Hi
The log file refers to the proportion of the variance that a 26 dim decomposition retains from the concatenated data. In case of a single ICA the data itself has 100% variance so the two reported numbers are identical. In the case of a concat-ICA run the data only represents some of the variability due to the first subject-wise data reduction. This in the case of multi-subject ICA the 'eigenvalue_percent' values are lower as they get multiplied by the amount of variance reduced by the first set of PCAs.
hth
Christian
On 22 Jan 2010, at 11:11, Mikaël Naveau wrote:
> Dear Melodic users,
>
> I used group-ICA with concat approach on 10 fMRI acquisition of 240 TR.
> I would like to report percent of explained variance extracted by the
> second data reduction stage.
>
> I found two different types of information:
> - The “eigenvalues_percent” file which contains 10 * 26 values (26 =
> model order estimated by Laplace approximation). I suppose it refers to
> the second reduction step applied on concatenated first reduced data
> ([10*26]*voxels). This file indicates a proportion of explained variance
> equal to 0.2635 (corresponding to column 26).
> - The log file that indicates 66.6968 % of explained variance…
>
> With a single subject ICA, this two values match.
> Could anyone explain what I missed? Is one value is % of explained
> variance extracted by the first reduction step?
>
> Best regards,
>
> Mikael
>
> PS : here the command line used to compute gICA :
> melodic -i input_files -o output_dir --nomask --approach=concat --report
> --tr=2.000 --Ostats
> (data sets are already masked)
>
> <naveau.vcf>
|