You can take the filtered_func_data (or the var_filtered_func_data)
from a gfeat/cope directory, which has the measures for each
individual subject concatenated, and convert these to a zscore value
across subjects (by dividing calculating a standard deviation map and
dividing). This gives a multisubject map at each voxel which can be
easily paged through in fslview, and thresholded at any zscore you
choose (eg, Z>3 implies "outliers" where that subjects cope or varcope
value was 3 standard deviations above the mean. This method reveals
subjects and brain regions that may be outliers/artifacts.
It is harder to decide what to do with this information -ie, how do
you decide whether to exclude a subject based on this information.
One question I have is whether it is worthwhile to try and exclude
subjects who are "outliers", since I believe FSL 2nd level analysis
using FLAME has the effect of "deweighting" the influence of those
subjects who have high levels of noise, in a sophisticated voxelwise
mannter. I would be interested to hear from FSL experts whether
excluding "noisy" subjects increases or decreases the statistical
significance of FLAME analyses.
Dan
On 8/22/07, Christopher Nuņez <[log in to unmask]> wrote:
> What procedures do people use for searching for outliers in higher level analyses? Thank you.
>
> -------------------------------------------------------------------
> S. Christopher Nuņez, PhD
> Postdoctoral Fellow
> Laboratory of Neuro Imaging
> UCLA Dept of Neurology - #1 in NIH Funding
> David Geffen School of Medicine
> Neuroscience Research Building 1, Suite 225
> 635 Charles Young Drive South
> Los Angeles, CA 90095-7334
> (310) 267-5555
> -------------------------------------------------------------------
>
|