Dr. Nichols,
Thank you for responding and the reference. Would it be reasonable
in your opinion to calculate an FDR threshold within an ROI? Or would
another correction for muliple comparisons method be more appropriate?
thank you,
Jason.
On Mon, 14 Feb 2005, Thomas E Nichols wrote:
> Hi Jason,
>
> > Hello. I am using FSL scripts to extract information from
> > statistical map images in certain ROIs. I am doing this by placing a
> > threshold on the map and then finding information like: max z-score, mean
> > z-score, number of voxels etc. However, by chosing a threshold this way
> > (for example z=2.3) I am essentially using an uncorrected threshold. I
> > know that if I want to do a Bonferroni correction in the ROI I can divide
> > alpha by the number of voxels in the ROI.
>
> Yup.
>
> > But what if I want to do an FDR correction at alpha=0.5? Can I find
> > the 95th percentile of all the z values in the ROI and use that as
> > the threshold for the ROI? Would this then be considered FDR
> > corrected at the ROI level?
>
> No, it's not that simple. You have to compare the ordered P-values
> p(i) to i/V (where V=number of voxels). There's a script FDR.m here
>
> http://www.sph.umich.edu/~nichols/FDR
>
> that will give you the threshold.
>
> Hope this helps!
>
> -Tom
>
>
> -- Thomas Nichols -------------------- Department of Biostatistics
> http://www.sph.umich.edu/~nichols University of Michigan
> [log in to unmask] 1420 Washington Heights
> -------------------------------------- Ann Arbor, MI 48109-2029
>
Jason Steffener
Department of Radiology
New Jersey Medical School
Biomedical Engineering Department
New Jersey Institute of Technology
Newark, NJ USA
(973) 676-1000 x1545
http://web.njit.edu/~jrs3
