Hi
> my dr_stage2_ic[#ICA].nii.gz images should contain the subject representation/equivalent of the group networks defined in the groupmelodic.ica/melodic_IC.nii.gz ???.
>
If this is what you used as the maps feeding into DR then yes.
> so for each (cognitive) parameter i'm investigating (looking for it's correlation with my networkS), I should define one or few networks.
> then i look into melodic_IC.nii.gz, visually recognize the networks, note its component number (eg: 0011) and then look at (and report in my paper) the significant voxels of the :
>
> dr_stage3_ic0011_tfce_corrp_tstat2.nii.gz
>
> image (assuming i'm interested in contrast 2)
>
Yes
> in general, if I got it, my significant voxels must always be put in relation to a specific network, and cannot be reported regardless of their RSN and of course i cannot create ONE result map x covariate after mixing different RSN.
> am i now correct?
>
Yes
> but what happens if a i found significant voxel in a component which is not a common network or not a network at all?
> in some way, that voxel, in that 6 minute recording, correlated with my covariate value.
>
It's entirely possible that the initial ICA has split off some part of an 'common network' into a separate component because the underlying signal in that subset of voxels was significantly altered compared to the main network dynamics. You could test this by feeding in melodic_IC maps that - instead of being derived from the full set of data sets available - were derived only from the control population (or even better, from a separate control population). In that case you might find that the effect now becomes directly associated with a 'common network'
hth
Christian
> thanks!
> Alberto
>
>
>
>
>
>
>
>
>
>
|