[color=#000000]Hi Shady,[/color]
[color=#000000]That is strange, you should be in fact getting exactly (or almost exactly) the same results. Did you check that your seed-to-voxel analyses used multivariate regression measures (not the default bivariate correlation)? For the seed-to-voxel results to be equivalent to the GICA subject-level maps you need to use all of those ICA seeds in a single multivariate analyses -i.e. selecting 'multivariate regression'-, rather than using each seed separately in a different univariate analysis -e.g. selecting 'bivariate regression/correlation' (sorry the manual does not go into these details, it just briefly mentions "using these ROIs as seeds in multivariate analyses")[/color]
Hope this helps
Alfonso
[i]Originally posted by Shady El Damaty:[/i][quote]The manual states (page 34 last paragraph) that the resulting ROI time series and connectivity maps obtained from importing the ICA components as ROIs in the Setup tab more closely represent the properties of the networks identified in the ICA procedure. The manual also states that the resulting seed-to-voxel maps resulting from the imported ROIs should be exactly the same as the individual subject-level spatial maps estimated in the ICA procedure.
If this is true, then do I get different second-level results for contrasts performed under ICA networks -> Spatial Components vs using a Seed to Voxel analysis with ROIs imported as described in method #3 on page 34 of the manual?
This inconsistency makes it a bit difficult to collapse across results from masked ICA results since you would have to reimport them into CONN as ROIs before you can look at the main effect of different ICA maps (i.e. components in left and right hemispheres obtained from a masked ICA). Any advice on how to proceed with this issue?[/quote]
[color=#000000]That is strange, you should be in fact getting exactly (or almost exactly) the same results. Did you check that your seed-to-voxel analyses used multivariate regression measures (not the default bivariate correlation)? For the seed-to-voxel results to be equivalent to the GICA subject-level maps you need to use all of those ICA seeds in a single multivariate analyses -i.e. selecting 'multivariate regression'-, rather than using each seed separately in a different univariate analysis -e.g. selecting 'bivariate regression/correlation' (sorry the manual does not go into these details, it just briefly mentions "using these ROIs as seeds in multivariate analyses")[/color]
Hope this helps
Alfonso
[i]Originally posted by Shady El Damaty:[/i][quote]The manual states (page 34 last paragraph) that the resulting ROI time series and connectivity maps obtained from importing the ICA components as ROIs in the Setup tab more closely represent the properties of the networks identified in the ICA procedure. The manual also states that the resulting seed-to-voxel maps resulting from the imported ROIs should be exactly the same as the individual subject-level spatial maps estimated in the ICA procedure.
If this is true, then do I get different second-level results for contrasts performed under ICA networks -> Spatial Components vs using a Seed to Voxel analysis with ROIs imported as described in method #3 on page 34 of the manual?
This inconsistency makes it a bit difficult to collapse across results from masked ICA results since you would have to reimport them into CONN as ROIs before you can look at the main effect of different ICA maps (i.e. components in left and right hemispheres obtained from a masked ICA). Any advice on how to proceed with this issue?[/quote]