3dttest++ on resting state data for clusters extraction

AFNI Message Board
1

Hi everybody,
I already run a seed-based correlation from animals resting state data.
I would like to extract the connected clusters with a threshold corrected for FDR

I have two fMRI data for each animal.
I already coregistrate and pretreated my fMRI data and performed a classical seed-based correlation.

Now, I want to use 3dttest++ on all my z-fisher transformed data (14 animals x2)
So I first try to use 3dttest++ with Clustsim :

‘3dttest++’, ‘-setA’, subprocess.list2cmdline(listemap), ‘-Clustsim’, ‘-prefix’, ‘clustersim-stat_fisher_’ + atlas_name2 + ‘.nii.gz’

and then I try to use the output of this analysis in clusterize to extract them

The problems I found where :

  • Could I use 3dttest++ with non-independent (2 fold the same animal) data
  • Who could I use a threshold non-corrected pvalue at p<0.05 to found the good cluster-thresholding FDR corrected ?
  • I do not know if clusterize already use the output 1D table from this analysis ? If not who could I use it ?
  • The nii output contains 2 images ('SetA_mean, and SetA_Zscr), do I have to apply clusterize on the first, second or load both of them in afni ?

I also found that there is an ETAC option,

It is maybe better to use this one ?

So my other questions are:

  • Which kind of mask is needed ? My data were already masked but it seems obligatory…
  • Is there specific option that I need to use ? -ACF ?
  • The default threshold is p<0.05 ?

Thanks for your time !

Dough

2

Hi Dough,

The row is based on an uncorrected p-value threshold, listed
in the first column. The column is based a corrected p-value
threshold, listed above the dashed lines. The combination
would be required cluster size.

For example, using an uncorrected p-value of 0.001 for the
voxelwise threshold, to get a corrected p-value of 0.05 at
the cluster level would require a minimum cluster size of
22 voxels (from 21.9 in the table).

  • Such a cluster table is produced by 3dttest++ -Clustsim,
    though it is up to you to apply the threshold/cluster step
    (it does not assume which p-values to apply).

  • The thresholding and clustering is applied to a t-stat or
    z-score volume. The mean volume would probably be used for
    colorization, showing the effect size.

  • The choice of ETAC is up to you. The ETAC output is not
    a cluster table but a mask of surviving voxels.

  • It is appropriate to specify your group mask in the
    clusterizing operation (e.g. to 3dttest++). It is nice to
    generate unmasked results as well, for review.

  • The -ACF option applies when you have computed ACF
    estimates, probably at the single subject level, from
    afni_proc.py, and when you are applying them in 3dClustSim.
    Previously (in your post), 3dClustSim was not mentioned.

  • There is no default threshold, except in the ETAC case.
    But a corrected p of 0.05 is the most common.

  • rick

3

This reply helps a lot. My query is (which might be silly), since you said:

“”"

  • Such a cluster table is produced by 3dttest++ -Clustsim,
    though it is up to you to apply the threshold/cluster step
    (it does not assume which p-values to apply).

  • The thresholding and clustering is applied to a t-stat or
    z-score volume. The mean volume would probably be used for
    colorization, showing the effect size.
    “”"

I got multiple tables after running 3dttest++ -Clustsim on 20*2 subjects, how do you proceed from there i.e. how do I apply thresholding and clustering now on z-score maps for each subject? I don’t know how to interpret the results. If I open AFNI, select some p-value and cluster-size value, I get a Report which says peak-x, peak-y, peak-z values which again I don’t how to interpret. Any help or direction is appreciated.

4

You should choose a cluster table depending on
the desired neighbors and sidedness of the test.
That table would give you a minimum cluster size,
based on the uncorrected and corrected p-values
you are using.

Then that cluster size is applied to the 3dttest++
statistical output (either in the afni GUI or in a
3dclust or 3dClusterize command).

  • rick
5

Thanks a lot for your precise answer! It really help!
Happy to see that i was not alone in that situation!
Dough