Clustsim Failure in 3dttest++, any tips for small volume correction?

AFNI Message Board
1

I am using the -Clustsim option in 3dttest++ to try and do a small volume correction for a bilateral dlPFC mask. Input files are from a seed-based resting functional connectivity analysis. I’ve successfully implemented the correction using e.g. a whole brain mask, but with the smaller mask am running into problems in which 3dClustSim “fails” for some seed regions and/or group comparisons. If I re-run the identical analysis, different seeds and/or group comparisons fail. However, for some seeds/comparisons on each run through it works just fine, suggesting the mask is being read in to the program okay.

In the terminal window I’m getting errors like this - that one of the .sdat files isn’t long enough:
Example 1:
++ loading -setB datasets
562019.52562** FATAL ERROR: data file ‘./CS.AD-Risk.N146b.Amg+Ce_Bilateral_D1E1_Oct2020_reorient.0trxb.NwcmdfX0.4.aw.s6.mni.0181.sdat’ isn’t long enough
** Program compile date = Jan 7 2021
645*+ WARNING: 3dttest++ -setB :: 234 vectors are constant
++ t-test randomsign/permute:0** FATAL ERROR: ===== 3dClustSim command failed :-((( =====

Example 2:
** FATAL ERROR: data file ‘./CS.Risk-Control.N146b.2mm_newCe_D1E1_Bilateral.0trxb.NwcmdfX0.4.aw.s6.mni.0186.sdat’ isn’t long enough
** Program compile date = Jan 7 2021
18** FATAL ERROR: ===== 3dClustSim command failed :-((( =====

Are there any options known to improve Clustsim performance with smaller masks? Any insight into what this error means would also be helpful - the .sdat files are not easily read (binary files?) so comparing one that is long enough versus one that is not has not been informative.

Thanks!

2

These “isn’t long enough” messages are caused by the .sdat files written during the 3dttest++ randomization runs being truncated somehow. Not empty and not non-existent, because those are different messages.

But how that file truncation happened, I don’t understand. The only thing I can think of is that the disk storage is full, but that seems very unlikely, and you would almost certainly have a whole host of other messages in that situation. Although the program has many error messages built in, most of them have never been actually seen in practice – including these “isn’t long enough” messages.

Do you have the .sdat files saved? How big are they? How many voxels in your mask? The files should be exactly 2*number of mask voxels long – the ‘s’ means ‘short’ and these files are ‘short integers’ = 16 bit = 2 byte values.