Dear AFNI experts,
I am analyzing a data set for which I want to estimate the probability of false positive clusters by means of running 3dClustSim. I am aware that 3dClustSim has been updated so that instead of assuming the spatial auto-correlation function (ACF) is given by a Gaussian-shaped function, it simulates the noise volume assuming the ACF is given by a mixed-model for which the parameters (a, b and c) can be calculated by the program 3dFWHMx.
The 3dFWHMx help page indicates that a good input to the program would be the errts-files, which gives the residuals left over after the GLM fitted signal model is subtracted out from each voxel’s time series. Therefore, I used the errts-files that were output by 3dREMLfit. I ran the 3dFWHMx program with the following parameters for each subject in our experiment.
3dFWHMx –mask mask+tlrc –ACF –input errts.subject1+tlrc
Subsequently, I calculated the median of the a, b, c values that were output by 3dFWHMx (a = 0.94, b = 1.78, c = 6.85). The “old style” FWHM parameters that were output seem rather low though, i.e., around 3 (whereas we smoothed the data with a Gaussian kernel of 9mm). Then I ran 3dClustSim with these values that I obtained.
3dClustSim –mask mask+tlrc –acf 0.94 1.78 6.85 -iter 10000
The output of 3dClustSim, however, seems a bit odd. NN2 2-sided, at pthr 0.01 and alpha 0.05, gives a cluster size of only 8 voxels which seems way too lenient. It seems like I am following all the steps as indicated on the help pages of 3dFWHMx and 3dClustSim, so I am not sure what causes these results.
Any feedback on whether we are running the newest version of 3dClustSim and the 3dFWHMx program correctly would be greatly appreciated.
Best,
Wessel