Dear Donald, thank you for your answer. As far as I understand a SVC is already implemented in the wfu pickatlas ROI Analysis.
I am still confused about the appropriate p-value that should be reported. Is it the uncorrected cluster level p-value or the uncorrected peak level p-value or the FDR/FWE corrected cluster-level (peak voxel) p-value? Since choosing the right p-values has a high impact on the results I am very concerned about doing the right way.
Thank you for your help,
Silke Matura
Dr. Dipl.-Psych. Silke Matura
Laboratory of Neuroscience
Department of Psychiatry, Psychosomatic Medicine and Psychotherapy
University of Frankfurt
Heinrich-Hoffmann Str. 10
60528 Frankfurt am Main
Tel. 069-6301- 83839
[log in to unmask]
________________________________________
Von: MCLAREN, Donald [[log in to unmask]]
Gesendet: Dienstag, 30. Oktober 2012 16:42
Bis: Matura, Silke
Cc: [log in to unmask]
Betreff: Re: [SPM] ROI Analysis VBM
On Tue, Oct 30, 2012 at 6:13 AM, Matura, Silke <[log in to unmask]> wrote:
> Dear SPM users,
>
> I am currently analysing ROIs in regard to grey matter density using the wfu
> pickatlas in SPM8. As far as I understand the ROI Analysis does not employ a
> voxelwise comparison as in VBM but uses an average value that is generated
> based upon the respecting ROI.
Researchers use the term ROI analysis in two ways that leads to
confusion. The traditional meaning of an ROI analysis is that you
extract the values of your dependent variable from the ROI and analyze
those values. The statistics are done outside of SPM/FSL/AFNI in a
statistical package like SAS/STATA/SPSS. You can correct for the
number of ROIs/tests in a number of ways - Bonferroni, Tukey, FDR,
etc.
The second approach is to limit the voxels that you analyze to a
specific ROI or set of ROIs. This isn't an ROI analysis per se as you
are still analyzing the voxels. If this is what you have doen, then
you should use SVC. When you take this approach, you need to include
all the ROIs at one time, otherwise you will inflate the significance
of each finding. The inference is made on clusters on voxels in the
small volume that you have searched.
Hope this helps to clarify your confusion.
Is that correct? Now I wonder how I should
> deal with the correction for multiple comparisons. FWE does not seem
> appropriate since there are no multiple comparisons in a ROI analysis. I
> would go for a small volume correction and report the respecting p-value (no
> FWE and no FDR correction). Could you give me some advice if that is the
> right approach?
>
> Thank you,
> Silke
>
> Dr. Dipl.-Psych. Silke Matura
>
> Laboratory of Neuroscience
>
> Department of Psychiatry, Psychosomatic Medicine and Psychotherapy
>
> University of Frankfurt
>
>
>
> Heinrich-Hoffmann Str. 10
>
> 60528 Frankfurt am Main
>
>
>
> Tel. 069-6301- 83839
>
> [log in to unmask]
|