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

Thu May 11 19:49:49 PDT 2017

Dear Marco,

Thank you for your valuable comments here. As I can not change the number
of channel because my device is fix in that case (Emotiv EPOC headset),
from here I can conclude that in my case ADJUST is not recommended. I must
say, ADJUST is a nice tool for EEG based research community.

One thing more, some talk about the value of *K. *can you advice a specific
value of this parameter in my case, As I am confuse how to get this value.

Thank you for the advice, Yeah I will increase the registration time to get
enough training points for ICA decomposition and do the necessary steps for
ICA decomposition.

Thank you.

Best Regards,

Rabnawaz

On Tue, May 9, 2017 at 5:58 PM, Marco Buiatti <marco.buiatti at gmail.com>
wrote:

> Dear Rabnawaz,
>
> unfortunately, 14 channels do not have the spatial resolution necessary
> for ADJUST spatial features to be correctly computed, no matter the time
> duration of the registration.
>
> The importance of the duration of EEG recordings is relative to the
> quality of ICA decomposition. Please follow EEGLAB's indications on ICA
> decomposition here: https://sccn.ucsd.edu/wiki/Chapter_09:_Decomposing_
> Data_Using_ICA
> where you can also find a "rule of thumb" for the minimum duration of the
> recordings:
> "As a general rule, finding *N*stable components (from N-channel data)
> typically requires *more than* *kN^2* data sample points (at each
> channel), where N^2 is the number of weights in the unmixing matrix that
> ICA is trying to learn and *k* is a multiplier. In our experience, the
> value of *k* increases as the number of channels increases. In our
> example using 32 channels, we have 30800 data points, giving 30800/32^2 =
> 30 pts/weight points."
>
> My advice is:
> - use longer recordings (whatever measure you plan to use, 180 s are
> really short for any analysis!)
> - clean your data from paroxystic, non-stereotyped artifacts (crucial for
> a good ICA decomposition)
> - run ICA and test whether it efficiently isolates artifacts from
> neural-like components.
>
> Good luck,
>
> Marco
>
> On 9 May 2017 at 03:22, Rabnawaz khan <13mseerabnawaz at seecs.edu.pk> wrote:
>
>> Dear Marco,
>>
>> Hoping all is well, I would like to get a bit more in my previous query.
>> there are two condition, one is number of channel (which is fix in my case,
>> 14 channels) the second is time duration of data(which I can increase). can
>> you suggest minimum duration of EEG recordings with 14 channels to properly
>> apply ADJUST and get good results.
>>
>>
>>
>>
>> Best Regards,
>>
>> Rabnawaz
>>
>> On Wed, May 3, 2017 at 11:48 AM, Rabnawaz khan <
>> 13mseerabnawaz at seecs.edu.pk> wrote:
>>
>>> Dear Armand and Marco,
>>>
>>> Thank you so much for your valuable comments. Actually my task is to
>>> work out for Alpha absolute power values for this data. for that reason i
>>> wanted to use automatic artifact removal in my pre-processing phase, but i
>>> was surprised after getting un-expecting values for alpha absolute power
>>> from my data.
>>>
>>> considering the scenario, there is much noise present in the data. So
>>> working by visual inspection and rejecting the the noisy portion from my
>>> data now its seems better as I get expected values for alpha band of EEG.
>>> Thanks
>>>
>>>
>>>
>>>
>>> Best Regards,
>>>
>>> Rabnawaz
>>>
>>> On Tue, May 2, 2017 at 3:05 PM, Marco Buiatti <marco.buiatti at gmail.com>
>>> wrote:
>>>
>>>> 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.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.uc
>>>>> sd.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 <+39%200464%20808178>
>>>> https://sites.google.com/a/unitn.it/marcobuiatti/
>>>>
>>>> ***********************************************
>>>>
>>>
>>>
>>
>
>
> --
> 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 <+39%200464%20808178>
> https://sites.google.com/a/unitn.it/marcobuiatti/
>
> ***********************************************
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20170512/fd41d858/attachment.html>