[Eeglablist] On ICA based artifact rejection
Gunseli, E.
e.gunseli at vu.nl
Wed Sep 12 04:02:24 PDT 2012
Dear all,
I have a question about the steps that should be taken before running ICA.
If we are going to epoch the data (step 5), why are we manually rejecting the extra noisy parts earlier (step 2).
Since these extra noisy portions are mostly at beginning and end of trial blocks, they will be gone away during epoching anyway.
So, epoching the critical time window can save a fair amount of time that else we would have spent on manual inspection.
At this point I have another question; I have read that, it is better to run ICA on continuous, non-epoched data.
One of the problems of running ICA on epoched data is that, "the baseline correction changes relative values across channels" (S. Luck, ERP Boot Camp Lecture Slides). But probably that is not the only reason to run ICA on continuous data because this problem can easily be overcome via removing the baseline after running ICAs.
So I guess there should be other problems related to running ICAs on epoched data.
Can anyone provide information about these potential problems?
Kind regards,
Eren
From: eeglablist-bounces at sccn.ucsd.edu [mailto:eeglablist-bounces at sccn.ucsd.edu] On Behalf Of Makoto Miyakoshi
Sent: Friday, September 07, 2012 00:58
To: IMALI THANUJA HETTIARACHCHI
Cc: eeglablist at sccn.ucsd.edu
Subject: Re: [Eeglablist] On ICA based artifact rejection
Dear Imali,
Pipeline check?
1. Re-reference continues data (The data is collected on a bipolar montage so re-reference to the common average)
2. Reject unsuitable portions of data by visual inspection
3. High pass filter the data(cut-off @ 0.5Hz to preserve ERP components), low pass filter the data(cut-off 30Hz)
I would say filtering should be done before data rejection, since the rejection creates boundaries which can confuse filtering.
4. Extract epochs (without baseline removal???)
According to Groppe et al., whole-epoch baseline is better than usual short pre-stimulus baseline. Scott says no baseline correction is even better. So without baseline removal may be better.
5. Run ICA
6. A. Tools>Remove components to subtract ICA components or should I do
B. Tools> Reject data Epochs> reject data(all methods),( but if I do this how can that be an artifact rejection by ICA) or
C.Tools>Reject data epochs>export marks to ICA reject and then Tools>Reject data epochs>Reject marked epochs?
6A is not necessary if you are going to use STUDY (it will create clusters for artifacts) so don't bother to do this. 6B is recommended. Theoretically, 2nd ICA after 6B improves the quality of decomposition, but in my experience it rarely changes the result unless drastic rejection, either quantitatively or qualitatively, is performed.
Could you please make it clear to me how I should reject epochs using ICA after the first decomposition.
Just take a look at independent component activities as you check your channel EEG data. It is a good idea to use 'all methods' for obtaining statistical suggestions (but don't take them blindly). You may want to discard 5-10 % of data here, depending your data quality. Improbability test is good but hopefully it is done after thresholding on channel EEG (therefore I recommend mild amplitude threshold on channel EEG before ICA, and then improbability test on IC activities).
7. Then Tools>Remove Baseline and Plot>Channel ERP's steps will give me the ERP for a particular stimulation?
Yes.
8. Now to do dipole localisation Run ICA on the pruned data set and run DIPFIT, here won't I get the same remaining ICA components from the first epoched data set?
Again, whether or not running the 2nd ICA depends on how much you care about data quality. DIPFIT does not care your IC activities. It only cares about scalp maps ICA generated. Therefore, epoch rejection does not affect DIPFIT performance.
You should locate two dipoles manually when you find bilateral topographies using an interactive 'fine fit' GUI of DIPFIT.
Makoto
2012/9/6 Stephen Politzer-Ahles <politzerahless at gmail.com<mailto:politzerahless at gmail.com>>
Hi Imali,
I don't have experience with using "reject based on ICA", but the first option you pointed out (6A--using Tools>Reject components to remove the IC(s) with artifact) works. What I have typically done is first use Tools>Reject components to do that, and then use Tools>Reject epochs (by inspection) on the cleaned data to go through and reject any epochs that contain other artifact. (In my case, I use ICA to remove the blink artifacts, but then must reject by inspection to remove artifact that's left over such as skin potentials or EMG).
Maybe some others on the list can give you some more information about the other methods, which I am not familiar with.
Best,
Steve
On Wed, Sep 5, 2012 at 10:28 PM, IMALI THANUJA HETTIARACHCHI <ith at deakin.edu.au<mailto:ith at deakin.edu.au>> wrote:
Thank you very much Makoto, really appreciate your guidance and help.
I have further some questions regarding ICA artifact rejection and localisation.
In EEGLAb we use ICA for both artifact rejection and source localisation? As I want to use EEGLAB for my dipole localisation, I am a bit confused with the steps that I should follow, I read the details on artifact rejection given on the wiki and a thread on 'Pipeline of processing to optimize ICA for artrifact removal' on the discussion list, but still not clear on the steps. Below I will briefly give the steps which I understood that I should follow, can you please tell me whether my understanding is correct and comment if I have gone wrong somewhere?
1. Re-reference continues data (The data is collected on a bipolar montage so re-reference to the common average)
2. Reject unsuitable portions of data by visual inspection
3. High pass filter the data(cut-off @ 0.5Hz to preserve ERP components), low pass filter the data(cut-off 30Hz)
4. Extract epochs (without baseline removal???)
5. Run ICA
Now I get confused, after the ICA decomposition I will be able to view the ICA components with Tools> Reject data using ICA>Reject components by map
With this window I can detect the components for eye artifacts, muscle artifacts etc. Then is it,
6. A. Tools>Remove components to subtract ICA components or should I do
B. Tools> Reject data Epochs> reject data(all methods),( but if I do this how can that be an artifact rejection by ICA) or
C.Tools>Reject data epochs>export marks to ICA reject and then Tools>Reject data epochs>Reject marked epochs?
Could you please make it clear to me how I should reject epochs using ICA after the first decomposition.
7. Then Tools>Remove Baseline and Plot>Channel ERP's steps will give me the ERP for a particular stimulation?
8. Now to do dipole localisation Run ICA on the pruned data set and run DIPFIT, here won't I get the same remaining ICA components from the first epoched data set?
Many thanks,
Imali
From: Makoto Miyakoshi [mailto:mmiyakoshi at ucsd.edu<mailto:mmiyakoshi at ucsd.edu>]
Sent: Wednesday, 5 September 2012 7:02 AM
To: IMALI THANUJA HETTIARACHCHI
Cc: eeglablist at sccn.ucsd.edu<mailto:eeglablist at sccn.ucsd.edu>
Subject: Re: [Eeglablist] ERP localisation with BESA and DIPFIT with ICA
Dear Imali,
ICA returns 'one map/IC per a component' which does not change across recording time.
A static location corresponds to a brain region.
If you think of averaged ERP topo, for example, scalp topography changes from timepoint to timepoint. Independent components are not like that.
> 2. Do independent components for cognitive activity in brain represents ERP components(P1,N1, etc)?
Not necessarily. One IC can explain 3 ERP peaks (P1/N1/P2 as one burst).
> 3. Since I have minimal(correct to say no..) experience in ERP, how do I know my dipole localisations with ICA are correct? For instance, in a visual task I would expect to see one or more dipoles in visual area, but when changing the conditions such as colour or shape where else do I get dipoles? Or simply, how do I have a hypothesis for the ICA component related dipoles?
How do I know my dipole location is correct?
When you calculate dipole fit, you'll have residual variance. If this value is small, that means your dipole location is good.
For symmetrical two dipoles, when the topography show bilateral pattern you should place two dipoles (This may require some prior knowledge about somatosensory mu, alpha, and EOG).
> 4. With very limited neuroscience knowledge how do I get around with localisations to extract a task related neuronal activity?
If you don't have time to read Scott Makeig, Arnaud Delorme, or Julie Onton etc, then
1. run ICA
2. run dipfit (autofit)
Remember, 1 dipole for 1 (or bilateral 2) IC(s). They are always paired. ICA generates time-invariant scalp topo, and dipfit calculates the associated dipole(s) that is also time-invariant (ICs don't change their locations throughout your data just as your brain regions don't).
If you have further questions please ask further.
Makoto
2012/8/31 IMALI THANUJA HETTIARACHCHI <ith at deakin.edu.au<mailto:ith at deakin.edu.au>>
Dear EEGLAB list,
While reading through papers for my experiments, I just became curious (with some confusion) on the dipole fitting approach of the ERP data(for a specific task).
According to my understanding the ERP wave consists of several components such as P1,N1, P2 , N2 and P3 mainly (stimulus dependent). As I am intending to use ICA based source localization(using DIPFIT plugin) I wanted to find out on what degree the two dipole fitting approaches are differing in BESA and DIPFIT with ICA.
1. Am I correct if I say that with BESA, dipoles can be fitted to individual components of the ERP waveform?
2. Do independent components for cognitive activity in brain represents ERP components(P1,N1, etc)?
3. Since I have minimal(correct to say no..) experience in ERP, how do I know my dipole localisations with ICA are correct? For instance, in a visual task I would expect to see one or more dipoles in visual area, but when changing the conditions such as colour or shape where else do I get dipoles? Or simply, how do I have a hypothesis for the ICA component related dipoles?
4. With very limited neuroscience knowledge how do I get around with localisations to extract a task related neuronal activity?
Sorry about throwing a lot of questions at the list, but I have always found EEGLAB list as very friendly and a very expertized group. So, your advice will be highly appreciated to move forward in my work.
Best regards
Imali
Imali Thanuja Hettiarachchi
PhD Candidate
Centre for Intelligent Systems research
Deakin University, Geelong 3217, Australia.
Email: ith at deakin.edu.au<mailto:ith at deakin.edu.au>
www.deakin.edu.au/cisr<http://www.deakin.edu.au/cisr>
[Description: Description: Description: cid:1216BE20-1800-4A47-8B9F-E7B9D94831CD at deakin.edu.au]
_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu<mailto: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<mailto:eeglablist-request at sccn.ucsd.edu>
--
Makoto Miyakoshi
JSPS Postdoctral Fellow for Research Abroad
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu<mailto: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<mailto:eeglablist-request at sccn.ucsd.edu>
--
Stephen Politzer-Ahles
University of Kansas
Linguistics Department
http://www.linguistics.ku.edu/
--
Makoto Miyakoshi
JSPS Postdoctral Fellow for Research Abroad
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/20120912/cc4912cf/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.jpg
Type: image/jpeg
Size: 29155 bytes
Desc: image001.jpg
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120912/cc4912cf/attachment.jpg>
More information about the eeglablist
mailing list