[Eeglablist] EEG data processing questions

Makoto Miyakoshi mmiyakoshi at ucsd.edu
Wed Sep 2 10:19:37 PDT 2015


Dear Emmanuelle,

I agree with Stephen. I don't recommend 1-sec chunking the data for the
cleaning purpose.

For cleaning the continuous data, try clean_rawdata(). The artifact
subspace reconstruction is a very smart algorithm to correct artifacts.

http://sccn.ucsd.edu/wiki/Plugin_list_process
http://sccn.ucsd.edu/eeglab/plugins/ASR.pdf

This was written by Christian Kothe, which means quality guaranteed!

> P.S. I tried using ICA instead, but with only 8 channels (7 after
re-referencing), it is very hard to do!

What's the problem? Run ICA and remove at least a blink component, which
must be present as the top component unless eyes-closed condition. Blink
component has an ideal characteristics to be identified by ICA. I compared
33ch vs. reduced 3ch EEG data decompositions. ICA found the exactly same
component between the two (I mean ERPimage was identical).

For analysis, you may want to check out NBT by Simon Shlomo-Poil
https://en.wikipedia.org/wiki/Neurophysiological_Biomarker_Toolbox

Makoto



On Wed, Aug 26, 2015 at 7:00 AM, Emmanuelle Renauld <
emmanuelle.renauld.1 at ulaval.ca> wrote:

> Hi Stephen,
>
>
>
> Thanks for your answer.
>
>
>
> I don't know if we actually need to segment the recording. The problem was
> that our data seems to have many huge bugs that completely distorted the
> spectrum results. I was not aware that you can mark artifacts before the
> time-frequency decomposition. Can I do that in a script, or it has to be
> done manually? Is there a specific function?
>
>
>
> A little more info about the windows: We started with this idea after
> reading it in some articles in our field. (ex:
> http://www.ncbi.nlm.nih.gov/pubmed/23641206). In this article, they even
> filter each window separately. But in our method, all filtering steps were
> performed before, so we don't have to worry about the filter artifacts.
>
>
>
> In our analysis, I am not trying to average the signals together, as with
> ERPs. We only want information on the Fourier spectrum.  I thought that we
> could compute it on each window, considering that we are only interested in
> frequencies above 5 Hz anyways. But I had not realized that the spectrums
> would appear so smooth. (Ex, with spectopo, with our 500 Hz sampling rate,
> a 1 second window can be analyzed with a winsize of 2^8 max. 2 seconds
> allows 2^9. Visually, there is a huge difference!).
>
>
>
> That's when I came with the idea of a re-concatenated signal. I am not
> sure how I could assess the quality of an analysis on that whole signal.
> Did I miss some important discussions already done or articles already
> published on this subject?
>
>
>
> Thanks again,
>
>
>
>
>
> Emmanuelle
>
>
>
>
> ------------------------------
> *De :* politzerahless at gmail.com <politzerahless at gmail.com> de la part de
> Stephen Politzer-Ahles <spa268 at nyu.edu>
> *Envoyé :* 26 août 2015 04:26
> *À :* Emmanuelle Renauld
> *Cc :* eeglablist at sccn.ucsd.edu
> *Objet :* Re: [Eeglablist] EEG data processing questions
>
> Hi Emmanuelle,
>
> Windows of 1 second are pretty short, and keep in mind that there will be
> frequency issues near the edges (for example, if you use a filter, then
> even if you don't re-concatenate the epochs, there will be filter artifacts
> introduced at the edges of the epoch). So the usable window of analysis,
> practically speaking, is even shorter than 1 second.
>
> Do you actually need to segment the recording? I work in event-related
> potentials, so I'm not familiar with what is standard for analyzing data
> without events/epochs. But it seems to me like it might be better to just
> run your time-frequency decomposition on the continuous data, and then in
> later analysis (for example, if you want to average across timepoints over
> the entire recording---if there were no stimulus events, then maybe time is
> meaningless anyway) just ignore timepoints that are close to artifacts (you
> can mark artifacts in the continuous data, without actually removing the
> data, before running the time-frequency decomposition).
>
> Best,
> Steve
>
>
>
> Stephen Politzer-Ahles
> New York University, Abu Dhabi
> Neuroscience of Language Lab
> http://www.nyu.edu/projects/politzer-ahles/
>
> On Mon, Aug 24, 2015 at 11:23 PM, Emmanuelle Renauld <
> emmanuelle.renauld.1 at ulaval.ca> wrote:
>
>> Hi everyone
>>
>>
>>
>>
>>
>> I have many trouble processing some data, due to the many bugs in the
>> data. We don't know yet what causes them, maybe the cap is broken or some
>> electrodes work bad, still to see.
>>
>>
>>
>>
>>
>> Anyways, now here is what I do:
>>
>>
>>
>> - rereferencing
>>
>> - filtering (0.5 - 50 Hz)
>>
>> - Cutting the data into windows of 1 seconds. (They are not epochs; we
>> don't have events informations).
>>
>> - Removing the worst windows with some rejection criteria.
>>
>> - Keeping the subjects where < 10% of the windows are removed.
>>
>>
>>
>> We I want to compute the statistics (ex, Fourier spectrum), I read that
>> it was better not to re-concatenate the windows together since it could
>> cause false frequencies. So I compute Fourier on each window and average
>> their spectrum. However, smaller data = less precise fourier spectrum. I am
>> not really satisfied with the results. I feel that sometimes, the spectrum
>> with a concatenated signal would be better. In particular, when I tried, I
>> could see more alpha.
>>
>>
>>
>> What do you think?
>>
>>
>>
>>
>>
>> P.S. I tried using ICA instead, but with only 8 channels (7 after
>> re-referencing), it is very hard to do!
>>
>>
>>
>>
>>
>> _______________________________________________
>> 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
>>
>
>
> _______________________________________________
> 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/20150902/d089d394/attachment.html>


More information about the eeglablist mailing list