<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    <p>Hi Bob - <br>
    </p>
    <span
      style="font-size:12.0pt;line-height:107%;font-family:"Times
      New Roman",serif">It is my opinion that all EEG recordings
      are a mixture of artifact and brain activity.  The observation of
      consistent and reliable EEG results in many experiments reflects
      the robustness of EEG phenomena, in spite of the presence of
      artifacts, which vary from laborastory to laboratory and person to
      person. <br>
      <br>
      I think your are very mistaken to think the objective is to
      measure something relative phase very precisely when in fact the
      objective is to measure something robust in the face of a noisy
      recording situation. <br>
      <br>
      I'd recommend this discussion, if it continues, focus on
      robustness.   In my own work, I like to see that the effect of
      interest is present with and without ICA clean-up.    <br>
      <br>
      ramesh  <br>
    </span>
    <blockquote type="cite"
      cite="mid:474385387.1267677.1498323134968@mail.yahoo.com">
      <div style="font-family:Helvetica Neue, Helvetica, Arial,
        sans-serif;font-size:small;">
        <div>
          <p class="ydp71640538MsoNormal"><span
              style="font-size:12.0pt;line-height:107%;font-family:"Times
              New Roman",serif">There is another issue that arose
              that needs to be
              addressed, i.e., the belief by some that the original EEG
              recording is not
              valid to begin with and therefore this justifies
              distorting the phase
              differences.  </span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;">You stated: “Your
              phrase "distortion of the original time series" does make
              sense<span class="ydp71640538apple-converted-space"> </span></span>only
            if you believe that the
            original time series represents somehow a<span
              class="ydp71640538apple-converted-space"> </span>gold
            standard, something special.  In
            my view this is not<span
              class="ydp71640538apple-converted-space"> </span>justified,
            for instance because it disregards the fact that recording<span
              class="ydp71640538apple-converted-space"> </span>settings
            will influence how the
            data will be recorded, that is, they<span
              class="ydp71640538apple-converted-space"> </span>determine
            phase and amplitude!”</p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;">This is simply not
              true
              and the reason we know it is not true is that we use
              microvolt calibration sine
              waves at different frequencies and different amplitudes
              and different phase
              differences injected into the inputs of over 40 different
              amplifiers.   20
              microvolts in gives 20 microvolts out and a 30 degree
              phase shift in gives 30
              degree phase shift out, etc.  That is the
              amplifiers that we test as part of the FDA quality control
              process are
              excellent amplifiers that faithfully map one to one the
              input to the
              output.   Also, the vast majority of EEG
              amplifiers cannot be sold unless they past rigorous FDA
              quality control tests,
              for example, Cadwell, Nihon Khoden, Biosemi, Neuroscan,
              Nexus, Deymed, ANT,
              Nicolet.  None of these amplifiers
              distort phase differences or amplitudes.  
              If they did then FDA would fine the manufacturers.</span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;"> </span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;">You stated: “The
              second
              flaw in your reasoning is that you believe there are<span
                class="ydp71640538apple-converted-space"> </span></span>"artifact-free"
            intervals.<span class="ydp71640538apple-converted-space">”</span></p>
          <p class="ydp71640538MsoNormal"><span
              class="ydp71640538apple-converted-space"><span
                style="font-size: 12pt; line-height: 107%; font-family:
                "Times New Roman", serif; color: rgb(38, 40,
                42); background-image: initial; background-position:
                initial; background-size: initial; background-repeat:
                initial; background-attachment: initial;
                background-origin: initial; background-clip: initial;"> </span></span></p>
          <p class="ydp71640538MsoNormal"><span
              class="ydp71640538apple-converted-space"><span
                style="font-size: 12pt; line-height: 107%; font-family:
                "Times New Roman", serif; color: rgb(38, 40,
                42); background-image: initial; background-position:
                initial; background-size: initial; background-repeat:
                initial; background-attachment: initial;
                background-origin: initial; background-clip: initial;">The
                standard definition of artifact free are electrical
                potentials recorded from the scalp that were generated
                by the brain and only by
                the brain.   There have been millions of
                EEG recordings over the last 50 years and it is a false
                belief that 100% of EEG
                recordings are artifact and that the over 100,000 peer
                reviewed studies cited
                in the National Library of Medicine only involve
                artifact.</span></span></p>
          <p class="ydp71640538MsoNormal"><span
              class="ydp71640538apple-converted-space"><span
                style="font-size: 12pt; line-height: 107%; font-family:
                "Times New Roman", serif; color: rgb(38, 40,
                42); background-image: initial; background-position:
                initial; background-size: initial; background-repeat:
                initial; background-attachment: initial;
                background-origin: initial; background-clip: initial;">The
                skull does not generate electrical potentials and there
                is attenuation by the low conductivity but there is no
                fundamental change in
                the phase differences that are produced by synchronous
                synaptic potentials and
                various differences in white matter conduction
                velocities.  As the project manager at NIH for the
                development
                of the first 128 channel system there were occasions
                that we recorded EEG
                before and after a craniotomy.  We
                compared the ECoG electrical potentials recorded from
                the dura to those
                recorded from the scalp and there was a 20 to 50 times
                higher amplitude but the
                phase differences were essentially the same because they
                are produced by
                summated synaptic potentials and global cortico-cortical
                connectivity in the
                white matter.   </span></span></p>
          <p class="ydp71640538MsoNormal"><span
              class="ydp71640538apple-converted-space"><span
                style="font-size: 12pt; line-height: 107%; font-family:
                "Times New Roman", serif; color: rgb(38, 40,
                42); background-image: initial; background-position:
                initial; background-size: initial; background-repeat:
                initial; background-attachment: initial;
                background-origin: initial; background-clip: initial;"> </span></span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;">You stated: “In 
              your case, the average<span
                class="ydp71640538apple-converted-space"> </span></span>reference
            is far from zero because 19 10-20 channels, which are
            located<span class="ydp71640538apple-converted-space"> </span><span
style="font-size:12.0pt;line-height:107%;font-family:"Times New
              Roman",serif;color:#26282A"><br>
              <span style="background-image: initial;
                background-position: initial; background-size: initial;
                background-repeat: initial; background-attachment:
                initial; background-origin: initial; background-clip:
                initial;">on the top half of the head sphere, can hardly
                sum up to zero.”</span></span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;"> </span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;">This is correct and as
              Desmedt
              and others point out even 128 channel recordings do not
              produce a zero
              electrical field.   This would require
              placing electrodes on the bottom of the brain or skull or
              like Don Tucker jokes
              a valid average reference requires “Recording from the
              other side of the moon”.    This is one of the reasons
              that Desmedt and
              others argue against the use of an average reference and
              are in favor of a
              single common reference.</span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;"> </span></p>
          <p class="ydp71640538MsoNormal"><span style="font-size: 12pt;
              line-height: 107%; font-family: "Times New
              Roman", serif; color: rgb(38, 40, 42);
              background-image: initial; background-position: initial;
              background-size: initial; background-repeat: initial;
              background-attachment: initial; background-origin:
              initial; background-clip: initial;">Robert<br>
            </span></p>
          <!--EndFragment--></div>
        <div><br>
        </div>
        <div><br>
        </div>
        <div id="yahoo_quoted_8665233753" class="yahoo_quoted">
          <div style="font-family:'Helvetica Neue', Helvetica, Arial,
            sans-serif;font-size:13px;color:#26282a;">
            <div>On Saturday, June 24, 2017, 6:52:55 AM EDT, Stefan
              Debener <a class="moz-txt-link-rfc2396E" href="mailto:stefan.debener@uni-oldenburg.de"><stefan.debener@uni-oldenburg.de></a> wrote:</div>
            <div><br>
            </div>
            <div><br>
            </div>
            <div>
              <div dir="ltr">Dear Robert,<br clear="none">
                <br clear="none">
                Ok, I guess I have to give up on you. Of course you
                ascribe the <br clear="none">
                "original" times series some magic, and this is one of
                several flaws in <br clear="none">
                your reasoning. To cite your own published paper:<br
                  clear="none">
                <br clear="none">
                " PROBLEMS WITH RE-MONTAGING AND DISTORTIONS OF THE
                ORIGINAL TIME SERIES <br clear="none">
                If the original EEG/event-related potential (ERP) time
                series is <br clear="none">
                transformed into a second time series by using the
                average reference <br clear="none">
                then the original phase differences from three electrode
                locations may <br clear="none">
                be scrambled and lost. For example, with an average
                reference the entire <br clear="none">
                surface of the brain is not measured, thus the averaging
                does not create <br clear="none">
                a true zero potential at each instant of time."
                (citation taken from <br clear="none">
                <a shape="rect"
href="http://www.appliedneuroscience.com/Coh_phasediff&phase_resetinEEG-ERP.pdf"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Coh_phasediff&phase_resetinEEG-ERP.pdf</a>).<br
                  clear="none">
                <br clear="none">
                Your phrase "distortion of the original time series"
                does make sense <br clear="none">
                only if you believe that the original time series
                represents somehow a <br clear="none">
                gold standard, something special (something "magic",
                forgive my poor use <br clear="none">
                of English) that is magically close to the contributions
                from brain <br clear="none">
                generators to the surface-recorded signal. In my view
                this is not <br clear="none">
                justified, for instance because it disregards the fact
                that recording <br clear="none">
                settings will influence how the data will be recorded,
                that is, they <br clear="none">
                determine phase and amplitude! I argued before that
                attributing <br clear="none">
                something special to the "original" time series is
                highly misleading, <br clear="none">
                the "original" time series is not closer to the brain
                signal, in <br clear="none">
                contrast it may be pretty far away from it, not only
                because phase of <br clear="none">
                mixed brain generators does not make much sense, but
                also becasue of all <br clear="none">
                the artifactual influences not accounted for.  Now if
                you really believe <br clear="none">
                that one particular reference scheme is appropriate in
                bringing the <br clear="none">
                original data close to the brain while all others are
                rubbish, then I <br clear="none">
                wish you good luck in trying to convince the community.
                I predict that <br clear="none">
                nobody will take you serious. The reference discussion
                has been going on <br clear="none">
                for ages, and there are good reasons to change the
                reference (online or <br clear="none">
                offline) depending on what the purpose of the
                study/analysis is.<br clear="none">
                <br clear="none">
                I argue that any post-recording signal processing that
                changes the <br clear="none">
                morphology of a time series will change the phase values
                as well, NOT <br clear="none">
                just the average reference, and NOT just ICA. In  your
                case, the average <br clear="none">
                reference is far from zero because 19 10-20 channels,
                which are located <br clear="none">
                on the top half of the head sphere, can hardly sum up to
                zero. Only a <br clear="none">
                full equidistant spatial coverage of the head sphere
                would make the <br clear="none">
                spatial average approximate zero potential. Check this
                paper for <br clear="none">
                in-depth discussion on the average reference and the
                bias it introduces: <br clear="none">
                <a shape="rect"
                  href="https://www.ncbi.nlm.nih.gov/pubmed/10402104"
                  target="_blank" moz-do-not-send="true">https://www.ncbi.nlm.nih.gov/pubmed/10402104</a>).
                If you don't like the <br clear="none">
                average reference, fair enough, then simply take any
                other reference, <br clear="none">
                but you will observe the same "distortion" on phase
                values. By the way, <br clear="none">
                non-invasive EEG does not measure signals from the
                brain, there is a <br clear="none">
                skull in between and a couple other layers, all with
                different <br clear="none">
                conductivies...just another reason to be aware of the
                inverse problem <br clear="none">
                and trat signals originally recorded from the scalp with
                great care.<br clear="none">
                <br clear="none">
                The second flaw in your reasoning is that you believe
                there are <br clear="none">
                "artifact-free" intervals. The report I included showed
                (far from <br clear="none">
                perfect, but clearly evident!) heart-electrical
                activity. Recording <br clear="none">
                parameters strongly determine whether such "EKG"
                contributions show up <br clear="none">
                in ICA decompositions or not (such as sub-10/20 spatial
                coverage). Now, <br clear="none">
                only because you don't see heart-electrical activity in
                your "original" <br clear="none">
                19-ch 10/20 recordings, can you seriously claim that the
                heart of your <br clear="none">
                participant was not beating during recording? Of course
                not! All it says <br clear="none">
                is that the influcene may be stronger or weaker
                represented in your <br clear="none">
                recordings (depending on individual differences, and,
                again, recording <br clear="none">
                parameters). With your philosophy, you gonna miss it if
                it does not jump <br clear="none">
                into your eyes.<br clear="none">
                <br clear="none">
                Anyway, I hope others join the discussion, I am giving
                up.<br clear="none">
                <br clear="none">
                Best,<br clear="none">
                Stefan<br clear="none">
                <br clear="none">
                <br clear="none">
                Am 23.06.17 um 19:16 schrieb Robert Thatcher:<br
                  clear="none">
                ><br clear="none">
                > Dear Stefan,<br clear="none">
                ><br clear="none">
                >    Thank you for your dilligence and dedication to
                this important <br clear="none">
                > issue.  I am pleased that you are in agreement with
                myself and <br clear="none">
                > Georges and Gert and scienific publications that 
                "a spatial filter <br clear="none">
                > operation  such as ICA or other, the phase
                differences may indeed be <br clear="none">
                > different."  I would add that we have not yet found
                an instance where <br clear="none">
                > ICA reconstruction did not alter phase differences
                between channel and <br clear="none">
                > there I would change the word "may" and phrase to
                "are indeed different".<br clear="none">
                ><br clear="none">
                > You sated:  "As a toy example I include the common
                average reference. "<br clear="none">
                ><br clear="none">
                > In Neuroguide we do not allow one to compute phase
                diffferences or <br clear="none">
                > coherence when using a common reference.  We found
                with simulation and <br clear="none">
                > real data that the common reference mixes the
                phases differences <br clear="none">
                > between all channels and itself distorts phase
                differences and often <br clear="none">
                > in strange ways, for example, if one or two
                channels has a suddent <br clear="none">
                > large amplitude alpha or theta rhythm then the
                phase differences <br clear="none">
                > between channels that do not have a alpha or beta
                or theta rhythm are <br clear="none">
                > altered.  If one uses a single common reference
                then if an alpha <br clear="none">
                > rhythm appeas for example in O1/2 then there is not
                change in phase <br clear="none">
                > differences in channels where there is no alpha. 
                We also compared two <br clear="none">
                > sine waves at different phase differences, e.g., 30
                deg, 60 deg, 90 <br clear="none">
                > deg, etc and mixed different amounts of white noise
                in one of the <br clear="none">
                > channels we found linear reductions in coherence
                (i.e., the phase <br clear="none">
                > stability over time) as a function of the SNR and
                the mean phase <br clear="none">
                > differences were stable until the noise was too
                high and when <br clear="none">
                > coherencer was near zero.  In contrast, uses the
                averge refence and <br clear="none">
                > repeats the same signal and noise test with
                different phase <br clear="none">
                > differences then the mean phase difference is
                quickly lost and there <br clear="none">
                > is no valid measure of coherence.  Here is a url to
                a publication <br clear="none">
                > that discusses this topic: <br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Coh_phasediff&phase_resetinEEG-ERP.pdf"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Coh_phasediff&phase_resetinEEG-ERP.pdf</a><br
                  clear="none">
                ><br clear="none">
                > You stated:  "Your claim that ICA has somehow
                corrupted the data such <br clear="none">
                > that previously super reliable clinical effects all
                over a sudden <br clear="none">
                > vanished is not convincing either."<br clear="none">
                ><br clear="none">
                > I never said this. There is no ICA reconstruction
                in NeuroGuide and <br clear="none">
                > the over 3,000 users of Neuroguide have not
                complained about coherence <br clear="none">
                > or phase and they always obtain repeatable measures
                when the retest <br clear="none">
                > patients.  It was only the WinEEG users that are
                worried about ICA <br clear="none">
                > and they never said that they got "super reliable
                clinical effects". <br clear="none">
                > They just noticed that coherence using the WinEEG
                after ICA <br clear="none">
                > reconstruction was totally different than when they
                use NxLink or SKIL <br clear="none">
                > or Neurorep or Neuroguide or Brindx or other
                software, etc.<br clear="none">
                ><br clear="none">
                > You stated:  "Now which phase values are valid,
                those obtained by one <br clear="none">
                > particular reference scheme or those by another? In
                my view they are <br clear="none">
                > both arbitraty"<br clear="none">
                ><br clear="none">
                > See my earlier reply regarding average references
                and the Laplacian <br clear="none">
                > reference in regard to the accurate and
                reproducible measures of phase <br clear="none">
                > differences as opposed to using a single common
                reference.  I am <br clear="none">
                > recopying the link here: <br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Coh_phasediff&phase_resetinEEG-ERP.pdf%20"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Coh_phasediff&phase_resetinEEG-ERP.pdf
                </a><br clear="none">
                > here is another review on this topic: <br
                  clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Brain%20Connectivity-A%20Tutorial.pdf"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Brain%20Connectivity-A%20Tutorial.pdf</a><br
                  clear="none">
                ><br clear="none">
                > You stated:  “there is no such magically clean raw
                brainsignal <br clear="none">
                > available in the first place!”<br clear="none">
                ><br clear="none">
                > No one says that “magic” is involved in the EEG. 
                However, the physics <br clear="none">
                > of EEG is involved and during a 5 minute to 20
                minute EEG recording <br clear="none">
                > there is plenty of artifact free data.  There are
                over 100,000 QEEG <br clear="none">
                > publications in the National Library of Medicine
                with high effect <br clear="none">
                > sizes and high test retest reliability and highly
                reproducible <br clear="none">
                > findings.  My concern and the concern of many
                others is that ICA <br clear="none">
                > reconstruction alters phase differences in an
                entire EEG recording <br clear="none">
                > even if there are only a few instances of
                artifact.  The reason that <br clear="none">
                > QEEG has been so successful and so widely used
                since the late 1950s is <br clear="none">
                > because people have been successful in avoiding or
                deleting artifact <br clear="none">
                > and selecting multiple artifact free parts of the
                record and achieving <br clear="none">
                > 0.95 or higher test retest reliability.  Over
                reaction to the <br clear="none">
                > presence of small amounts of artifact is not a
                justification for <br clear="none">
                > altering the electricity of the brain including
                network dynamics such <br clear="none">
                > as average synaptic rise times and conduction
                velocities and couplings <br clear="none">
                > between groups of neurons.<br clear="none">
                ><br clear="none">
                > You stated: “Artifacts not accounted for adulterate
                EEG phase values”<br clear="none">
                ><br clear="none">
                > We agree on this but this is not what the
                discussion is about.  The <br clear="none">
                > concerns of many is that the phase differences in
                the artifact free <br clear="none">
                > sections are altered.  Rarely if ever do
                clinicians/scientists bother <br clear="none">
                > computing the phase differences during an eye
                movement artifact.  <br clear="none">
                > Usually phase differences are zero and this physics
                fact plus the <br clear="none">
                > electrical gradients from the eyes is how many
                people detect eye <br clear="none">
                > movement artifact and then omit the artifact from
                analyses without <br clear="none">
                > using ICA.  The main focus in QEEG is the parts of
                the recording <br clear="none">
                > where there is no artifact and the electrical
                potentials are generated <br clear="none">
                > by the brain inside the skull.<br clear="none">
                ><br clear="none">
                > Also, thank you for your use of the Hilbert
                transform, this is only of <br clear="none">
                > the tools that we use and it allows one to evaluate
                phase differences <br clear="none">
                > in every individual time sample in the artifact
                free sections and <br clear="none">
                > prove that each and every one of the phase
                differences for all <br clear="none">
                >  channel combinations is altered by ICA
                reconstruction.  You may be <br clear="none">
                > interested that we use the Hilbert transform to
                measure phase shift <br clear="none">
                > and phase lock duration that have a high
                correlation with Autism and <br clear="none">
                > intelligence in short distance connections and with
                use the Hilbert <br clear="none">
                > transform to measure the magnitude of information
                flow (phase slope <br clear="none">
                > index) and intelligence.  Here are some hyperlinks
                to these studies: <br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Intelligence-phase_reset_Nature.pdf"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Intelligence-phase_reset_Nature.pdf</a><br
                  clear="none">
                ><br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Autism%20Thatcher%20et%20al.pdf"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Autism%20Thatcher%20et%20al.pdf</a><br
                  clear="none">
                ><br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Default_Network_LORETA_Phase_Reset-Thatcher_et_al.pdf"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Default_Network_LORETA_Phase_Reset-Thatcher_et_al.pdf</a><br
                  clear="none">
                ><br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Intelligence%20&%20information%20flow-Thatcher%20et%20al%202016.pdf"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Intelligence%20&%20information%20flow-Thatcher%20et%20al%202016.pdf</a><br
                  clear="none">
                ><br clear="none">
                > We are also using the Hilbert transforms for
                cross-frequency network <br clear="none">
                > dynamics including phase-amplitude coupling.  I do
                not believe that <br clear="none">
                > we would have discovered these important network
                correlations if we <br clear="none">
                > had used ICA reconstruction.<br clear="none">
                ><br clear="none">
                > Robert<br clear="none">
                ><br clear="none">
                ><br clear="none">
                > On Friday, June 23, 2017, 9:53:01 AM EDT, Stefan
                Debener <br clear="none">
                > <<a shape="rect"
                  ymailto="mailto:stefan.debener@uni-oldenburg.de"
                  href="mailto:stefan.debener@uni-oldenburg.de"
                  moz-do-not-send="true">stefan.debener@uni-oldenburg.de</a>>
                wrote:<br clear="none">
                ><br clear="none">
                ><br clear="none">
                > Dear Robert,<br clear="none">
                ><br clear="none">
                > I have expanded my illustration and now consider
                the phase differences<br clear="none">
                > between two channels, slides 13 to 16 of the
                updated pdf:<br clear="none">
                > <a shape="rect"
href="https://www.dropbox.com/s/e70qhf91dgc5anu/Thatcher_summary_2.pdf?dl=0"
                  target="_blank" moz-do-not-send="true">https://www.dropbox.com/s/e70qhf91dgc5anu/Thatcher_summary_2.pdf?dl=0</a><br
                  clear="none">
                ><br clear="none">
                > Note that phase values were derived by the Hilbert
                transform of the<br clear="none">
                > bandpass filtered signal, as explained by W Freeman
                here:<br clear="none">
                > <a shape="rect"
href="http://www.scholarpedia.org/article/Hilbert_transform_for_brain_waves"
                  target="_blank" moz-do-not-send="true">http://www.scholarpedia.org/article/Hilbert_transform_for_brain_waves</a><br
                  clear="none">
                ><br clear="none">
                > More details on the particular implementation I
                used are here:<br clear="none">
                > <a shape="rect"
                  href="https://de.mathworks.com/help/signal/ref/hilbert.html"
                  target="_blank" moz-do-not-send="true">https://de.mathworks.com/help/signal/ref/hilbert.html</a><br
                  clear="none">
                ><br clear="none">
                > If you measure phase differences between two
                channels, consider the<br clear="none">
                > result as your gold standard, and then apply a
                spatial filter operation<br clear="none">
                > such as ICA or other, the phase differences may
                indeed be different. I<br clear="none">
                > assume any spatial filter (that effectively
                spatially filters the data)<br clear="none">
                > changes phase values and phase difference values.
                As a toy example I<br clear="none">
                > include the common average reference. If you apply
                a common average<br clear="none">
                > reference to the raw data, then bandpass filter as
                before, and compare<br clear="none">
                > the phase difference values to your "gold
                standard", then the phase<br clear="none">
                > differences will change as well. Now which phase
                values are valid, those<br clear="none">
                > obtained by one particular reference scheme or
                those by another? In my<br clear="none">
                > view they are both arbitraty, since recording
                settings as well as<br clear="none">
                > preprocessing steps may have a strong impact on the
                actually measured<br clear="none">
                > phase. There is no reason to assume that a change
                in phase, or in phase<br clear="none">
                > differences, "adulterates" a magically clean phase
                signal obtained from<br clear="none">
                > the raw data - simply because there is no such
                magically clean raw brain<br clear="none">
                > signal available in the first place!<br
                  clear="none">
                ><br clear="none">
                > Your claim that ICA has somehow corrupted the data
                such that previously<br clear="none">
                > super reliable clinical effects all over a sudden
                vanished is not<br clear="none">
                > convincing either. Artifacts such as eye blinks and
                lateral eye<br clear="none">
                > movements are very common, I hope you can agree at
                least here. Now, keep<br clear="none">
                > in mind that they contribute fixed spatial
                patterns  - as long as the<br clear="none">
                > electrodes cap does not shift during acquisition
                the projections of the<br clear="none">
                > sources of those artifacts do not change. My
                illustrations above show<br clear="none">
                > very clearly how artifacs indeed adulterate phase
                values, just as Arnos<br clear="none">
                > illustrations do! Now, if you disregard artifactual
                influences you may<br clear="none">
                > end up with highly reliable connectivity effects -
                but they tell you<br clear="none">
                > very little about brain function! Even more
                troubling, if you compare<br clear="none">
                > two individuals EEGs (say, one "healthy", one
                "abnormal"), then a<br clear="none">
                > different amount of artifacts in the data, if not
                carefully taken care<br clear="none">
                > of during preprocessing, will produce spurious
                results that are falsely<br clear="none">
                > attributed to differences in brain function.
                Actually, given that many<br clear="none">
                > artifacts often contribute much more variance to
                that raw signals than<br clear="none">
                > (reasonably well validated) brain signals, such as
                fronto-midline theta,<br clear="none">
                > this is actually very likely! So, what we learn is
                that:<br clear="none">
                ><br clear="none">
                > Artifacts not accounted for adulterate EEG phase
                values<br clear="none">
                ><br clear="none">
                > Best,<br clear="none">
                ><br clear="none">
                > Stefan<br clear="none">
                ><br clear="none">
                ><br clear="none">
                ><br clear="none">
                > Am 22.06.17 um 20:30 schrieb Robert Thatcher:<br
                  clear="none">
                > > Dear Stefan,<br clear="none">
                > >    The attachment did not contain any measures
                of phase differences<br clear="none">
                > > between channels.  It is very difficult to
                visually see differences<br clear="none">
                > > in phase differences.  One must use the
                cross-spectrum to calculate<br clear="none">
                > > phase differences and compare phase
                differences in degrees.  Phase<br clear="none">
                > > difference varies from -180 to 180 degrees and
                one must look at the<br clear="none">
                > > numbers.  Below is a url to the two power
                points that also show<br clear="none">
                > > visually similar EEG tracings but also
                computed the instantaneous<br clear="none">
                > > phase differences using the Hilbert transform
                (complex demodulation).<br clear="none">
                > >  Four identical time points were selected and
                they demonstrated<br clear="none">
                > > totally different phase differences with
                respect to the O1 channel and<br clear="none">
                > > the other 18 channels.  No matter what
                reference channel one selects<br clear="none">
                > > and no matter what identical time points one
                selects there are always<br clear="none">
                > > large differences in the phase difference
                between channels in all<br clear="none">
                > > frequency bands.  I also computed the average
                phase difference in the<br clear="none">
                > > artifact free parts of the record and the
                averages were statistically<br clear="none">
                > > significantly different at P < 0.0001 and
                the same for the FFT.<br clear="none">
                > ><br clear="none">
                > > Proof of phase difference adulteration is in
                the power points.  I am<br clear="none">
                > > again copying the hyperlink here:<br
                  clear="none">
                > ><br clear="none">
                > ><br clear="none">
                > > <br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Phase_Diff-Original_&_Delorme-Post-ICA-4_time_points.zip"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Phase_Diff-Original_&_Delorme-Post-ICA-4_time_points.zip</a><br
                  clear="none">
                > ><br clear="none">
                > ><br clear="none">
                > > This cannot be explained by a low quality ICA
                reconstruction because<br clear="none">
                > > the ICA reconstruction was conducted by Arnu
                using EEGLab software.<br clear="none">
                > ><br clear="none">
                > > Robert<br clear="none">
                > > On Thursday, June 22, 2017, 2:00:19 PM EDT,
                Stefan Debener<br clear="none">
                > > <<a shape="rect"
                  ymailto="mailto:stefan.debener@uni-oldenburg.de"
                  href="mailto:stefan.debener@uni-oldenburg.de"
                  moz-do-not-send="true">stefan.debener@uni-oldenburg.de</a>
                <br clear="none">
                > <mailto:<a shape="rect"
                  ymailto="mailto:stefan.debener@uni-oldenburg.de"
                  href="mailto:stefan.debener@uni-oldenburg.de"
                  moz-do-not-send="true">stefan.debener@uni-oldenburg.de</a>>>
                wrote:<br clear="none">
                > ><br clear="none">
                > ><br clear="none">
                > > Dear Robert,<br clear="none">
                > ><br clear="none">
                > > I looked up some own data and find absolutely
                no evidence in favour of<br clear="none">
                > > your ICA phase adulteration claim, see the
                attached pdf report. I guess<br clear="none">
                > > you simply used a poor ICA implementation,
                and/or a poor component<br clear="none">
                > > selection. The attached example is in full
                accordance with Arnos reply,<br clear="none">
                > > with the difference that I zoom into a clearly
                visibile alpha<br clear="none">
                > > oscillation, to have a reference brain signal.
                The example shows no<br clear="none">
                > > evidence that occipital alpha phase is biased
                by ICA eye blink<br clear="none">
                > > correction. This is a very typical example and
                based on a quick and<br clear="none">
                > > dirty ICA decomposition, nothing fancy, to
                keep this demo simple. Better<br clear="none">
                > > preprocessing and component selection would
                easily further improve the<br clear="none">
                > > signal quality.<br clear="none">
                > ><br clear="none">
                > > Best,<br clear="none">
                > ><br clear="none">
                > > Stefan<br clear="none">
                > ><br clear="none">
                > ><br clear="none">
                > ><br clear="none">
                > > Am 20.06.17 um 19:53 schrieb Robert Thatcher:<br
                  clear="none">
                > > ><br clear="none">
                > > > Dear Arno,<br clear="none">
                > > ><br clear="none">
                > > > 1)*On Phase Differences in the Original
                vs the Delorme ICA<br clear="none">
                > > > Reconstruction: *We can agree or disagree
                about whether or not some<br clear="none">
                > > > small eye movement artifact was in the
                hand selection that I did.  But<br clear="none">
                > > > that misses the main point here.  That is
                the ICA reconstruction<br clear="none">
                > > > alters each and every data point in the
                entire record including all<br clear="none">
                > > > artifact free portions no matter what one
                selects. For example, the<br clear="none">
                > > > record is 6 minutes and 51 seconds = 411
                seconds. The Mitsar sample<br clear="none">
                > > > rate was 250 samples per second = 102,750
                data samples. Phase<br clear="none">
                > > > difference for each frequency band for
                each and every one of the<br clear="none">
                > > > 102,750 data samples has been altered by
                your own ICA reconstruction<br clear="none">
                > > > in the EDF file that you emailed to me.
                Unless you were to sit next to<br clear="none">
                > > > me or if we do a Team Viewer it is not
                possible for me to demonstrate<br clear="none">
                > > > this for all of the data points and then
                create a power point for all<br clear="none">
                > > > of these data samples.  However, I can
                show some exemplars, for<br clear="none">
                > > > example, I have created two figures at 4
                different time points (1 sec;<br clear="none">
                > > > 2:27 sec; 42 sec & 5:49 sec) that you
                can download. You can extract<br clear="none">
                > > > each screen capture and expand them so
                that you can see that the exact<br clear="none">
                > > > same time points were selected and the
                Hilbert transform JTFA for the<br clear="none">
                > > > 4 time points resulted in different phase
                differences in all channel<br clear="none">
                > > > combinations with respect to O1 for all
                frequencies.  The same is true<br clear="none">
                > > > no matter which channel is selected to
                compute the phase differences<br clear="none">
                > > > in degrees.  The same is true also if one
                computes averages of the<br clear="none">
                > > > instantaneous phase differences or if one
                uses the FFT.  Here is the<br clear="none">
                > > > download URL:<br clear="none">
                > > ><br clear="none">
                > > ><br clear="none">
                > > <br clear="none">
                > <a shape="rect"
href="http://www.appliedneuroscience.com/Phase_Diff-Original_&_Delorme-Post-ICA-4_time_points.zip"
                  target="_blank" moz-do-not-send="true">http://www.appliedneuroscience.com/Phase_Diff-Original_&_Delorme-Post-ICA-4_time_points.zip</a><br
                  clear="none">
                > > ><br clear="none">
                > > ><br clear="none">
                > > > 2)*On the WinEEG ICA Reconstruction: *I
                agree that having access to<br clear="none">
                > > > ICA components themselves and the
                topography is critical in<br clear="none">
                > > > understanding exactly what the WinEEG
                software did. Unfortunately, I<br clear="none">
                > > > personally do not have access to the
                WinEEG software.<br clear="none">
                > > > Clinician/Scientists in Australia use the
                WinEEG software and they<br clear="none">
                > > > were the ones that expressed concern
                about phase difference distortion<br clear="none">
                > > > at a workshop in Adelaide and gave me the
                original and the WinEEG ICA<br clear="none">
                > > > eye movement corrected files in EDF
                format. They explained that they<br clear="none">
                > > > removed only one ICA component for eye
                movement before they<br clear="none">
                > > > reconstructed a new time series.  At
                first, I was impressed because<br clear="none">
                > > > the eye movements were absent in the
                reconstructed time series.  I<br clear="none">
                > > > then was able to use JTFA (Hilbert
                transform) to compare the two edf<br clear="none">
                > > > files and discovered that all of the
                phase differences for all<br clear="none">
                > > > channels for all frequencies had been
                altered by the ICA<br clear="none">
                > > > reconstruction including artifact free
                periods.  I could demonstrate<br clear="none">
                > > > this by individual time comparisons or
                averages of instantaneous phase<br clear="none">
                > > > differences or by the FFT.  A user of
                WinEEG explained that they do<br clear="none">
                > > > not throw away the original raw digital
                data, however I was told that<br clear="none">
                > > > they believe that the ICA reconstructed
                times series is artifact free<br clear="none">
                > > > and therefore they compute means and
                standard deviations for their<br clear="none">
                > > > normative database using the ICA
                reconstructed data and not the hand<br clear="none">
                > > > edited or artifact deleted original data
                samples like other commercial<br clear="none">
                > > > companies do.  Your ICA reconstructed
                time series is actually less<br clear="none">
                > > > different than the original phase
                difference in comparison to the<br clear="none">
                > > > WinEEG ICA.  Nonetheless, both your ICA
                reconstruction and the WinEEG<br clear="none">
                > > > reconstructions are significantly
                different than the original <br clear="none">
                > recording.<br clear="none">
                > > ><br clear="none">
                > > > Best regards,<br clear="none">
                > > ><br clear="none">
                > > > Robert<br clear="none">
                > > ><br clear="none">
                > > > Cp���<br clear="none">
                > > ><br clear="none">
                > > ><br clear="none">
                > > > On Tuesday, June 20, 2017, 1:12:41 AM
                EDT, Arnaud Delorme<br clear="none">
                > > > <<a shape="rect"
                  ymailto="mailto:arno@ucsd.edu"
                  href="mailto:arno@ucsd.edu" moz-do-not-send="true">arno@ucsd.edu</a>
                <mailto:<a shape="rect"
                  ymailto="mailto:arno@ucsd.edu"
                  href="mailto:arno@ucsd.edu" moz-do-not-send="true">arno@ucsd.edu</a>>
                <mailto:<a shape="rect"
                  ymailto="mailto:arno@ucsd.edu"
                  href="mailto:arno@ucsd.edu" moz-do-not-send="true">arno@ucsd.edu</a>
                <br clear="none">
                > <mailto:<a shape="rect"
                  ymailto="mailto:arno@ucsd.edu"
                  href="mailto:arno@ucsd.edu" moz-do-not-send="true">arno@ucsd.edu</a>>>>
                wrote:<br clear="none">
                > > ><br clear="none">
                > > ><br clear="none">
                > > > Dear Robert,<br clear="none">
                > > ><br clear="none">
                > > > 1) *On my ICA decomposition analysis on
                your data.* You have selected<br clear="none">
                > > > a subset of the file where there is 1
                minute and 41 second data of eye<br clear="none">
                > > > free data. I was only able to select 40
                seconds in the same file, and<br clear="none">
                > > > I also showed that even in this short
                file, there was some residual<br clear="none">
                > > > eye movements. Jason and Stefan agreed
                with me. This is the reason why<br clear="none">
                > > > ICA components power spectrum over
                frontal channels (and frontal<br clear="none">
                > > > channels only) was affected below 10 Hz
                frequency band in my data<br clear="none">
                > > > analysis. So on my ICA decomposition, our
                disagreement comes from the<br clear="none">
                > > > interpretation. You feel that the power
                we remove at low frequency in<br clear="none">
                > > > frontal channel is not eye movement. In
                an attempt to convince you, I<br clear="none">
                > > > have picked up a clean region from your
                EDF dataset, and did some<br clear="none">
                > > > dipole localization at this latency. We
                see that in the clean data,<br clear="none">
                > > > the best dipolar fit (with 2 symmetrical
                dipoles) ends up near the eye<br clear="none">
                > > > balls with a residual variance of 6.9%.
                Hopefully this convinces you<br clear="none">
                > > > that your data is not free of eye
                movement artifacts. If you are<br clear="none">
                > > > willing to take a step further you might
                contemplate the idea that ICA<br clear="none">
                > > > can remove this residual spurious
                activity.<br clear="none">
                > > ><br clear="none">
                > > > 2) *On the WinEEG ICA decomposition
                analysis.* It is critical for us<br clear="none">
                > > > to see the scalp topography (and if
                possible continuous activity) of<br clear="none">
                > > > the components the people at the
                Australia workshop selected. Without<br clear="none">
                > > > this, it is not possible for us to
                comment on the cleaned data. I<br clear="none">
                > > > agree with you that there was some phase
                distortion in alpha (visible<br clear="none">
                > > > directly in the raw data in the first
                email you sent) and that this<br clear="none">
                > > > should not be the case. However, without
                seing the ICA decomposition,<br clear="none">
                > > > it is not possible for us to conclude as
                to wether people selected the<br clear="none">
                > > > wrong ICA components or if the ICA
                decomposition implemented in this<br clear="none">
                > > > software is buggy (ICA is not a simple
                algorithm and it is sensitive<br clear="none">
                > > > to numerical imprecision and a lot of
                other parameters - a suboptimal<br clear="none">
                > > > implementation could easily explain the
                WinEEG results). Also, you<br clear="none">
                > > > seem to imply that the WinEEG people were
                running ICA on their data<br clear="none">
                > > > then throwing away the raw data (which is
                why their ICA biased<br clear="none">
                > > > neurofeedback database is useless for
                practical purposes). Is that<br clear="none">
                > > > correct? One should never throw away the
                raw data. If they did throw<br clear="none">
                > > > away the raw data, it is an indication
                that the WinEEG are not<br clear="none">
                > > > rigorous in their approach and therefore
                might not have implemented<br clear="none">
                > > > ICA in an optimal way. If it is not the
                case, one may easily<br clear="none">
                > > > reconstruct the database of measures with
                or without ICA decomposition<br clear="none">
                > > > (assuming ICA is done right which does
                not seem to be the case) then<br clear="none">
                > > > assess data measure distoritions (power,
                phase index, etc…) in a<br clear="none">
                > > > statistical fashion.<br clear="none">
                > > ><br clear="none">
                > > > Best wishes,<br clear="none">
                > > ><br clear="none">
                > > > Arno<br clear="none">
                > > ><br clear="none">
                > > > <a shape="rect"
href="http://sccn.ucsd.edu/%7Earno/download/clean_edf_file_analysis2.pdf%20"
                  target="_blank" moz-do-not-send="true">http://sccn.ucsd.edu/~arno/download/clean_edf_file_analysis2.pdf
                </a><br clear="none">
                > <<a shape="rect"
href="http://sccn.ucsd.edu/%7Earno/download/clean_edf_file_analysis2.pdf%20"
                  target="_blank" moz-do-not-send="true">http://sccn.ucsd.edu/%7Earno/download/clean_edf_file_analysis2.pdf%20</a>><br
                  clear="none">
                > > <<a shape="rect"
href="http://sccn.ucsd.edu/%7Earno/download/clean_edf_file_analysis2.pdf%20"
                  target="_blank" moz-do-not-send="true">http://sccn.ucsd.edu/%7Earno/download/clean_edf_file_analysis2.pdf%20</a>><br
                  clear="none">
                > > > <<a shape="rect"
href="http://sccn.ucsd.edu/%7Earno/download/clean_edf_file_analysis2.pdf"
                  target="_blank" moz-do-not-send="true">http://sccn.ucsd.edu/%7Earno/download/clean_edf_file_analysis2.pdf</a>><br
                  clear="none">
                > > ><br clear="none">
                > > >> On Jun 18, 2017, at 11:44 AM, Robert
                Thatcher<br clear="none">
                > > <<a shape="rect"
                  ymailto="mailto:rwthatcher2@yahoo.com"
                  href="mailto:rwthatcher2@yahoo.com"
                  moz-do-not-send="true">rwthatcher2@yahoo.com</a>
                <mailto:<a shape="rect"
                  ymailto="mailto:rwthatcher2@yahoo.com"
                  href="mailto:rwthatcher2@yahoo.com"
                  moz-do-not-send="true">rwthatcher2@yahoo.com</a>> <br
                  clear="none">
                > <mailto:<a shape="rect"
                  ymailto="mailto:rwthatcher2@yahoo.com"
                  href="mailto:rwthatcher2@yahoo.com"
                  moz-do-not-send="true">rwthatcher2@yahoo.com</a>
                <mailto:<a shape="rect"
                  ymailto="mailto:rwthatcher2@yahoo.com"
                  href="mailto:rwthatcher2@yahoo.com"
                  moz-do-not-send="true">rwthatcher2@yahoo.com</a>>>
                <div class="yqt6084624675" id="yqtfd48737"><br
                    clear="none">
                  > > >> <mailto:<a shape="rect"
                    ymailto="mailto:rwthatcher2@yahoo.com"
                    href="mailto:rwthatcher2@yahoo.com"
                    moz-do-not-send="true">rwthatcher2@yahoo.com</a>
                  <mailto:<a shape="rect"
                    ymailto="mailto:rwthatcher2@yahoo.com"
                    href="mailto:rwthatcher2@yahoo.com"
                    moz-do-not-send="true">rwthatcher2@yahoo.com</a>>
                  <br clear="none">
                  > <mailto:<a shape="rect"
                    ymailto="mailto:rwthatcher2@yahoo.com"
                    href="mailto:rwthatcher2@yahoo.com"
                    moz-do-not-send="true">rwthatcher2@yahoo.com</a>
                  <mailto:<a shape="rect"
                    ymailto="mailto:rwthatcher2@yahoo.com"
                    href="mailto:rwthatcher2@yahoo.com"
                    moz-do-not-send="true">rwthatcher2@yahoo.com</a>>>>>
                  wrote:<br clear="none">
                  ><br clear="none">
                  > ><br clear="none">
                  > > >><br clear="none">
                  > > >> <Pre-ICA-Hand Artifact free
                  selections.edf><br clear="none">
                  > ><br clear="none">
                  > > ><br clear="none">
                  > > ><br clear="none">
                  > > ><br clear="none">
                  > > > Dieser Nachrichteninhalt wird auf
                  Anfrage komplett heruntergeladen.<br clear="none">
                  > ><br clear="none">
                  > ><br clear="none">
                  ><br clear="none">
                  <br clear="none">
                </div>
              </div>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
  </body>
</html>