[Eeglablist] Re. Using ICA with interpolated channels

Arnaud Delorme arno at ucsd.edu
Thu Sep 30 11:02:17 PDT 2010


Dear Grega and Robert,

concerning interpolating channels before running ICA. Robert's analysis of pros and cons of interpolating channels before running ICA makes perfect sense. We personally at the Swartz center never interpolate data channels and rarely reduce the dimensionality of the data matrix before running ICA (unless we are using 256 channels and then we reduce the dimensionality to 150 before running ICA). As Robert pointed out both reducing the dimensionality using PCA and interpolating channels introduce non linearities. Spherical interpolation introduces non-linearities because a non-linear algorithm is used to interpolate channels. Pre-processing with PCA introduces non-linearities because some of the PCA components - the ones with the lowest eigenvalues - are discarded. Since PCA does not model the structure of the data (i.e. the brain sources), this introduces non linearity. It is hard enough to run ICA and get a clean decomposition for the purpose of analyzing brain source that it is better not to apply any procedure that would potentially introduce non-linearity. When running ICA for the purpose of removing artifacts, this is probably less critical.

Just wanted to add my grain of salt,
Cheers,

Arno

On Sep 29, 2010, at 4:36 AM, Robert Whelan wrote:

> Dear Grega, 
> 
> That is a great suggestion -- thank you. Jordi Costa Faidella emailed me directly with the same suggestion yesterday and we've already started coding up the approach that you describe -- should be done and tested in a day or two. We will also run our EEG data through the new approach and quantify the difference between interpolating channels vs. removing channels before ICA.
> 
> Re. Joseph Dien's ERP PCA toolkit. At the time of writing our paper we wanted to pick a method from the literature for comparison (although the ERP PCA toolkit has been available for a while), and with the publication of Dien (2010) we will definitely compare the two approaches. Although I haven't used the Toolkit yet, I read the Dien (2010) paper recently and it looks great.
> 
> Thanks again, 
> 
> Rob & Hugh
> 
> On Tue, Sep 28, 2010 at 9:14 PM, Grega Repovs <grega.repovs at psy.ff.uni-lj.si> wrote:
> Dear Rob & Hugh,
> 
> Since there seem to be arguments against using the problematic channels both before as well as after interpolation, why not run ICA without those channels. So the procedure would be:
> 
> 1/ Identify and remove bad channels
> 2/ Perform ICA on good channels only
> 3/ Remove bad ICA components
> 4/ Reconstruct good channels
> 5/ Interpolate bad channels
> 
> This way neither noise nor non-linearities would affect the ICA solution and bad channels can still be interpolated based on cleaned data.
> 
> I also have one other question with regards to FASTER. In your paper you compared it to SCADS. I was wondering, why did you not compare it to ERP PCA Toolkit by Joseph Dien, which also performs fully automated data preprocessing and employs algorithms similar to FASTER. I myself would be quite interested in that comparison.
> 
> All the best,
> 
> Grega Repovs
> 
> 
> 
> On Sep 28, 2010, at 12:04 PM, Robert Whelan wrote:
> 
>> Jordi Costa Faidella wrote "Is it correct to perform an ICA on a dataset in which some of the channels have been interpolated?"
>> This is an interesting question and we considered both orders (each order has some advantages and disadvantages) for the FASTER method. Ultimately, we decided to run interpolation first followed ICA. Here was our rationale:
>> As the EEGLAB manual recommends – “ICA works best when given a large amount of basically similar and mostly clean data.” (see p.59). Therefore, an ICA on a dataset in which some channels are noisy (perhaps with a lot of non-stereotypic data due to a problem with the electrode) may decrease the quality of the ICA (i.e., dissimilar activations are mixed into the ICs).
>> On the other hand, interpolating before ICA raises a couple of issues 1) it reduces the dimensionality of the data and 2) introduces some non-linearity into the data (if the interpolation method was not linear), which is detrimental to the ICA solution. We dealt with Issue 1in FASTER by restricting the maximum number of ICs to correspond with the reduced rank of the data after interpolation.
>> The choice then was between reducing the quality of the ICA by introducing noisy channels or reducing the quality of the ICA by the non-linearity introduced due to spherical interpolation. Although ICA assumes linearity, there is almost certainly some non-linearity in the signals recorded at the scalp, and the non-linearity introduced by spherical interpolation is likely only a small contributer to the overall non-linearity. In any case, based on pilot testing we found that when the ICA was done with noisy channels included (i.e., not interpolated out) the resulting components were less useful than when the data were cleaner (i.e., the channels were interpolated). As an aside, testing algorithms on real data proved much more informative than testing on the simulated data, perhaps due to the inclusion of non-stereotypic artefacts in the real data. 
>> That said, we are certainly open to persuasion on this issue and/or suggestions about how to quantify which order is better. Also, might there be situations in which one order is superior to the other, perhaps depending on the maximum number of ICs that can be generated? 
>> If there is demand, we can also configure FASTER so that the user can select the order of the processing steps. Email me directly robert.whelan at tcd.ie or whelanrob at gmail.com if this is something that people might want or with any other suggestions.
>> 
>> Best Regards, 
>> 
>> Rob & Hugh
>> 
>> -- 
>> Robert Whelan, PhD
>> Senior Research Scientist
>> 
>> Trinity Centre for Bioengineering
>> Trinity College Dublin
>> 
>> Department of Neurology
>> St. Vincent's University Hospital
>> Elm Park, Dublin 4
>> 
>> webpage: http://www.mee.tcd.ie/~neuraleng/People/Robert
>> _______________________________________________
>> 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
> 
> 
> 
> 
> -- 
> Asist. Prof. Grega Repovš, Ph.D.
> Department of Psychology
> University of Ljubljana
> Aškerčeva 2
> SI-1000 Ljubljana
> tel: +386 1 241 1175
> email: grega.repovs at psy.ff.uni-lj.si
> 
> 
> 
> 
> -- 
> Robert Whelan, PhD
> Senior Research Scientist
> 
> Trinity Centre for Bioengineering
> Trinity College Dublin
> 
> Department of Neurology
> St. Vincent's University Hospital
> Elm Park, Dublin 4
> 
> webpage: http://www.mee.tcd.ie/~neuraleng/People/Robert
> _______________________________________________
> 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

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20100930/c285fa69/attachment-0001.html 


More information about the eeglablist mailing list