimage by Pomponio et al. (2019).
https://pypi.org/project/neuroHarmonize/
MIT License
82 stars 28 forks source link

How many subjects per scanner are required? #34

Closed Melissa1909 closed 9 months ago

Melissa1909 commented 1 year ago

Dear Raymond, thanks for providing this awesome repo!

I was wondering how many subjects are necessary to get substantial estimations for the scanner effect? I've read this post (https://github.com/rpomponio/neuroHarmonize/issues/16) and have seen your response saying that fitting the model on healthy controls alone would make sense and that the amount has to be large enough. What would you consider "large enough"? I'd have a dataset with 4 sites (n(HC)=140), with one of them having only 25 healthy controls. Best, Melissa

rpomponio commented 9 months ago

I'm not aware of a consensus threshold for the amount of subjects needed per site in harmonization, but 25 Healthy Controls is near the lower-limit, especially if you are including multiple covariates (age, sex, etc.).

The empirical Bayes step should mitigate small sample effects by shrinking the site effects towards the pooled mean. It would be interesting to see results with and without EB.

Closing this issue for now, but feel free to open with additional comments.