gPPI: IRF and detrending

AFNI Message Board
1

Hi AFNI experts,

I had few questions about the https://afni.nimh.nih.gov/pub/dist/edu/data/CD.expanded/AFNI_data6/FT_analysis/PPI/cmd.ppi.2.make.regs script and generating IRF.
set dur = 15 # use a 15-second curve for BLOCK
set nt_irf = ccalc -i "$dur/$TRup"

set hrf_file = x.BLOCK.1D
3dDeconvolve -nodata $nt_irf 0.1 -polort -1
-num_stimts 1
-stim_times 1 1D:0 “BLOCK(0.1,1)”
-x1D $hrf_file -x1D_stop

  1. Should the 0.1 within the -nodata option and the BLOCK() function be the same as $TRup? Currently, my TRup is 0.5 and without changing those options (using 0.1 for both), it creates the same IRF when using TRup of 0.1 but with 30 time points instead of 150.

  2. If so, should the change of 0.1 to $TRup apply to both the -nodata option and BLOCK() function or only the -nodata option. Using a TRup of 0.5, each option provides a slightly different IRF that looks similar to when using a TRup of 0.1 but with only 30 time points instead of 150.

  3. Also, is time being controlled for twice for the physiological variable? Once when the physiological seed time series is detrended and also again during the polort option of 3dDeconvolve?

Thanks!,
Ekarin

2

Hi Ekarin,

  1. Yes, I missed that when generalizing the script. So
    both of those 3dDeconvolve -nodata commands should have
    -nodata $nt_irf $TRup

You should be able to see the entire curve in those
output files (like x.BLOCK.1D), and the TR should match
the data.

  1. The BLOCK(0.1,1) is fine, that 0.1 does not change.
    It is meant to be the shape of an impulse response (and
    a duration of 0 is not allowed, so I used 0.1).

  2. Do you mean when the errts dataset is used to generate
    the seed time series, along with the final regression?
    That is certainly okay. We could almost run a regression
    on the reduced model, but it is viewed as safer to apply
    the entire model. The results should be very similar,
    but note that the temporally partitioned PPI regressors
    will not necessarily be orthogonal to motion, just for
    example. Anyway, it seems safer to use the full model.

  • rick
3

Thanks, Rick, that was my question!

4

If I wanted to use SPMG1 as the IRF, what should the ideal duration be?

Thanks!,
Ekarin

5

Hi Ekarin,

That is a 25 s curve. Is that the same basis function
that was used in the main regression (or at least one
of them, e.g. via SPMG2)?

  • rick
6

Thanks, Rick, I was planning on using the same basis function for the main regression. I was interested in comparing the differences in results using different basis functions (i.e., GAM, BLOCK, and SPMG1).

7

That is great, keeping the basis function consistent.

  • rick
8

I had a follow-up question on the IRF basis. If I was interested in comparing the results of not centering, mean centering, and zero-centering the PHYS term, how could I convolve the mean- and zero-centering PHYS term with one of the IRF bases and take that into account in the subject level analysis?

Thank you,
Ekarin