[Eeglablist] What to do with more than one IC per subject in a cluster

[Makoto Miyakoshi]

Dear Salim,

Sorry for slow response.

> for PCA-clustering I removed the noisy ICs and then I preselected ICs
which are located in the Sensorimotor cortex, after that I created a STUDY
and used a PCA for clustering and ensured each cluster has at most 2
components from the same subject to avoid biasing.

Interesting. As long as you choose them by anatomical locations, I believe
you are fine.

> and Whether these results are comparable to those of MPT?

First you have to check how much overlap between MPT-selected ICs and your
selections.

MPT is, to my understanding, a 'similarity filter'. Therefore, I expect
that MPT-selected ICs could be more consistent in the selected measure than
k-mean clustering (well, you can still tweak the parameters in k-mean
clustering so that the result is maximally similar to that of MPT).

Sorry it could be confusing that there are two solutions for the
group-level analysis.
By the way, before running MPT I recommend you clean the ICs in the
following way (did I tell this to you before? If so excuse me)
http://sccn.ucsd.edu/wiki/Backproject_clustered_ICs

Makoto



On Thu, Feb 26, 2015 at 2:29 AM, Salim Al-wasity <salim_alwasity at yahoo.com>
wrote:

> Dears
> As I understood from the following discussion (
> http://sccn.ucsd.edu/pipermail/eeglablist/2013/006353.html), its not
> recommended to pre-select the components for a PCA-clustering approach?
>
> I am comparing between the ERSP results of using MPT approach and
> PCA-clustering approach. In the first STUDY, I used the MPT to cluster the
> ICs (after removing the noisy ones). and for PCA-clustering I removed the
> noisy ICs and then I preselected ICs which are located in the Sensorimotor
> cortex, after that I created a STUDY and used a PCA for clustering and
> ensured each cluster has at most 2 components from the same subject to
> avoid biasing.
> Is this technique of pre-selecting ICs based on ROI (Region Of Interest)
> prior to PCA-clustering is a good way to avoid biasing?  and Whether these
> results are comparable to those of MPT?
>
> Sincerely
> Salim
>
>
> _______________________________________________
> 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
>



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20150319/c20faa4a/attachment.html>