I’m wondering what you mean by ‘projection components are created here by projecting the good ones out of the bad ones’ in the @extract_meica_ortvec help? And do I need to add the orts regressors to my 3dDeconvolve? thx!


It might be good to run this script in the context of afni_proc.py, and see how it handles things. The proc script would handle projecting the components in the final regression.

The meica.py/tedana.py that comes with AFNI partitions the components into ignored, accepted, midk_rejected and rejected. By default, the accepted components are considered “good”, and the 2 rejected ones are considered “bad” (this can be altered via -reject_midk 0). A typical usage of tedana.py would have it project the rejected components by:

  1. fit all components to the data
  2. subtract the fit of the rejected ones

But the “bad” and the “good” components are not orthogonal. So it might make sense to project the “good” components out of the “bad” ones. That might be a less aggressive projection, hopefully keeping more of the desired signal in the data.

That is what this script does. It creates a file of bad components that are orthogonal to the good ones.