Hi Varina,

I'm afraid this may not be right... For fMRI data from just 1 subject, permutation isn't as straightforward because of temporal correlation (that persists on the time series after ICA). Moreover, the call to randomise you suggest will test where the weights in the spatial maps across all components are on average different than zero. This doesn't seem to tell interesting things...

If you'd like a probability assigned to the spatial maps of each component, use the mixture modelling that is available in Melodic. If you have many subjects and would like to see how well each fits a group map, perhaps you could consider the dual regression, where the inference will be based on permutations (via randomise), with including correction for multiple testing.

Hope this helps!

All the best,

Anderson


On Mon, Apr 23, 2012 at 2:16 PM, Varina Wolf <[log in to unmask]> wrote:
Good morning,

I would like to check to make sure what I understand about the basics of correcting for mulitiple comparisons is true.

Let say I have one subject with no-task resting state data analysed by Melodic with IC Map thresholded at 0.99.  Then:
- Several of the IC Maps are found to be noise
- Those remaining should go on to correction for multiple comparisons
- If I use randomise the command for the single subject would be filled in as such:

randomise -i <input> -o <output> -d <design.mat> -t <design.con> [options]
randomise melodic_IC.nii.gz  -o melodic_IC_randomiseoutput -1 -1 -T

Would it be correct to say that this would:
1.  performThreshold-Free Cluster Enhancement of the threshzstat images of the melodic derived independant components?
2. Is TFCE the best choice for multiple comparison correction in this case, or is there a more suitable tool, such as FDR?
2. Should I first try to remove the components from melodic_IC that are noise prior to performing this step?
3. If this is correct, how to I generate the corrected individual IC maps from the single output image of the above step?
4. Is there a tool to optimize the orginal choice of IC Map threshold  in Melodic?

Most gratefully,
Varina