On Friday, January 06, 2012 15:53:59 you wrote:
> Dear Benjamin,
>
> I did what you suggested by choosing the images from " reg_standard" folder
> of each subject (from my first completed analysis) and disabling all of
> "Pre-Stats" icons. However, since Melodic doesn't let us skip from
> "Registration state" I just chose 3DOF (for increasing speed) in the
> registration of Atlas . But the results are strange.
Can't you just uncheck all the boxes under Registration?
> Although it is reasonable to have a difference between each time you run
> ICA analysis, the difference between when I do the whole process (I have
> done that twice) and when I use the registered_filtered_images and skip
> the pre-processing (I have done that 3 times) is too much. For instance,
> in the first situation I have 48 components in both tries but for second
> situation there are 68 component for all of 3 times.
>
> Do you know what the problem is?
It looks like you're doing ICA on two datasets. Set "A" is some dataset with
unspecified preprocessing being done to it. Set "B" is your
registered_filtered_images, sans preprocessing. Both times you run ICA on Set A
you get the same number of components. All three times you run ICA on Set B
you get the same number of components. So it seems that melodic is being
remarkably consistent when you run ICA on the same data set.
Your complaint, if I understand you correctly, is that you get a different
number of components when running ICA on Set A and Set B. This is hardly a
problem, though. In fact, it's what we expect from running melodic on two
different datasets. It looks like preprocessing is reducing the dimensionality
of Set A compared to Set B.
FYI, on the command line you can override automatic dimensionality estimation
and force melodic to generate the desired number of components with the "--
dim" option.
|