log transform option

[Hwamee Oh]

Hi Chris,

I have a question regarding log transformation option on the target
images during step2.
Based on my short reading on SSMPCA, all target images may undergo
log transformation prior to subsequent analysis procedures.
Since there is an option for this, instead of being a standard
procedure, I'd like to know when to apply log transformation.

For an initial analysis, I'm interested in running PCA with subsequent
fit of a behavioral variable.

Thanks!
Hwamee

[Christian Habeck]

Hi Hwamee,

it's pretty simple, the original SSM dealt with PET images and assumed multiplicative variability, thus a log transform achieves additive variability in the transformed images and the normalization chosen removes global scaling factors. For fMRI this does not work for the simple fact that often subject images are parametric maps coming out of a GLM time series model and can have negative values, so the log transofrm will give imaginary numbers which is no good.

If you are dealing with fMRI images that are parametric maps and have negative numbers in them, leave out the log transform. Any other modality with positive numbers should work. Whether it's the right thing to do (i.e., the variability can be assumed to be multiplicative) for non-PET data is not clear though, maybe try it with and without log transform and see what you get.

Chris

--

You received this message because you are subscribed to the Google Groups "GCVA" group.
To post to this group, send email to gc...@googlegroups.com.
To unsubscribe from this group, send email to gcva+uns...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/gcva?hl=en.

[Hwamee Oh]

Thanks, Chris, for your reply. Since I'm currently dealing with PET
data, your explanation really helped. Though, I'll keep in mind when
dealing with fMRI data.
Thanks!

> > gcva+uns...@googlegroups.com <gcva%2Bunsu...@googlegroups.com>.