[Eeglablist] Artifact removal using ADJUST plugin (Rabnawaz khan)

Marco Buiatti marco.buiatti at gmail.com
Tue May 2 00:05:28 PDT 2017

Dear Rabnawaz,

Armand is right, 180 seconds are definitely not enough to obtain a good ICA
decomposition, and 14 channels are too few for ADJUST working properly. The
reason for this is that ADJUST is based on spatial features which are not
correctly computed for montages below 32 channels.

A final, general consideration: the term "automatic" refers to the fact
that no parameter tuning is needed, not that it works magically (neither
ADJUST nor any other automatic method pretends to be 100% reliable). When
using automatic methods, you should always review the results to
double-check their efficacy.

All the best,

Marco Buiatti (main ADJUST developer)

On 29 April 2017 at 13:16, Armand Mensen <research.mensen at gmail.com> wrote:

> Hi Rabnawaz,
> So with only 14 channels and 180 seconds of recording, its going to be
> very difficult for you to "clean" your data.
> For ICA: While there is no set rule for just how many channels or length
> of recording will result in a good separation of independent components...
> considered that the maximum number of independent components you are able
> to find is equal to the rank of your dataset (in your case 14 channels).
> Thus it seems highly unlikely that ICA will find a single component that
> captures the spurious (as you descrive it) artefact without also captures
> some genuine neural activity with it.
> In terms of time, the longer your recording the better the ICA algorithm
> can separate truly independent sources. 180 seconds is very little
> information to go on for the ICA learning. The quality of the resulting
> independent components will also depend on what sort of sources are in your
> data. An active component that is consistent throughout the 180 seconds of
> your recording is more likely to be captures than a few random artefacts
> etc.
> There are complete books and 100s of papers written about ICA, and so I
> cannot describe all the pitfalls in this email, and I would suggest having
> a good look through the literature before blindly applying a tool like
> ADJUST and hoping for the best.
> My best advice is that you attempt with different filtering settings to
> eliminate the artefacts you find in your data. If your filters intrude in
> your frequencies of interest, then I would suggest rejecting those samples
> in your data with artefacts and working with the remaining time that is
> artefact free (although with only 180 seconds, you don't have much to work
> with).
> Good luck!
> Armand
> On 29 April 2017 at 07:13, Rabnawaz khan <13mseerabnawaz at seecs.edu.pk>
> wrote:
>> Dear Mensen,
>> Thank you for your response.
>> After plotting my data using EEGlab plot (scroll plot option), I can see
>> some portion of my data is very noisy, most of the channels are giving huge
>> peaks for some duration (I am not sure how to differentiate these noise but
>> I can guess these are not the eye or muscle artefacts). Moreover, I am
>> using Emotiv EPOC headset for recordings, which is 14 channels data, and
>> the duration of my recordings is 180sec.
>> any good advice dealing with it?
>> thanks,
>> Best Regards,
>> Rabnawaz
>> On Wed, Apr 26, 2017 at 4:49 PM, Armand Mensen <research.mensen at gmail.com
>> > wrote:
>>> Dear Rabnawaz,
>>> A few things here to comment on.
>>> 1) I think you may be expecting too much from any automatic artefact
>>> removal (or manual ones for that matter). Which sorts of artefacts any tool
>>> can remove from your data may depend on a multitude of factors. For
>>> example, ICA is generally quite good at finding eye blinks and eye
>>> movements. However even there it depends on the number of channels you
>>> recorded, and the length of your recording time, whether those sorts of
>>> artefacts can be removed without taking too much good data with them. So
>>> what sort of artefacts are you trying to deal with? How many channels do
>>> you have? How long is your recording?
>>> 2) It is generally unwise to run multiple ICA analysis one after
>>> another. There are a couple of reasons for this, but the main one (I think)
>>> is that you are reducing the rank of your data each run. This rank
>>> reduction is not always so easily or accurately detectable and will lead to
>>> problems.
>>> 3) I've used ADJUST sparingly in the past, and wasn't overly impressed.
>>> While examining components manually does take longer than just running some
>>> automatic script... you will get a good feeling for the quality of your
>>> data as well as the sorts of strong independent sources that are there
>>> (whether artefactual or not). I completely see the utility of using the
>>> same artefact criteria and taking some of the subjective decision making
>>> out of using ICA to remove artefacts... however only to the extent that
>>> these standard criteria are really generalisable, and useful in the first
>>> place [apologies to any ADJUST creator or enthusiast; I'd be happy to be
>>> convinced otherwise].
>>> Good luck with your analysis!
>>> Armand
>>>> ---------- Forwarded message ----------
>>>> From: Rabnawaz khan <13mseerabnawaz at seecs.edu.pk>
>>>> To: eeglablist at sccn.ucsd.edu
>>>> Cc:
>>>> Bcc:
>>>> Date: Tue, 25 Apr 2017 18:34:43 +0800
>>>> Subject: [Eeglablist] Artifact removal using ADJUST plugin
>>>> Dear All,
>>>> I am using ADJUST plugin for artifact removal from raw data. I follow
>>>> the instruction listed in the ADJUST manual to process the data. after
>>>> running ADJUST I get the artifacted ICs in a new pop-up window, I mark
>>>> these ICs for rejection and then I go to tools to remove these ICs (via
>>>> EEGlab GUI menu>>tools>>remove components). A new dataset is created
>>>> (eegdata pruned with ICA ), according to ADJUST tutorial this is the clean
>>>> eeg data, but I see from the plots that there is still artifact present in
>>>> the data. When I run ICA again and then again run ADJUST I get a new pop-up
>>>> window in which some other ICs are identified as artifacts. I remove these
>>>> ICs again. But still, artifact are present in data. I repeat running
>>>> ADUST many time and each time I get new ICs marked as an artifact.
>>>> I would be happy if you guide me with this problem.
>>>> Best Regards,
>>>> Rabnawaz
>>> _______________________________________________
>>> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
>>> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.uc
>>> sd.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

Marco Buiatti

Neonatal Neuroimaging Unit
Center for Mind/Brain Sciences
University of Trento,
Piazza della Manifattura 1, 38068 Rovereto (TN), Italy
E-mail: marco.buiatti at unitn.it
Phone: +39 0464-808178

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20170502/0c7c85cc/attachment.html>

More information about the eeglablist mailing list