Hi,
There is a misconception here: ICA (unlike PCA) does not give a partitioning of variance. That is, on average (acorss space) mutiple components can potentially express the saem bit of variance. Therefor their total will _not_ add to 100%. What this means in your case is that you seem to have strong correlations between components. These numbers are only indicative relative to strong vs weak effects (i.e. can be used for ranking) but not much else... so no, not a bug, varainces don;t need to add to 100% or less.
Wrt your second question: The difference between TICA and CONCAT is strong for taks related effects only. If most components are not task related then the overall difference between TICA and CONCAT output can be small.
hth
Christian
On 20 May 2013, at 12:01, David Soto <[log in to unmask]> wrote:
> hi - sorry about the persistent inquiry
> this follows from my previous post 052819
> for which I got no reply. Here it is the latest
> bug found on this saga
>
> TICA( FSL5.0.4) run on LionOS
> gives ICs 'explaining' very high
> proportion of the variance i.e.
> over 30% of the variance, for each the first 10 ICs (out of the 75 found)
>
> The same result is not found under FSL5.0.4 run on Ubuntu
>
> while CONCAT run on Ubuntu and on Lion with FSL5.0.4 on the same images gives exactly the same output
>
> This leads me to conclude that TICA falters under Lion.
>
> I am running MELODIC through the commnad line. Preprocessing done
> independently and output resampled to 4 mm and then registered to MNI
> melodic -i /Users/dsoto/Documents/fMRI/wmbias_rs_task/taskdata4mm/taskicalist.txt -o /Users/dsoto/Documents/fMRI/wmbias_rs_task/task_melodictica4mm -v --nobet --bgthreshold=10 --tr=2.5099999905 --report --mmthresh=0.5 -a tica --Tdes=/Users/dsoto/Dropbox/imaging/Exps/uWM_WMgui/wmgui/design.mat --Tcon=/Users/dsoto/Dropbox/imaging/Exps/uWM_WMgui/wmgui/design.con
>
>
> Just thought you may be interested to know
>
> david
|