Dear Stefanie,
For any of the areas that survive the multiple-comparison-based thresholding then you can reject the null hypothesis at these locations (e.g. anterior caudate nucleus), and therefore declare a significant correlation between changes in shape/geometry in these locations and your performance measure.
Be aware that the values returned by randomise are actually 1-p values, so voxels over 0.95 in the corrp image represent locations where there is a significant effect. In this way is it exactly the same as what you would get out of FSL-VBM.
I hope this helps.
All the best,
Mark
> On 19 Nov 2014, at 02:05, Stefanie K <[log in to unmask]> wrote:
>
> Hi All,
> I performed a shape analysis using FSL_First to correlate performance in a navigation task with shape differences in the caudate. I followed the instructions listed in the wiki for the design matrix:
>
> "Design matrix with one column (EV) where each row contains the value of the correlating measure for each subject in the same order as the subject's bvars (in the concatenated bvars file)".
>
> This worked well, but now I struggle with the interpreation of those results. For example after correction for multiple comparison I reveal a difference in shape in the anterior caudate nucleus.
> But how would I interprete that?
> The wiki gives the following explanation:
> "Interpretting the results of a correlation: the values in the p-value images will give the probability of a zero correlation (the null hypothesis) at each vertex."
> My null hypothesis would be that performance would not correlate with shape of the caudate
> Let's assume I would get a p-value of 0.05, would that mean that in the anterior caudate I can reject the null hypothesis?
>
> I would be happy for any suggestions to that topic. I read several papers that used a correlation with VBM but unfortunately non that did the same for shape analysis.
> Many Thanks
> Stefanie
|