[Eeglablist] Problem with channel detection in Run ICA

Wed Jun 27 07:22:34 PDT 2012

Yes, now that I am using sensitive measures for identifying linked  
sites I am finding that in dense array EEG it is more common than I  
had hoped. I think that the linking of scalp sites could also relate  
to the interesting phenomenon that Jan initially mentioned:

"... the channels that are not displayed in the channel selection
dialog before running the ICA is not systematic, once it was for example the
EEG channel F5, once the EEG channel P7."

In the case of a pair of linked channels, the selection of which site  
is accused of being rank deficient is arbitrary. For example, if sites  
F3 and F5 are linked and sites P7 and P5 are linked, it makes sense to  
me that in one run you could find that F3 and P7 are deemed rank  
deficient, then on a second run of the identical data set find that  
sites F5 and P5 are deemed rank deficient.

Of course... this could not account for Jan's exact example without a  
very "fancy" (and unlikely) bridge between sites F5 and P7.

Quoting Makoto Miyakoshi <mmiyakoshi at ucsd.edu>:

> Dear James,
>
>> Linked channels can be overly correlated. electrolyte bridges cause
>> neighbouring channels to share the same information resulting in each
>> bridged pair decreasing the rank of the data by 1.
>
> This reminded me of a story I heard before that someone once saw a
> scalp topo that showed a huge red bar present around from parietal to
> occipital due to massive bridging :-)
>
> Makoto
>
> 2012/6/26 James Desjardins <jdesjardins at brocku.ca>:
>> Hi All,
>>
>> I have been dealing with this as well.
>>
>> When the rank of the scalp data is smaller than the number of channels
>> it means that there are highly predictable channels in the data (they
>> do not have information that could not be derived from other channels
>> in the data).
>>
>> Re-referencing reduces the rank of the data by 1 (e.g. in average
>> referenced data each channel is perfectly negatively correlated with
>> the average of all other channels).
>>
>> Interpolating data channels does not increase the rank of the data. If
>> you have 128 channels and you interpolate 5 bad channels your rank
>> will be 128-5 at best.
>>
>> Linked channels can be overly correlated. electrolyte bridges cause
>> neighbouring channels to share the same information resulting in each
>> bridged pair decreasing the rank of the data by 1. I have started
>> checking my data for unusually large and invariant correlation
>> coefficients across neighbouring sites.
>>
>>
>> --
>> James Desjardins, MA
>> Technician, Cognitive and Affective Neuroscience Lab
>> Department of Psychology, Behavioural Neuroscience
>> Brock University
>> 500 Glenridge Ave.
>> St. Catharines, ON, Canada
>> L2S 3A1
>> 905-688-5550 x4676
>>
>>
>> Quoting Makoto Miyakoshi <mmiyakoshi at ucsd.edu>:
>>
>>> Dear Jason,
>>>
>>> In short, Jan is asking why he sometimes has different ranks though
>>> having the same number of channels. I'm interested in this question
>>> too. I appreciate your help.
>>>
>>> Makoto
>>>
>>> 2012/6/21 Remi, Jan Dr. <Jan.Remi at med.uni-muenchen.de>:
>>>> Dear EEGLAB users,
>>>>
>>>> I am using EEGLAB to run an ICA on my EEG data that I acquire in   
>>>> an EEG-fMRI
>>>> environment to ultimately get rid of the cardioballistogram   
>>>> artifact that is
>>>> typical for recording EEG inside the strong magnet of an MRI machine.
>>>>
>>>> Recently I get a message that reads as follows:
>>>> "EEGLAB has detected that the rank of your data matrix is lower [than] the
>>>> number of input data channels. This might be because you are including a
>>>> reference channel or because you are running a second ICA   
>>>> decomposition. The
>>>> proposed dimension for ICA is 57 (out of 62 channels). Rank   
>>>> computation may
>>>> be inaccurate so you may edit this number below. If you do not understand,
>>>> simply press OK below."
>>>>
>>>> Besides being very thankful for the last sentence, I really do not
>>>> understand the problem. Actually the number of channels that   
>>>> EEGLAB proposes
>>>> varies between 57 and 60 (out of the actual 62 channels) for the 6 files I
>>>> want to run the ICA on. These files differ only in the stimulus condition,
>>>> the EEG properties are not changed at all, they are recorded on   
>>>> the same EEG
>>>> machine (Neuroscan Maglink), with the exact same setup for   
>>>> approximately the
>>>> same time (about 9:45 minutes each). So while I of course do   
>>>> expect the EEG
>>>> to differ in some properties of the EEG signal, i.e. changes in gamma band
>>>> etc., the recording setup conditions are the same. So I do not see where
>>>> there would be a systematic mistake in the recording, especially since I
>>>> have had the same failure notice on a data set, where I had used the ICA
>>>> before without any problem and then 2 weeks later, when I wanted   
>>>> to redo the
>>>> ICA on the same EEG data, where I had only applied a different   
>>>> filter in the
>>>> Neuroscan software before running the ICA analysis (a different low
>>>> frequency filter), I get the same failure notice.
>>>> More over, the channels that are not displayed in the channel selection
>>>> dialog before running the ICA is not systematic, once it was for   
>>>> example the
>>>> EEG channel F5, once the EEG channel P7.
>>>>
>>>> The ICA itself gets me great decomposition, I can get rid of the artifact
>>>> very nicely, I am happy with the resulting data, but I don't like  
>>>>  the idea,
>>>> that I am possibly systematically missing data. I do read the EEG in a
>>>> clinical way, I am a medical researcher.
>>>>
>>>> Any ideas where my mistake could be?
>>>> A similar question had been asked in 2011 and 2009, mainly pertaining to a
>>>> problem of displaying all channels in a 32 bit dataset.
>>>>
>>>> In case you need screenshots of my problem I will be happy to   
>>>> answer emails
>>>> to my email-adress directly.
>>>>
>>>> Thank you all, I enjoy EEGLAB and its community a lot,
>>>>
>>>> Jan Rémi
>>>> Epilepsy and Sleep Center, Department of Neurology, University of Munich
>>>> currently: Department of Neurology, University of Coimbra, Portugal
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> 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
>>>
>>>
>>>
>>> --
>>> 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
>>> 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
>
>
>
> --
> 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
> For digest mode, send an email with the subject "set digest mime" to  
>  eeglablist-request at sccn.ucsd.edu
>