I want to use 3dClustSim to perform multiple comparisons correction. Before doing so, I need to use 3dFWHMx to approximate the ACF. However, I’m not sure what kind of residual file I should pick up as input file. It will be greatly appreciated if you could give me some advice.
I used lme4 package in R to perform a hierarchical linear modeling for each voxel ( the sample was a group of subjects other than time series). I wonder if I need to use the square residuals (I assumed that sum of square rediduals could be used as a group statistic for residual file) as input for 3dFWHMx or I need to extract the residual for every data point (a subject) and save the complete set of residuals for each voxel, feeding a four-dimensional dataset (3d space for spatial coordinates and the fourth dimension indicating different data point (subject) ) to 3dFWHMx?
I am not an R-afficionado, so I don’t quite understand those output options; perhaps someone else will chime in about that.
However, on the use of 3dFWHMx: the program is used to estimate the average spatial extent of structure of time series, and in particular we use it on the residual time series (to judge how much spatial structure exists in the noise; when we perform clustering, we are trying to estimate clusters that are bigger than what would be expected to be given by spatial structure in the data). So, the input to 3dFWHMx should be a full data set of residuals (4D, AKA 3D+time dataset), probably with a mask of the region of interest in which you are looking, like a brain mask.
If you are using this for group analysis, then in general you would
process each subject and get a 4D residuals (“errts”, for “error time series” in AFNI-speak) dset for each, presumably all in the same standard space
generate a group mask of interest in standard space (probably based on overlap of all individual subject masks)
calculate ACF parameters using 3dFWHMx within each subject’s errts dset within that group level mask
average each of ACF parameters across group-- the estimates are very likely quite similar across a group
… and then you can use those three group-average ACF parameters in 3dClustSim.
Does that map on to outputs you have from lme4?
–pt
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.