<html>
  <head>
    <meta content="text/html; charset=windows-1252"
      http-equiv="Content-Type">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">Hi Makoto, <br>
      <br>
      Just to clarify, while the process of developing my method does
      use inputs from the ICLabel website, the final product that people
      will use will <b><i>only</i></b> look at the EEG measures and
      nothing from the website. Just like the other methods, it will not
      require people to actively contribute labels to work once
      released. The label collection is only for the development of the
      classier.<br>
      <br>
      Luca<br>
      <br>
      On 06/30/2017 05:52 PM, Makoto Miyakoshi wrote:<br>
    </div>
    <blockquote
cite="mid:CAEqC+SXVK48QjNJ5KS3z09KLz44XBgosLjOEn=ewTJM7chfs-A@mail.gmail.com"
      type="cite">
      <meta http-equiv="Content-Type" content="text/html;
        charset=windows-1252">
      <div dir="ltr">Dear Marius,
        <div><br>
        </div>
        <div>Thank you for your opinion.</div>
        <div><br>
        </div>
        <div>My colleague Luca has been working on automatic IC labeling
          using machine learning algorithm: but it still uses <i>user
            input </i>as the data to be tranied. So it is NOT a
          solution to develop a better or perfect algorithm that judges
          what is what based on EEG measures (he tested all the kinds of
          algorithms available at the timepoint of last year, and found
          nothing was perfect.) See this page for his data collection
          scheme (which is also a educational tool)</div>
        <div><br>
        </div>
        <div><a moz-do-not-send="true"
            href="http://reaching.ucsd.edu:8000/auth/login">http://reaching.ucsd.edu:8000/auth/login</a><br>
        </div>
        <div><br>
        </div>
        <div>I have been collaborating with radiologists and
          neurologists. They diagnose patients, particularly
          neurologists even determines which brain tissue to remove.
          What algorithm do they use? They use eyeballs. After all,
          humans are still the best learning machine today (though it is
          very tempting to make my own criteria for IC labeling for
          non-aggressive data cleaning)</div>
        <div><br>
        </div>
        <div>Makoto</div>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On Fri, Jun 30, 2017 at 12:54 AM,
          Marius Klug <span dir="ltr"><<a moz-do-not-send="true"
              href="mailto:marius.s.klug@gmail.com" target="_blank">marius.s.klug@gmail.com</a>></span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div dir="ltr">
              <div>
                <div>
                  <div>
                    <div>Hi Gian, Makoto, list<br>
                      <br>
                    </div>
                    the quest for more reproducibility in EEG is one of
                    my personal topics of interest, so here are a few
                    comments from my point of view: I too would consider
                    an automated method for IC classification a valuable
                    tool, and dismissing it is not as easy as Makoto
                    suggests if you ask me. But I agree with Makoto:
                    there are no perfect solutions and manual inspection
                    to date is hard to get around since experienced
                    inspectors are just better than the algorithms.
                    However the reasoning that it's not reproducible is
                    valid, and also maybe a trained eye is better than
                    automated methods, but a novice might not exactly
                    know where to look at and the quality may suffer...<br>
                    <br>
                    You didn't specify if you just want to reject
                    eye-movement and other artifacts or want to go the
                    other way round and just keep ICs that are
                    specifically generated by the brain. In the latter
                    case you might want to try out a few things, because
                    there are some that work okay-ish: <br>
                    First, you can use dipole fitting and check the
                    residual variance for each IC, and have a threshold
                    of <0.15 (typically) for brain ICs. This is not a
                    fail-safe method, however, since eye-ICs and
                    sometimes also very narrow (not spread around the
                    whole head) muscle or other artifactual ICs can have
                    a small residual variance. You can also check if the
                    located dipoles lie inside the brain of the dipfit
                    model. <br>
                    Then there's SASICA, an EEGLAB plug-in also
                    published (Chaumon, Bishop & Busch, 2015). I
                    highly recommend the paper also for a deeper
                    understanding of the IC classification. Playing
                    around with SASICA I found it to be okay - not
                    perfect - for an automated method. You can play
                    around with the different classifiers (it uses
                    several methods in combination - dipfit being one of
                    them, but not mandatory) and their respective
                    thresholds and check if it suits you. <br>
                    <br>
                  </div>
                  Unfortunately, EEG methods do have a high degree of
                  subjectivity in several steps, and the automated
                  methods are usually not as good as manual inspection
                  yet (bad channel detection and time-domain artifacts
                  as well), but it's a trade-off between clear
                  reproducibility and best quality, the topic as a whole
                  needs to be treated with care. I hope you find some
                  valuable information in my suggestions and wish you a
                  successful study and smooth data analysis process! ;-)<br>
                  <br>
                </div>
                Best,<br>
              </div>
              Marius<br>
              <div>
                <div>
                  <div>
                    <div>
                      <div><br>
                      </div>
                    </div>
                  </div>
                </div>
              </div>
            </div>
            <div class="HOEnZb">
              <div class="h5">
                <div class="gmail_extra"><br>
                  <div class="gmail_quote">2017-06-30 4:02 GMT+02:00
                    Makoto Miyakoshi <span dir="ltr"><<a
                        moz-do-not-send="true"
                        href="mailto:mmiyakoshi@ucsd.edu"
                        target="_blank">mmiyakoshi@ucsd.edu</a>></span>:<br>
                    <blockquote class="gmail_quote" style="margin:0 0 0
                      .8ex;border-left:1px #ccc solid;padding-left:1ex">
                      <div dir="ltr">Dear Gian Marco,
                        <div><br>
                        </div>
                        <div>Ah I missed your important email! Sorry for
                          being so late.</div>
                        <div>Automated ICA rejection is not more
                          reliable than manual inspection, because the
                          programmer implemented his or her own criteria
                          (or his or her research results on criteria)
                          to the application. Therefore, for rebuttal
                          you can say 'Then why do neurologists still
                          use their eyeballs to identify epileptic
                          spikes, given the great advance of machine
                          learning technology today' etc... Trained eyes
                          are still one of the best solutions.</div>
                        <div><br>
                        </div>
                        <div>> Another question is about bad channel
                          interpolation. In EEGlab there is the kurtosis
                          method but it does not work so good. Do you
                          known any other automatic method that
                          recognize bad channel?<br>
                        </div>
                        <div class="gmail_extra"><br>
                        </div>
                        <div class="gmail_extra">I recommend either the
                          one implemented in clean_rawdata() plugin or
                          the one in PREP plugin. The former was
                          developed by Christian Kothe and the other by
                          Nima Bigdely-Shamlo, both are former SCCN
                          colleagues (now in Qusp). Let me share a piece
                          of my recent writing.</div>
                        <div class="gmail_extra"><br>
                        </div>
                        <div class="gmail_extra"><span
id="m_1741620722056554548m_-7186107844428593281gmail-docs-internal-guid-f5720a7e-f6bc-1c2f-80b1-d91b88a0feab"><span style="font-size:11pt;font-family:Arial;color:rgb(0,0,0);background-color:transparent;vertical-align:baseline;white-space:pre-wrap">We first performed outlier channel detection, rejection, and interpolation using the </span><span style="font-size:11pt;font-family:Arial;color:rgb(0,0,0);background-color:transparent;font-style:italic;vertical-align:baseline;white-space:pre-wrap">clean_rawdata</span><span style="font-size:11pt;font-family:Arial;color:rgb(0,0,0);background-color:transparent;vertical-align:baseline;white-space:pre-wrap"> plug-in (contributed by Christian Kothe) also available through the EEGLAB Extension Manager. This plug-in calculates each scalp channel signal’s correlation to its random sample consensus (RANSAC) estimate computed from nearby scalp channel signals in successive 5-s segments. Channel signals exhibiting low correlation to signals in neighboring scalp channels (e.g., here r < 0.8 at more than 40% of the data points) were rejected and then replaced with an interpolated channel using the spherical option in </span><span style="font-size:11pt;font-family:Arial;color:rgb(0,0,0);background-color:transparent;font-style:italic;vertical-align:baseline;white-space:pre-wrap">eeg_interp</span><span style="font-size:11pt;font-family:Arial;color:rgb(0,0,0);background-color:transparent;vertical-align:baseline;white-space:pre-wrap"> which makes use of Legendre polynomials up to degree 7 to calculate unbiased expected channel values (see Mullen et al., 2015).</span></span><br>
                        </div>
                        <div class="gmail_extra"><br>
                        </div>
                        <div class="gmail_extra">Makoto</div>
                        <div class="gmail_extra"><br>
                        </div>
                        <div class="gmail_extra"><br>
                        </div>
                        <div class="gmail_extra"><br>
                          <div class="gmail_quote">On Mon, Mar 27, 2017
                            at 7:58 AM, Gian Marco Duma <span dir="ltr"><<a
                                moz-do-not-send="true"
                                href="mailto:gmduma90@gmail.com"
                                target="_blank">gmduma90@gmail.com</a>></span>
                            wrote:<br>
                            <blockquote class="gmail_quote"
                              style="margin:0px 0px 0px
                              0.8ex;border-left:1px solid
                              rgb(204,204,204);padding-left:1ex">
                              <div dir="ltr">Dear EEGlab community, I'm
                                writing because I need a suggestion. ICA
                                works very well for eye blink and eye
                                movements correction, and indeed it
                                works very well too for artifacts
                                identification as muscle contraction,
                                electrical noise and so on. Thanks to
                                the experience it becomes possible to
                                recognize the specific components by
                                visual inspection, even if an
                                experimenter must be very careful in
                                components rejection. I submitted a
                                pre-registered reports to Cortex
                                journal, and they asked me for an
                                automatized ICA components rejection
                                method because visual inspection is not
                                considered as a reproducible method. So
                                I'm writing to ask for a suggetion about
                                possible automatized components
                                rejection methods, specially for eye
                                blink and eye movements. 
                                <div>Another question is about bad
                                  channel interpolation. In EEGlab there
                                  is the kurtosis method but it does not
                                  work so good. Do you known any other
                                  automatic method that recognize bad
                                  channel?</div>
                                <div>Thanks for your help</div>
                              </div>
                              <br>
                              ______________________________<wbr>_________________<br>
                              Eeglablist page: <a
                                moz-do-not-send="true"
                                href="http://sccn.ucsd.edu/eeglab/eeglabmail.html"
                                rel="noreferrer" target="_blank">http://sccn.ucsd.edu/eeglab/ee<wbr>glabmail.html</a><br>
                              To unsubscribe, send an empty email to <a
                                moz-do-not-send="true"
                                href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu"
                                target="_blank">eeglablist-unsubscribe@sccn.uc<wbr>sd.edu</a><br>
                              For digest mode, send an email with the
                              subject "set digest mime" to <a
                                moz-do-not-send="true"
                                href="mailto:eeglablist-request@sccn.ucsd.edu"
                                target="_blank">eeglablist-request@sccn.ucsd.e<wbr>du</a><span
                                class="m_1741620722056554548HOEnZb"><font
                                  color="#888888"><br>
                                </font></span></blockquote>
                          </div>
                          <span class="m_1741620722056554548HOEnZb"><font
                              color="#888888"><br>
                              <br clear="all">
                              <div><br>
                              </div>
                              -- <br>
                              <div
                                class="m_1741620722056554548m_-7186107844428593281gmail_signature">
                                <div dir="ltr">Makoto Miyakoshi<br>
                                  Swartz Center for Computational
                                  Neuroscience<br>
                                  Institute for Neural Computation,
                                  University of California San Diego<br>
                                </div>
                              </div>
                            </font></span></div>
                      </div>
                      <br>
                      ______________________________<wbr>_________________<br>
                      Eeglablist page: <a moz-do-not-send="true"
                        href="http://sccn.ucsd.edu/eeglab/eeglabmail.html"
                        rel="noreferrer" target="_blank">http://sccn.ucsd.edu/eeglab/ee<wbr>glabmail.html</a><br>
                      To unsubscribe, send an empty email to <a
                        moz-do-not-send="true"
                        href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu"
                        target="_blank">eeglablist-unsubscribe@sccn.uc<wbr>sd.edu</a><br>
                      For digest mode, send an email with the subject
                      "set digest mime" to <a moz-do-not-send="true"
                        href="mailto:eeglablist-request@sccn.ucsd.edu"
                        target="_blank">eeglablist-request@sccn.ucsd.e<wbr>du</a><br>
                    </blockquote>
                  </div>
                  <br>
                </div>
              </div>
            </div>
          </blockquote>
        </div>
        <br>
        <br clear="all">
        <div><br>
        </div>
        -- <br>
        <div class="gmail_signature" data-smartmail="gmail_signature">
          <div dir="ltr">Makoto Miyakoshi<br>
            Swartz Center for Computational Neuroscience<br>
            Institute for Neural Computation, University of California
            San Diego<br>
          </div>
        </div>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
Eeglablist page: <a class="moz-txt-link-freetext" href="http://sccn.ucsd.edu/eeglab/eeglabmail.html">http://sccn.ucsd.edu/eeglab/eeglabmail.html</a>
To unsubscribe, send an empty email to <a class="moz-txt-link-abbreviated" href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu">eeglablist-unsubscribe@sccn.ucsd.edu</a>
For digest mode, send an email with the subject "set digest mime" to <a class="moz-txt-link-abbreviated" href="mailto:eeglablist-request@sccn.ucsd.edu">eeglablist-request@sccn.ucsd.edu</a></pre>
    </blockquote>
    <p><br>
    </p>
  </body>
</html>