Hi Jim, thanks for your questions;
> 1. In generating the group data, does feat use the raw (not thresholded)
> data from the 1st level .feat directories? Or does the thresholding that
> was used in the 1st level analysis affect the group comparison?
the group stats generally take as input unthresholded data from the first
level, namely the cope (contrast of parameter estimate) images and the
varcope (variances) of these (I say generally because if you used IRVA at
first level then it's the unthresholded zstats that get used). The cope
and varcope images are transformed into standard space before being
combined to do fixed and random effects group stats.
> 2. If I wnat to re-run a group comparison with a different threshold, do I
> simply select "contrast, thresholding, rendereing only", point to
> the .gfeat directory, change the threshold, and re-run? Or do I have to re-
> run the entire comparison with my new desired threshold, pointing to each
> of the 1st level .feat directories, and calmly sit through the registration
> again?
aha - not quite either option :)
once registration has been run for the first level FEAT sessions then
these FEAT directories contain the registration information (ie the
transformation files) and you do not need to redo-registration when you
run group stats. Thus if you are redoing group FEAT only you can turn off
the registration button.
but - yes, to redo different thresholding on group FEAT you do do the
group FEAT part from scratch - ie re-transforming first level stats images
into standard space and then combining, but because you don't need to
refind the registration transformations (ie can turn off the registration
button) this takes very little time. the next version will make this a
little neater though.
note - each time you re-run group feat you don't have to select all the
first level feat directories again - you can just take the last .gfeat
directory that you created and load in its design.fsf into FEAT, make sure
that registration is turned off, change any details (eg thresholding) and
press GO.
> Also, does anyone have a suggestion on z threshold for a simply motor task
> (finger to thumb)? When I use the default of 2.3 I get 1 massive activated
> cluster and I would like to see a little more definition between motor
> planning/SMA and motor cortex activation.
indeed - visual/motor goes way above 2.3 typically, however the choice is
statistically arbitrary, so its up to you. You could try 4.6 and see if
you like the blobs better! ;-) note that if you are doing cluster-based
thresholding then the p=0.01 cluster-threshold stays valid even if you
vary the Z threshold.
Thanks, Steve.
|