[Eeglablist] ICA reconstruction and alteration of the artifact free portions of the original EEG

Robert Thatcher rwthatcher2 at yahoo.com
Mon Jun 19 06:39:06 PDT 2017

Hi Jason,

   Thank you and lets agree to a few basicsthat I am sure you will agree to and mayhelp in our discourse: 
Letsagree that once the fundamental Physics of an electrical potential has beenreplaced by a different and altered reality then this invalidates allsubsequent analyses because the original measures of mother of the brain is thephysics of the universe and not a mathematical concept like ICA used to replace the originalnatural physics of the brain.
My replies are below your questions:


Thanks for theclarification. So do I understand correctly that for the comparisons you areshowing, you (1) manually selected non-blink segments and ran the “pre”measures, then (2) used EEGLAB to run ICA on the complete datasets, then (3)selected the (single) eye-blink component from the independent components (ICs)and removed it, then (4) again extracted the same previously manually selectednon-blink segments from the (back-projected, ICA blink-free) EEG.data, and ranthe measures again to produce the “delorme” results?”

Reply:  Yes on #1 but no on #2 because it wasclinicians in Australia that used WinEEG ICA reconstruction after removing oneICA component that represented the eye movement artifact.  The clinicians had noticed large differences incoherence and phase differences between the original artifact free selectionsas compared to the ICA reconstructed data during a workshop.  As explained by Ramesh Srinivasan eyemovement artifact is broad band and superimposed on all EEG channels and notjust the frontal leads.  Delormeperformed his own EEGLAB ICA reconstruction and he did not use the WinEEGICA.   Both the Australian clinicians andArno used 18 ICA components to reconstruct 19 channels – both created somethingout of nothing or like Georges states: “I will give you 18 bitcoins if you giveme 19”.


“Focusing on thecoherence results, it is surprising to me that removing only a frontallyconcentrated eye-blink component would significantly change the coherence innon-blink segments between say P4 and T6 electrodes, in non-blink, or evenblink segments. I believe the impact of these findings would be greatlyenhanced if you or your colleagues at Applied Neuroscience could produce (1)the ICA component maps that were rejected in this analysis, (2) the actualcoherence values (not just the p-values), and (3) the datasets you used,including the original data and the manually selected blink-free data, so thatthe results could be reproduced and verified by others. Apologies again if youhave been over this previously in the discussion.”


Reply:  As explained by Ramesh Srinivasan eyemovement artifact is broad band and superimposed on all EEG channels and notjust the frontal leads.  I have posted .zipfiles that contain all the numbers that you are mentioning.   This is how Arno was able to perform his ownindependent ICA reconstruction which also altered all of the phase differenceslike the Australian clinicians did.  Hereis a url to the .zip files so that you can download and do your own analysesand comparisons. 


“One of the centralmotivations of the ICA approach is that channel level coherence isfundamentally misleading since brain sources in a particular area may projectwidely across most or all electrodes, so that coherence between electrodes atdifferent channel locations is not a valid measure of coherence between brainregions corresponding to those locations. Non-zero phase lag is not assusceptible to this criticism as single sources will project with zerophase-lag, though still the attribution of the “source” of the activity to thebrain region below the channel is suspect. However, that said, ICA advocateswould not expect the channel coherence that exists between non-frontal channelsto be significantly altered by the removal of an eye-blink IC, and, particularlygiven the consternation created in the clinical field, it would be of greatimportance to determine the cause.”

Reply: I disagree that“coherence is fundamentally misleading”. Volume conduction occurs at zero phase lag and one can use the imaginarycomponent of the cross-spectrum to eliminate the volume conduction component.  Also, one can use directed coherence or thephase slope index or lagged coherence, etc.  Unfortunately, ICA adulteration of the original time points invalidatesthese non-volume conduction measures as much if not more than coherence.  

Best wishes,

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20170619/420346ac/attachment.html>

More information about the eeglablist mailing list