I donšt know what --migpN defaults to as far as number of PCA components
goes, but my guess is that a reduction in unstructured noise from the MIGP
leads the automatic dimensionality estimation to choose more components.
If you manually set the dimensionality to be the same, I would expect you
would get relatively similar ICA outputs.
Peace,
Matt.
On 8/9/15, 12:44 AM, "FSL - FMRIB's Software Library on behalf of Parnesh
Raniga" <[log in to unmask] on behalf of [log in to unmask]> wrote:
>The command line is below.
>
>/usr/share/fsl/5.0/bin/melodic -v -i .filelist
>--bgimage=/usr/share/fsl/data/standard/MNI152_T1_2mm_brain.nii.gz
>--bgthreshold=5.000000 -m
>/usr/share/fsl/data/standard/MNI152_T1_2mm_brain_mask.nii.gz --migp
>--nobet -o melodic.gica --Ostats --report --sep_vn --tr=0.754000
>
>vs
>
>/usr/share/fsl/5.0/bin/melodic -v -i .filelist
>--bgimage=/usr/share/fsl/data/standard/MNI152_T1_2mm_brain.nii.gz
>--bgthreshold=5.000000 -m
>/usr/share/fsl/data/standard/MNI152_T1_2mm_brain_mask.nii.gz --nobet -o
>melodic.gica --Ostats --report --tr=0.754000 -a concat
>
>
>I modified the std_pca function to use BLAS library. Unfortunately there
>is overhead of copying data from the different formats but no need to
>pipe data in and out. Note the modification was not used to perform the
>tests above. That is straight fsl.
|