Thanks for rapid response rick. For "a bit unconventional" I was referring to my decision to both try and remove the lowest, hopefully intrinsic, frequencies, and doing so inside a task based functional connectivity analysis between task conditions. Working in a lab that focuses mostly on resting state FC this is all way out in left field. *laugh*

OK so, I am doing this all the old-school step by step 3dDeconvolve way because all the original analyses for this dataset were completed two years ago this way. There were some strange FC findings during null vs strong condition ANOVA that I suspect is intrinsic connectivity so I am simply trying to redo the analysis but with a Bandpass filter slipped in, notably in the correct place, so I can show a direct comparison of analyses (apples to apples) to test my theory.

Yes, that little autocensor during step1 was just creation of .1D binary file and did not involve any removal till 3dDeconvolve.

From your comments would this then be a more correct way to order things, while adhering at much as possible to original processing? Does censoring early keep any spikes from "ringing across time"?

1. basic preprocessing including: Skull strip, autocensoring, 3dvolreg, creating motion regressors, epi + anat alignment, spatial normalization, automask + FWHM spatial smoothing.
2. 3dcalc func * 1D autocensor to remove bad time points early
3. Apply scaling to func data
4. 3dBandpass this preprocessed data
5. 1dBandpass the motion regressors
6. Concatenate runs of both
6. 3dDeconvolve -errts with the 1dBandpassed motion regressors, NO censors, and the stim_times of those task conditions of no-interest
7. 3dmaskave the relevant ROIs
8. Move .1D outputs into Matlab to generate necessary correlation coefficents and ttests.

If it is absolutely necessary to run censors and bandpass concurrently, would it be possible to just swap in 3dTproject to for 3dDeconvolve or something of the like?

~`Dane



Edited 1 time(s). Last edit at 07/01/2014 12:13AM by d6anders.