ETAC output file interpretation

I am running ETAC for the first time and am having trouble figuring out what all the output files are, even after reading the help and a slide deck I found.

I ran ETAC with the following commands:
-ETAC_blur 0 2
-ETAC_opt NN=1:sid=2:hpow=1:pthr=0.05,0.01:name=Fred1
-ETAC_opt NN=2:sid=2:hpow=1:pthr=0.05,0.01:name=Fred2 \

I got the following output files:


This seems to be suggested as the file we want to look at:
This should be: a binary mask file indicating which voxels survived the multi-thresholding process. Now does that mean the various smoothing and thresholds specified? and hpow?

If I want to investigate which combination of smoothing and threshold value a voxel survived, then I can look at:
Where in theory, if I combined the b0 .05 and the b2 .05 maps they should match

From reading the help it seems like you are suggesting the global threshold, even though one of the cool things when ETAC first came out seemed to be the threshold that varied across cortex? We are seeing some pretty major differences between the local and global masks.
In that there is much more activation that survives in the ‘local’ thresholded file. However, I am also not sure I know what to look at, as there are now 8 layers instead of 4 (two each for each smoothing and threshold level).

I also ran this once with hpow1,2, and I could not tell in the output files what happened to the hpow values. I didn’t get any more layers or files, so I wasn’t sure how the two levels were represented.

Thank you!
Stephanie McMains

ETAC does a lot of things, and the output files reflect various combinations of those.

First of all, I cannot recommend the “local” option, as I was unable to get it to control the false positive rate reliably enough, and eventually I gave up on it. (It was a lot of code, and I left it in the 3dXClustSim program as a potential avenue for future work.) The local option is not vetted by publication, so you can’t really cite any paper to support using it - whereas, the global ETAC option does have a paper you can cite to fight off the reviewers.

Second, the file is, as you say, the binary mask of voxels that survived any of the thresholding cases. And the file shows which voxels passed each individual thresholding case – collapsing across cases should give the same as

Third, running with multiple values of hpow is something I hoped would help, but does not in fact make much difference in the cases I tested - that is, the clusters that result don’t differ much between different values of hpow.

Thanks so much for the clarification. I had one follow-up about some weirdness I am seeing within the global output.

I tried it first with two hpowers based on something I had seen in an example:

-ETAC 8 -ETAC_blur 0 2 -ETAC_opt NN=1:sid=2:hpow=0,2:pthr=0.05,0.01:name=Fred1
This resulted in a fair amount of activity surviving the mask.
But then I saw the comments in the etac help which suggested running hpow=0 for ease of explaining and hpow=2 if you expect small clusters (one main area of interest is small, so I thought this might be good). I ran them both separately.

-ETAC 8 -ETAC_blur 0 2 -ETAC_opt NN=1:sid=2:hpow=0:pthr=0.05,0.01:name=Fred1
nothing survived

-ETAC 8 -ETAC_blur 0 2 -ETAC_opt NN=1:sid=2:hpow=2:pthr=0.05,0.01:name=Fred1
nothing survived

For ‘fun’ I tried a few more:

-ETAC 8 -ETAC_blur 0 2 -ETAC_opt NN=1:sid=2:hpow=1:pthr=0.05,0.01:name=Fred1
nothing survived

-ETAC 8 -ETAC_blur 0 2 -ETAC_opt NN=1:sid=2:hpow=0,1,2:pthr=0.05,0.01:name=Fred1
mostly overlapping with the initial run using hpow=0,2

So this leaves me pretty confused. I obviously want to do things correctly, and I would love for something to survive!! Any Idea why there is a difference if you run them separate versus combined? And there is no way to see the individual contributions from the hpow settings when you use more than one in an etac command?

For reference, this is a paired t-test on 16 subjects using AFNI 18.3.05.

Thank you!
Stephanie McMains

Just wanted to bump this up, as it is hanging me up and blowing my mind :).

There are a few points that come to my mind.
[li] You only have 16 samples (since a paired test is equivalent to a 1-sample test using A-B as the inputs). The program allows that, but it is really too few for good statistics in the permutation/randomization. That may be one cause of your lack of results.
[/li][li] Was any blurring applied to the data during the processing that lead to the 3dttest++ run? That is, if you used, did you specify blurring in the time series analysis? I ask because the amount of blurring you use is very low (0 and 2 mm) – if this is the only blurring ever applied, it is too small to show FMRI results in most cases.
[/li][li] I would use a wider range of pthr values, for example ‘pthr=0.01/0.001/19’ to specify 19 thresholds evenly spaced from 0.01 to 0.001, with step size (0.01-0.001)/(19-1)=0.0005. Using a range of pthr values is the main point of ETAC, and is what I was thinking of when I developed the idea while hiking in the Grand Canyon.
[/li][li] Although using multiple hpow values seemed like a good idea to me at the time, it doesn’t actually seem to make much difference. I now usually just use ‘hpow=2’

I hope this helps, and apologize for the delay. If you have any followup questions, please email me. (The homework is to somehow find my email address.)