Hi Anderson,
Thank you for the response.
Kindly, I have the following two follow up questions.
I run voxel-wise analysis using (randomise 5000 permutations and TFCE) to study the difference between two groups in PET images. I threshold the statistical maps at 0.95 then I used the remnant significant of the statistical map as a posthoc mask to calculate FA values for every subject.
I am interested in whole brain voxel-wise correlation analysis between FA values in this posthoc mask and the PET signal. In other words a correlation between FA in this posthoc mask and PET signal within each voxel in the brain.
I used the command randomise (n=5000 and TFCE)
My questions are:
1. In this analysis do I need to correct for multiple comparison like any analysis between two groups? Is it enough to use the non-corrected maps if the results support the hypothesis? Or I need the final output images in randomise must be corrected.
2. Can the voxel-wise correlation analysis in randomise accept covariates. For example, if I want to add to the design, age, gender and other covariates. Is this procedure still correct like adding the covariates in design to study the difference between two groups? How many caovariates (as maximum) do you recommend to be included in randomise.
3. If I feed to GLM demeaned data. Do I need to use the flag -D in GLM?
Thanks!
Jon
Hi Jon,
Yes, it is.
All the best,
Anderson
On 26 October 2016 at 10:52, John anderson <[log in to unmask]> wrote:
Dear FSL experts
This link in FSL wiki http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Single-Group_Average_with_Additional_Covariate
explain how to create design matrix for a voxel wise correlation analysis between clinical scale ( please correct me if the terms used to describe the analysis are not accurate ) and the whole brain.
If we replace the clinical scale in the model with FA values derived from an ROI. In want to know if the concept of the analysis still valid?
Thanks!
Jon
|