[Eeglablist] Baselining and filtering for ICA with epoched data

Wed Aug 29 18:47:07 PDT 2018

Dear Jumana,

> I thought on the page it mentioned that ERP researchers could consider
training on 1Hz then reapplying to 0.1Hz to avoid the situation described
by Steve Luck. I did training on 1Hz, applied to 0.1, and removed blink and
saccadic responses. Would you not advise this?

I do not recommend this trick as a part of standard pipeline. It seems too
tricky to me. I want to take complete control of data processing so that I
know what I am doing. This way, my intuition works better, and if I see
something unexpected I see what was wrong. Also, ASR requires good
suppression of < 1Hz too.

Makoto

On Mon, Aug 20, 2018 at 1:13 PM Ahmad, Jumana <jumana.ahmad at kcl.ac.uk>
wrote:

> Dear Makoto,
> I thought on the page it mentioned that ERP researchers could consider
> training on 1Hz then reapplying to 0.1Hz to avoid the situation described
> by Steve Luck. I did training on 1Hz, applied to 0.1, and removed blink and
> saccadic responses. Would you not advise this?
>
> Best wishes,
> Jumana
>
> > On 20 Aug 2018, at 19:56, Eric Fields <eric.fields at bc.edu> wrote:
> >
> > Hi all,
> >
> > I looked into this a bit with a sort of simulation. I ran ICA on a
> dataset and then I artificially added a big linear drift to one channel in
> one epoch.
> >
> > Here is the original data:
> >
> > [Original.jpg]
> >
> > And here is the big linear drift added to AF4 in the third epoch:
> >
> > [B4_ICA.jpg]
> >
> > I then added the ICA weights (which had been calculated without this
> drift artifact). Here is what the ICA component activations look like:
> >
> > [IC_activations.jpg]
> >
> > As you can see, the drift ends up in several ICs.
> >
> > Here's what it looks like if I remove IC 7:
> >
> > [Remove7.jpg]
> >
> > You can see that there is now drift on Fp1, which didn't appear in the
> pre-ICA corrected data. Removing other ICs affected by the drift does
> something similar, but the effects of each differs a bit.
> >
> > Of course, any time you have a idiosyncratic artifact in a single epoch,
> something like this will happen because the ICA solution will not account
> well for such an artifact. Such epochs just have to be thrown out.
> >
> > But, imagine there were systematic low frequency information throughout
> the study—either noise (e.g., skin potentials) or low frequency signal of
> interest (e.g., part of an ERP slow wave that appears on every trial), and
> this low frequency information is not part of the data you run ICA on. The
> example above suggests this will get distributed across different ICs,
> including perhaps ICs of interest (if you look above, it looks like part of
> the drift ends up on IC 1, which is a blink component). And then if you
> remove one of those ICs, you might remove part of that low frequency
> information and/or change it's scalp distribution, either of which could
> lead to wrong conclusions.
> >
> > I don't claim to be an expert here, but I remain skeptical that
> filtering at 1-2 Hz for running ICA and then applying that solution to an
> 0.01 or 0.1 Hz filtered dataset is a good idea. I've been running ICA on
> 0.1 Hz filtered data, which is the same thing I use for ERP analysis.
> Outside of really noisy/drifty datasets, that generally works fine, at
> least for removing blinks and ocular artifact. Of course, if you are
> working the frequency domain and you aren't looking at low frequency bands
> anyway, then you can just use a more stringent high pass filter at all
> processing steps and you don't have to worry about any of this.
> >
> > Eric
> >
> > -----
> > Eric Fields, Ph.D.
> > Postdoctoral Fellow
> > Cognitive and Affective Neuroscience Laboratory<
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww2.bc.edu%2Felizabeth-kensinger%2F&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=%2FROWp%2BmrAtRt2E7J%2FqeOXN2tLdy8wMihVVIjEM4b98E%3D&reserved=0>,
> Boston College
> > Aging, Culture, and Cognition Laboratory<
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.brandeis.edu%2Fgutchess%2F&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=jyD53q%2BQEJWrt9E%2FL8ASKjYjxQGVhJIRHkgjZ2n%2Fm%2BM%3D&reserved=0>,
> Brandeis University
> > eric.fields at bc.edu<mailto:eric.fields at bc.edu>
> >
> >
> > On Mon, Aug 20, 2018 at 1:03 PM Makoto Miyakoshi <mmiyakoshi at ucsd.edu
> <mailto:mmiyakoshi at ucsd.edu>> wrote:
> > Dear Jumana,
> >
> >> The recommendation on the page is to apply 1Hz to 0.1Hz for ERPs. Is
> this still the recommendation?
> >
> > No. I thought I have never recommended 'ICA on 1-Hz data, then copy it
> to 0.1Hz data' strategy as a part of my preprocessing pipeline. This is
> because not only ICA but also ASR also benefits from using a high-pass
> filter at 1-2 Hz (see the lower-end of the IIR spectral weighting on
> https://sccn.ucsd.edu/wiki/Makoto%27s_preprocessing_pipeline#ASR_FAQ_.28updated_03.2F19.2F2018.29
> <
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fsccn.ucsd.edu%2Fwiki%2FMakoto%2527s_preprocessing_pipeline%23ASR_FAQ_.28updated_03.2F19.2F2018.29&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=m7tGnnnQCyP7nCCmVAMho2%2FuMvRyDtS62nzf91T0CrA%3D&reserved=0>).
> The developer of ASR told me that I should use 0.25 Hz to 0.75 Hz as
> transition band when applying high-pass filter, which is twice as sharp as
> applying the '1-Hz high-pass filter' with the default EEGLAB filter
> function (whose transition band is 0 to 1 Hz). An unmixing matrix resulting
> from ICA can be copied from one dataset to others, but ASR result is the
> time-series itself, you can't copy it to others. For this reason, I do not
> use 'ICA on 1-Hz data, then copy it to 0.1Hz data' strategy myself, and (I
> thought) I have never regarded as a 'necessary process as a default'. I
> still think it is an interesting idea and creative attempt. Certainly there
> could be situation I could benefit from using that strategy.
> >
> > Another reason is that I pay more attention to results in time-frequency
> domain which is usually > 3 Hz (this is from limitation of time-frequency
> analysis: resolving 3 Hz already requires 1.1s-long sliding window with our
> recommended parameters. If you want to resolve 1 Hz, you'll need 3.3s-long
> window, which would be longer than SOAs in most of ERP paradigms). An
> averaged ERP is a broadband phonemenon, broad toward the lower end. If you
> want to investigate ERP in the traditional sense, the analysis deserves a
> dedicated pipeline.
> >
> > Makoto
> >
> >
> >
> > On Sat, Aug 18, 2018 at 12:03 AM Ahmad, Jumana <jumana.ahmad at kcl.ac.uk
> <mailto:jumana.ahmad at kcl.ac.uk>> wrote:
> > Hi Makoto,
> > The recommendation on the page is to apply 1Hz to 0.1Hz for ERPs. Is
> this still the recommendation?
> >
> > Best wishes,
> > Jumana
> >
> > On 18 Aug 2018, at 05:29, Makoto Miyakoshi <mmiyakoshi at ucsd.edu<mailto:
> mmiyakoshi at ucsd.edu>> wrote:
> >
> > Dear Eric and Jumana,
> >
> > I summarized my explanation to this wiki page. Sorry I did not notice I
> made so many typos...
> >
> https://sccn.ucsd.edu/wiki/Makoto%27s_preprocessing_pipeline#What_happens_to_the_.3C_1_Hz_data_if_ICA_is_calculated_on_.3E_1_Hz_data_and_applied_to_0.1_Hz_data.3F_.2808.2F17.2F2018_Updated.29
> <
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fsccn.ucsd.edu%2Fwiki%2FMakoto%2527s_preprocessing_pipeline%23What_happens_to_the_.3C_1_Hz_data_if_ICA_is_calculated_on_.3E_1_Hz_data_and_applied_to_0.1_Hz_data.3F_.2808.2F17.2F2018_Updated.29&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=bD%2BQdQDp2TTG%2FNrwNmFPzm1LGG9LBISI6pNjEcR%2Fc2A%3D&reserved=0
> >
> >
> > Makoto
> >
> > On Tue, Feb 27, 2018 at 9:14 AM Eric Fields <eric.fields at bc.edu<mailto:
> eric.fields at bc.edu>> wrote:
> > Hi Jumana and others,
> >
> > To be clear, my question isn't about implementation (I know it is
> relatively easy to calculate ICA weights from one dataset and apply them to
> another). My question was about how this effects the data.
> >
> > If I calculate the ICA weights using data filtered at 1 Hz, by
> definition this solution will not give maximally independent components
> when it is applied to 0.1 Hz filtered data. In the 0.1 Hz filtered data
> there is low frequency information in the data that ICA was not trained on
> and therefore doesn't "know" anything about. That low frequency information
> must, mathematically, end up somewhere in the IC activations when I
> multiply the unmixing matrix with the 0.1 Hz filtered data. What are the
> consequences and pitfalls of this? Is it something to worry about? Under
> what circumstances?
> >
> > My guess is that the low frequency information gets divided across the
> ICs in a somewhat a priori unpredictable way depending on the scalp
> distribution of the low frequency information and the scalp distribution of
> the ICs. If I then remove one of the ICs, I may remove some of this low
> frequency information with it. If so:
> >
> >  1.  If the low frequency information in question is purely noise, this
> could lead to some pattern of noise/artifact in the data that is hard to
> interpret or move noise to electrodes that didn't originally include it.
> >  2.  One of the reasons for using the 0.1 Hz filter is that part of the
> effects I am interested in (e.g., later ERP components) contain information
> below 1 Hz (see Tanner et al., 2015). Can I be confident that ICA does a
> good job of isolating artifact from neural activity of interest if part of
> that activity of interest was not present in the training dataset?
> >
> > Have these issues been addressed anywhere in the literature or does
> anyone have recommendations?
> >
> > Eric
> >
> > -----
> > Eric Fields, Ph.D.
> > Postdoctoral Fellow
> > Cognitive and Affective Neuroscience Laboratory<
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww2.bc.edu%2Felizabeth-kensinger%2F&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=%2FROWp%2BmrAtRt2E7J%2FqeOXN2tLdy8wMihVVIjEM4b98E%3D&reserved=0>,
> Boston College
> > Aging, Culture, and Cognition Laboratory<
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.brandeis.edu%2Fgutchess%2F&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=jyD53q%2BQEJWrt9E%2FL8ASKjYjxQGVhJIRHkgjZ2n%2Fm%2BM%3D&reserved=0>,
> Brandeis University
> > eric.fields at bc.edu<mailto:eric.fields at bc.edu>
> >
> > On Wed, Feb 21, 2018 at 1:00 PM, Ahmad, Jumana <jumana.ahmad at kcl.ac.uk
> <mailto:jumana.ahmad at kcl.ac.uk>> wrote:
> >
> > It depends if you want to examine component activity below 1Hz. Most
> artifacts of interest, such as blinks and saccades should be higher
> frequency etc.
> >
> >
> > ------------------------------------------
> > Jumana Ahmad
> > Post-Doctoral Research Worker in Cognitive Neuroscience
> > EU-AIMS Longitudinal European Autism Project (LEAP) & SynaG Study
> > Room M1.26<
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fmaps.google.com%2F%3Fq%3DM1.26%26entry%3Dgmail%26source%3Dg&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=RvEHY3M14EglMnxHC3C6RFxHHbNT8kpJUqVv2lT6OMA%3D&reserved=0>.Department
> of Forensic and Neurodevelopmental Sciences (PO 23) | Institute of
> Psychiatry, Psychology & Neuroscience | King’s College London | 16 De
> Crespigny Park | London SE5 8AF<
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fmaps.google.com%2F%3Fq%3D16%2BDe%2BCrespigny%2BPark%2B%257C%2BLondon%2BSE5%2B8AF%26entry%3Dgmail%26source%3Dg&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=ujMU2igp6fpHWIONdmqOEu%2FYGRwhaRWgaRyKH47l03A%3D&reserved=0
> >
> >
> > Phone: 0207 848 5359| Email: jumana.ahmad at kcl.ac.uk<mailto:
> antonia.sanjose at kcl.ac.uk> | Website: www.eu-aims.eu<
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.eu-aims.eu%2F&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=QGq%2BoGijlQDyjO6RvjFJWgPUvnhzOvq0RoWwq6g4OGI%3D&reserved=0>
> | Facebook: www.facebook.com/euaims<
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.facebook.com%2Feuaims&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=NrKv3msyU%2BG%2BMxjR%2F3vBw9UovRq0KMnV%2FtY78q282gA%3D&reserved=0
> >
> >
> >
> > ________________________________
> > From: eeglablist <eeglablist-bounces at sccn.ucsd.edu<mailto:
> eeglablist-bounces at sccn.ucsd.edu>> on behalf of Eric Fields <
> eric.fields at bc.edu<mailto:eric.fields at bc.edu>>
> > Sent: 21 February 2018 03:45:30
> > To: EEGLAB List
> > Subject: [Eeglablist] Baselining and filtering for ICA with epoched data
> >
> > Hi,
> >
> > I know there have been other threads related to this, so I apologize if
> this has been addressed directly and I missed it.
> >
> > Groppe et al. (2009) showed that ICA gives more reliable results if you
> use the full epoch instead of the prestimulus period to baseline. The
> reason generally given for this is that baseline correction changes the
> scalp distribution of sources depending on what is happening in the
> baseline period. By this logic, using the full epoch should improve ICA
> (because longer periods are less affected by random variations), but no
> baseline correction at all should be even better.
> >
> > Meanwhile, Winkler et al. (2015) have suggested that ICA works best on
> data high pass filtered at 1-2 Hz.
> >
> > Assuming I prefer to use a 0.1 Hz high pass filter (because of
> distortions 1 Hz filters can cause in the ERP: Tanner et al., 2015), I have
> two questions:
> >
> >
> >  1.  Does the removal of additional low frequency noise you get from
> using a full epoch baseline (vs no baseline) outweigh the downsides of
> baseline correction for ICA?
> >  2.  Alternatively, is it appropriate to apply a 1 or 2 Hz filter to the
> data used for ICA training, and then apply the ICA solution to an EEGset
> filtered at 0.1 Hz? Winkler et al. suggest this, but what happens to the
> low frequency information in the data when the ICA solution that has been
> learned without it is applied? Can this cause problems?
> >
> > Thanks!
> >
> > Eric
> >
> > -----
> > Eric Fields, Ph.D.
> > Postdoctoral Fellow
> > Cognitive and Affective Neuroscience Laboratory<
> https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww2.bc.edu%2Felizabeth-kensinger%2F&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=%2FROWp%2BmrAtRt2E7J%2FqeOXN2tLdy8wMihVVIjEM4b98E%3D&reserved=0>,
> Boston College
> > Aging, Culture, and Cognition Laboratory<
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.brandeis.edu%2Fgutchess%2F&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=jyD53q%2BQEJWrt9E%2FL8ASKjYjxQGVhJIRHkgjZ2n%2Fm%2BM%3D&reserved=0>,
> Brandeis University
> > eric.fields at bc.edu<mailto:eric.fields at bc.edu>
> >
> > _______________________________________________
> > Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html<
> https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fsccn.ucsd.edu%2Feeglab%2Feeglabmail.html&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7Cbd6a2cf749634ef9bc7308d606ce9f40%7C8370cf1416f34c16b83c724071654356%7C0&sdata=4TnEu326gWzZ0lmu%2BfQvIY3cVq%2BEPm9GiwjKqx5zpHE%3D&reserved=0
> >
> > 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
> > 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
> > <Original.jpg>
> > <B4_ICA.jpg>
> > <IC_activations.jpg>
> > <Remove7.jpg>
>

-- 
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/20180829/fd1f8ade/attachment.html>