[Eeglablist] PSD of resting state data

Fri Jan 25 17:08:54 PST 2019

Dear Katherine,

> 1. Currently I have epoched my data into pre and post and created a study
with 2 conditions per subject. Is it still reasonable to cluster from this
juncture?

Yes. Use STUDY.design to specify the factorial design.

> 2. Should I interpolate channels at the study level?

If you are performing independent component clustering, no. It does not use
channel information any more.

> 3. Finally, because I wanted to maintain the continuity of the continuous
data I did not remove noisy epochs (time periods - I don't really have
epochs) in hopes that the amount of noise would be similar in the pre and
post conditions (just wanted to know if you had any thoughts on that).

What's your question here? Maybe you are not sure if it is acceptable or
not?
It depends on how clean your data are. If you can provide additional
statistics or something that guarantees you don't have terrible outliers in
your data analyzed, it is fine. For example, use trimoutlier() plugin for
continuous data. It gives you at-a-glance summary for all channels and time
points.
https://sccn.ucsd.edu/wiki/TrimOutlier

Makoto

On Wed, Jan 23, 2019 at 6:32 AM Katherine Eskine <
eskine_katherine at wheatoncollege.edu> wrote:

> Hello,
>
> Thanks so much for writing such a fantastic plug in! I am excited to give
> that a try. I have a few quick clarification questions.
> 1. Currently I have epoched my data into pre and post and created a study
> with 2 conditions per subject. Is it still reasonable to cluster from this
> juncture?
> 2. Should I interpolate channels at the study level?
> 3. Finally, because I wanted to maintain the continuity of the continuous
> data I did not remove noisy epochs (time periods - I don't really have
> epochs) in hopes that the amount of noise would be similar in the pre and
> post conditions (just wanted to know if you had any thoughts on that).
>
> I am eternally grateful for this help. I've been working on these problems
> for an embarrassingly long time, and can finally see the light at the end
> of the tunnel. THANK YOU!
>
> Best,
>
> Kate
>
>
> Katherine E. Eskine
> Assistant Professor of Psychology
> Mars SC 1136 / t. 508-286-3636
> Wheaton College
>
>
>
> On Tue, Jan 22, 2019 at 4:59 PM Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
> wrote:
>
>> Dear Kate,
>>
>> > Can I access the PSD from the study GUI or is something that needs a
>> script?
>>
>> In the case of EEGLAB 14 or before...
>> After clustering the ICs in STUDY, 'Plot spectra' for all the clusters
>> from 'Edit and plot'. After plotting all the cluster spectra, close the
>> 'Edit and plot' window BY PRESSING OK. This creates
>> 'STUDY.cluster(3).icaspectra' or something like that under each cluster
>> which you can grab for your own statistics.
>>
>> > I actually have the same question about the ICA, how can I see if there
>> are differences? Where can I get the data to use in SPSS?
>>
>> Differences across conditions? Then you have to 'epoch' the trials
>> separately for each condition. This is needed regardless of whether you
>> perform sensor-level or ICA-decomposed comparison. If you are interested in
>> ICA-clustering STUDY, you may want to check out this plugin I wrote. This
>> supports data export to SPSS.
>>
>> https://sccn.ucsd.edu/wiki/Std_erpStudio
>>
>> Makoto
>>
>>
>> On Tue, Jan 22, 2019 at 11:57 AM Katherine Eskine <
>> eskine_katherine at wheatoncollege.edu> wrote:
>>
>>> First, thanks so much for your thoughtful responses. I am going to do a
>>> PSD and an ICA comparison.
>>>
>>> I have a few follow up questions I am hopeful you can help me with.
>>>
>>> Can I access the PSD from the study GUI or is something that needs a
>>> script? To do a repeated measures analysis where can I download the data
>>> from?
>>>
>>>  I actually have the same question about the ICA, how can I see if there
>>> are differences? Where can I get the data to use in SPSS?
>>>
>>> Thanks so much for your guidance, much obliged.
>>>
>>> Best,
>>>
>>> Kate
>>>
>>> On Jan 17, 2019, at 9:25 PM, Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
>>> wrote:
>>>
>>> Dear Katherine,
>>>
>>> > How can I best determine if there are differences between the
>>> distributions of frequencies bans (alpha, beta, etc.)?
>>>
>>> If you know how to code, of course you can take temporal dynamics of the
>>> alpha, theta, etc band power into consideration in an creative way!
>>> However, even just a standard power spectral density (PSD) would be
>>> sufficient to draw a conclusion. It is simpler too.
>>>
>>> > My original thinking was to find the average alpha, beta, theta, delta
>>> and gamma for the pre and the post then submit them to a repeated measures
>>> analysis. However, I am wondering if an analysis using the components might
>>> provide more information?
>>>
>>> Yes, ICA results come with their corresponding scalp topography (which
>>> is directly generated by ICA as inverse weight matrix, EEG.icawinv), on
>>> which you can perform dipole fitting using EEGLAB's infinitely close to
>>> official plugin Dipfit.
>>>
>>> > Can I identify significant differences in ICA's between the pre and
>>> the postconditions and then look at the dipoles for the brain source?
>>>
>>> You can at least compare pre and post conditions.
>>>
>>> > One follow-up question, will I run into problems because the pre and
>>> post conditions have the same ICA components? I assume that looking for
>>> different power of each component will get at any before and after
>>> differences, but does the structure violate vector parameters?
>>>
>>> If subjects did not take the cap between pre and post conditions, then
>>> you can run just one ICA on the two conditions. This is much more
>>> straightforward. If they did take the cap, then you MUST run ICA separately
>>> on pre and post, then you have to use IC clustering, hoping that a given
>>> cluster has both pre and post of the same subject (which is not
>>> guaranteed...) So the interpretation will be probabilistic.
>>>
>>> Makoto
>>>
>>>
>>> On Thu, Jan 17, 2019 at 11:24 AM Katherine Eskine <
>>> eskine_katherine at wheatoncollege.edu> wrote:
>>>
>>>> Dear all,
>>>>
>>>> I have been working through a data set and would deeply appreciate some
>>>> advice. I have 6 minutes of resting state data before and after exposure,
>>>> all recorded in the same session. I would like to see if there are
>>>> significant differences in frequency bands before as compared to after the
>>>> exposure.
>>>>
>>>> The EEG during the exposure was removed and then the continuous data
>>>> has been post-processed and submitted to ICA analysis, where heartbeat, eye
>>>> blinks, and other noisy components were removed. Then the data was split
>>>> into the 6 minutes before and after.
>>>>
>>>> How can I best determine if there are differences between the
>>>> distributions of frequencies bans (alpha, beta, etc.)?
>>>> * bandpass filtering & plotting per band
>>>> * average absolute power per band, or
>>>> * time-frequency transform using short-term Fourier transforms or
>>>> wavelets
>>>> * should I epoch the data into 3-second intervals and proceed from
>>>> there?
>>>>
>>>> My original thinking was to find the average alpha, beta, theta, delta
>>>> and gamma for the pre and the post then submit them to a repeated measures
>>>> analysis. However, I am wondering if an analysis using the components might
>>>> provide more information? Can I identify significant differences in ICA's
>>>> between the pre and the postconditions and then look at the dipoles for the
>>>> brain source?
>>>>
>>>> One follow-up question, will I run into problems because the pre and
>>>> post conditions have the same ICA components? I assume that looking for
>>>> different power of each component will get at any before and after
>>>> differences, but does the structure violate vector parameters?
>>>>
>>>> Thanks so much for your help. I have been following the discussion from
>>>> Mohith, but I think my continuous data might be a slightly different case.
>>>>
>>>> Best,
>>>>
>>>> Kate
>>>>
>>>>
>>>> Katherine E. Eskine
>>>> Assistant Professor of Psychology
>>>> Mars SC 1136 / t. 508-286-3636
>>>> Wheaton College
>>>>
>>>> _______________________________________________
>>>> 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
>>>
>>>
>>
>> --
>> Makoto Miyakoshi
>> Swartz Center for Computational Neuroscience
>> Institute for Neural Computation, University of California San Diego
>>
>

-- 
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/20190125/f43523b9/attachment.html>