[Eeglablist] SIFT for group analysis

Makoto Miyakoshi mmiyakoshi at ucsd.edu
Tue May 14 07:32:29 PDT 2024


Hi Bar,

My impression is that there is no EEGLAB's official solution for the
group-level analysis of SIFT data.
The official SIFT package comes with an old group-level analysis solution,
but it has serious limitations and was not usable for me at least. That
motivated me to develop my own.

In my opinion, the difficulty in the group-level SIFT analysis does not
necessarily reside in the statistics of connectivity measures per se. If
that were the case, using Fieldtrip functions would be sufficient to take
care of mass univariate analysis with some correction, like the current
EEGLAB STUDY does (I don't know how it works though, I have never tried it
myself). The real problem in the group-level SIFT is post-ICA inter-subject
inconsistency after all! The simple k-means clustering, or whatever
similar, does not work due to differences in sizes of the connectivity
matrices. For example, if there are 8, 12, 9, 15 brain ICs in 4 subjects,
the sizes of their connectivity matrices are 8x8, 12x12, 9x9, and 15x15.
How can you 'average' these things? If not averaging, what do you do?
That's the core problem on the calculation side.

groupICA is based on Nima Bigdely-Shamlo's probabilistic IC clustering
approach, originally called the Network Projection, which was only
'published' in his PhD dissertation. I took his framework, removed some
part of code that was related to the 'double dipping' problem, and added a
mass-univariate analysis framework to perform so called 'cluster-level
analysis' in the fMRI research. The official name is weak family-wise error
rate (wFWER) control. I used permutation test there.

If I have a chance to re-do the code, I would like to update many details.
I'm kind of waiting for a chance to do this project with someone younger
with an engineering background.

I developed groupSIFT for non-engineers. There are 8 publications already
using groupSIFT since 2024. I've supported all of these papers, but in most
of the cases they did not need my help. If you want to ask
their impressions about using gruopSIFT, I can give you their email address.

If you are interested in using groupSIFT with my support to write a paper,
let me know. I'd be happy to work with you.

Makoto

On Tue, May 14, 2024 at 8:22 AM Bar Lehmann via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> I would like to better understand how SIFT should or should not be used
> with group analysis. I see that SIFT has a group analysis button, yet in
> the SIFT wiki there is a reference to Makoto's GroupSIFT plugin, thus
> seeming to suggest that SIFT is only partially set up for group analysis,
> and GroupSIFT is needed for fully doing group analysis. Can someone please
> clarify the limitation of SIFT group analysis option, and how GroupSIFT
> adds to the possibilities for group analysis?
>
> Thanks very much for any insight you can provide,
> Bar
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
>


More information about the eeglablist mailing list