[pymvpa] spatial normalization for MVPA - is it good or bad?

[Vadim Axel]

Thanks for the insight!
I personally use always normalized and I do not think that this should
matter too much.
I think given that normalization introduces some smoothing, it may probably
even increase predictions - as Hans Op De Beeck showed that smoothing might
be helpful for prediction rate.


On Mon, May 5, 2014 at 10:21 PM, J.A. Etzel <jetzel at artsci.wustl.edu> wrote:

> I've done both: MVPA in subject space (no spatial normalization), and
> after normalization to an atlas. Even for searchlight analysis both are
> possible: run the searchlights in subject space and normalize for the
> group-level analysis, or normalize first.
>
> Ideally, it wouldn't matter at which stage of analysis spatial
> normalization was done. And it often doesn't seem to matter much in
> practice, particularly when working with fairly large ROIs.
>
> Sometimes working in subject space, with individually-defined ROIs seems
> most sensible, for example when a small, definable anatomic structure is of
> interest (such as the amygdala or hippocampus). Other times you need to
> work with spatially-normalized images, such as when doing a
> multiple-subject analysis.
>
> All spatial normalization algorithms are imperfect, and introduce
> additional blurring, distortions, and dependencies into the images. We
> accept those distortions for mass-univariate GLM analyses; should we accept
> them for MVPA as well? I usually do, hoping that they will be relatively
> minor compared to the distortions already added to the data (e.g. movement
> correction, slice-timing, scanner drift), and in BOLD itself.
>
> Do others have different experiences? Ever seen a dataset in which spatial
> normalization made a meaningful difference?
>
> Jo
>
>
>
>
> On 5/3/2014 9:31 AM, Vadim Axel wrote:
>
>> Hi,
>>
>> If one uses functional localizer or selects anatomical ROIs individually
>> he might run MVPA in a native space (without spatial normalization to
>> template of the data). Have you encountered any paper that compared
>> normalized vs. non-normlaized prediction rates? Or probably personal
>> observation? Obviously, for search-light one needs to normalize.
>>
>> Thanks,
>> Vadim
>>
>>
>>
>> _______________________________________________
>> Pkg-ExpPsy-PyMVPA mailing list
>> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
>> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-exppsy-pymvpa
>>
>>
> --
> Joset A. Etzel, Ph.D.
> Research Analyst
> Cognitive Control & Psychopathology Lab
> Washington University in St. Louis
> http://mvpa.blogspot.com/
>
> _______________________________________________
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-exppsy-pymvpa
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20140505/1d14364a/attachment.html>