I want to make sure that I understand the bandpass + censoring, as described in afni_proc.py:

1) Bandpass regressors (sin-wave functions of increasing order) are included in a GLM, along with all other regressors.

2) Censored time-points are removed from the voxel time-series and all other regressors.

3) GLM is calculated on remaining time-points.

This approach is to avoid FFT bandpassing first, then censoring, as data within the censored time-points would be smoothed into other data. Or sensoring first and then FFT bandpassing, as the temporal structure of the data would be different.

Does the bandpass + censoring approach (1,2,3 above) decribe the -regress_censor_motion option correctly?

Would the bandpass + censoring approach yield significantly different results from adding individual predictors for each censored time-point to the GLM?

If the results are the same, why is data-removal prefered?

Thanks very much,

Andrew



Edited 1 time(s). Last edit at 07/08/2015 03:48PM by mattare2.