3dClustSim from orig to standard voxel size

Hi Gurus:

I’ve gone though the entire process of using 3dFWHMx and 3dClustSim to get cluster thresholding values.
I understand that it is more accurate to run 3dClustSim with a mask that is the same dimensions as what you got off the scanner so that’s what I did. Basically I ran 3dFWHMx on the 4x1.8x1.8 .errts files and got the acf values from that and then used a 4x1.8x1.8 mask file with the acf values as the input to 3dClustSim.

My question is: We want to apply these values to a statistical dataset that is 1x1x1. In order to get the correct cluster sizes should we just multiply the cluster sizes in the table by 13 (1.8x1.8x4)?

I guess I’m just curious about “converting” cluster sizes among datasets of different dimensions.

Thanks for any insight!

Hi Lauren,

It is most accurate to report results at a voxel size that is
akin to the original one. That is to say, it is preferable not
to report results at 1x1x1. However, with voxels that are far
from isotropic, as yours are, the size would be skewed anyway.

In your case, were afni_proc.py to handle it, the voxels would
be resampled in standard space at 1.75 mm^3, which is where we
would suggest running 3dFWHMx and 3dClustSim. Resampling after
first level analysis adds additional blur to the data, making
the original blur estimates too small.

Oversampling the data before running 3dClustSim is cheating a
bit on the cluster sizes, too. Running it at 1 mm^3 would
lead to smaller volumes than would be seen with larger voxels
(unless the blur is pretty big).

Are you clustering in orig space? If so, why? Also, why go
to 1x1x1 at the end, and not something closer to the original

  • rick

Hi Rick, thanks for your response.

Te long story short of it is that we did all of our data analysis and write up on data using techniques from before that Ecklund paper came out - basically the whole processing pipeline and subsequent writeup was done 4(ish) years ago. We’re attempting to re-edit the manuscript for publication now and are re-doing the cluster thresholding to make sure our report is accurate. It was mainly a resting state project with an acquisition scan (which is what we’re currently analysing) so because of the RS component basically every analysis step was done in orig space up until the ANOVAs at the end. So basically we recently got the .errts and the acf parameters from data in orig space and ran 3dClustSim with those files/info. We have original stats files (from 4 years ago) that we wrote about and are basically just trying to ‘update’ with more conservative cluster information.