<div dir="ltr">Hi Romain,<div>Yes, that is correct, the first (ideal) approach would be to divide all DW and B=0 images by the mean CSF. For a recent AD study I actually used the 95th percentile of voxels within a CSF mask. Just to make sure I was not getting any PV with grey matter.</div><div><br></div><div>As Donald said, at typically DWI resolutions this can only be done with elderly populations. For cohorts below 60 years I normally use the median b=0 WM value.  As Luis suggested, you could also use the average DWI signal. However, if your WM mask includes voxels affected by pathology (eg a reduced DW signal), then your normalisation will be biased and you may remove some of the effect you are looking for. The b=0 intensity can also be affected by disease and may bias your normalisation, however our assumption is that pathology induced changes to T2 are less severe than the DW signal. As long as the pathology is not global, then using the median value for normalisation means you are less likely to introduce a bias. When computing the median you could also mask out WM regions you suspect are affected by the disease. A good sanity check to ensure that you are not introducing a bias is to compare the median WM value across patient groups with a t-test (to make sure there is no difference).</div><div><br></div><div>We normally use N4 to estimate the bias field on the b=0 and apply it to the DW volumes. This seems to work OK, however we don&#39;t have really severe bias fields.</div><div><br></div><div>Just to follow up on what Donald mentioned regarding problems with dividing by the voxel b=0 in voxels with CSF partial volume. This is not just an issue for spherical deconvolution and Apparent Fibre Density, but also for tensor FA and ADC (see <a href="http://www.ncbi.nlm.nih.gov/pubmed/11323803">http://www.ncbi.nlm.nih.gov/pubmed/11323803</a>). </div><div><br></div><div>Cheers,</div><div>Dave</div><div><br></div><div><br></div><div><br></div><div><br></div><div> <br><div class="gmail_extra"><br><div class="gmail_quote">On 17 December 2014 at 02:23, Luis Concha <span dir="ltr">&lt;<a href="mailto:lconcha@unam.mx" target="_blank">lconcha@unam.mx</a>&gt;</span> wrote:<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex"><div dir="ltr">I have used fsl&#39;s FAST to estimate the bias field on the b=0 and applied it to the DWIs and it seems to work well.<div><br></div><div>I guess one could also use some the segmentations provided by FAST to obtain the mean signal of WM voxels (of the average DWI) and use that to normalise the DWI signal between subjects.</div></div><div class="gmail_extra"><span class=""><br clear="all"><div><div><div dir="ltr">Dr. Luis Concha<br>Instituto de Neurobiología<br>Laboratorio C-13<br>UNAM, Campus Juriquilla<br>Boulervard Juriquilla 3001<br>Juriquilla, Querétaro.<br>C.P. 76230<br>México<br>Tel (442) 2 38 10 54<br>Fax (442) 2 38 10 46<br><a href="http://personal.inb.unam.mx/lconcha/" target="_blank">http://personal.inb.unam.mx/lconcha/</a><br></div></div></div>
<br></span><div><div class="h5"><div class="gmail_quote">On Tue, Dec 16, 2014 at 8:41 AM, romain valabregue <span dir="ltr">&lt;<a href="mailto:romain.valabregue@upmc.fr" target="_blank">romain.valabregue@upmc.fr</a>&gt;</span> wrote:<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
  
    
  
  <div bgcolor="#FFFFFF" text="#000000">
    <div>Hello<br>
      <br>
      This is an interesting thread (worth to put on the wiki ... since
      few people know that you do no take the B0 ...)<br>
      <br>
      Just to be sure I understand the first approach : you divide all
      DWI volume by the mean csf value taken from the b=0 images ?<br>
      <br>
      What about about bias field correction <br>
      The bias is a big concern to: it is very large with 32 channel
      coils<br>
      <br>
      A few input on this ?<br>
      <br>
      N4 on the B0 and apply it to the DWI ? does it works well.<br>
      <br>
      Many thanks <br>
      <br>
      Romain<br>
      <br>
      <br>
      Le 16/12/2014 14:00, J-Donald Tournier a écrit :<br>
    </div><div><div>
    <blockquote type="cite">
      <div dir="ltr">Hi Luis,
        <div><br>
        </div>
        <div>Well spotted. That is one of the issues that we still don&#39;t
          have a totally satisfactory answer for. Basically, we need an
          internal reference intensity for each subject so that we can
          rescale their data to a comparable scale. One approach was to
          use the CSF intensity, which works fine when the ventricles
          are sufficiently large that you can get a clean estimate of
          the CSF signal free from partial volume artefacts. While this
          works in older subjects, it&#39;s difficult to do in younger
          cohorts. Another approach that we use is to measure the b=0
          signal within a conservative white matter mask, and use the
          median or mean value as the reference intensity. This is
          effect scales the data to a single, subject-specific, global
          WM b=0 intensity, avoiding the issues introduced by voxel-wise
          normalisation to the b=0 signal. In all cases, the primary
          concern is whether the intensity used as a reference is
          affected by the condition under investigation, and hence
          whether this step may either mask a genuine effect or
          introduce an effect that is not present in the DW data itself.
          My guess is that the latter approach (normalisation to a
          subject-specific global WM b=0 signal) raises the fewest
          concerns since it&#39;s closest to what most people would naively
          expect anyway (i.e. normalisation to the voxel-wise b=0
          signal). </div>
        <div><br>
        </div>
        <div>Happy to hear your thoughts...</div>
        <div><br>
        </div>
        <div>Cheers,</div>
        <div>Donald.</div>
        <div><br>
        </div>
        <div><br>
        </div>
      </div>
      <div class="gmail_extra"><br>
        <div class="gmail_quote">On 15 December 2014 at 23:11, Luis
          Concha <span dir="ltr">&lt;<a href="mailto:lconcha@unam.mx" target="_blank">lconcha@unam.mx</a>&gt;</span>
          wrote:
          <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
            <div dir="ltr">This is a very interesting discussion, thanks
              Rob and Donald for the detailed answers.
              <div><br>
              </div>
              <div>Rob, you mention that for inter-subject studies you
                would need</div>
              <div>a) B1 field correction (presumably accomplished by N3
                or N4 or similar tools), and</div>
              <div>b) <span style="font-family:arial,sans-serif;font-size:13px">inter-subject
                  intensity normalisation.</span></div>
              <div><span style="font-family:arial,sans-serif;font-size:13px"><br>
                </span></div>
              <div><font face="arial, sans-serif">If not using a b=0
                  singnal normalisation, how can we get signal units
                  comparable between individuals?</font></div>
            </div>
            <div class="gmail_extra"><br clear="all">
              <div>
                <div>
                  <div dir="ltr">Dr. Luis Concha<br>
                    Instituto de Neurobiología<br>
                    Laboratorio C-13<br>
                    UNAM, Campus Juriquilla<br>
                    Boulervard Juriquilla 3001<br>
                    Juriquilla, Querétaro.<br>
                    C.P. 76230<br>
                    México<br>
                    Tel (442) 2 38 10 54<br>
                    Fax (442) 2 38 10 46<br>
                    <a href="http://personal.inb.unam.mx/lconcha/" target="_blank">http://personal.inb.unam.mx/lconcha/</a><br>
                  </div>
                </div>
              </div>
              <div>
                <div>
                  <br>
                  <div class="gmail_quote">On Mon, Dec 15, 2014 at 7:18
                    AM, J-Donald Tournier <span dir="ltr">&lt;<a href="mailto:jdtournier@gmail.com" target="_blank">jdtournier@gmail.com</a>&gt;</span>
                    wrote:<br>
                    <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
                      <div dir="ltr">Hi Jan,
                        <div><br>
                        </div>
                        <div>Just to expand a little further on what Rob
                          said, the main reason for not normalising to
                          the b=0 signal is essentially to preserve the
                          apparent fibre density. Really, the problem is
                          that normalising to the b=0 signal breaks the
                          linearity of the DWI signal to the FOD
                          (irrespective of the log transform), and that
                          is something we think should be avoided if at
                          all possible. This is an issue that I wish I&#39;d
                          described explicitly in my original 2004
                          spherical deconvolution paper (even back then,
                          all the processing was done on the raw
                          signal)... As Rob mentioned, we&#39;ll try to
                          rectify this is a future paper, but for now,
                          here&#39;s a brief description of my reasons for
                          this.</div>
                        <div><br>
                        </div>
                        <div>This is based on a fundamental aspect of
                          spherical deconvolution and mixture models in
                          general: that the DWI signal scales <i>linearly</i> with
                          the amount of tissue present. While the
                          simulations done by Dave Rafflelt in the paper
                          Rob mentioned do make the point very nicely,
                          their purpose is a lot more specific than is
                          required for this argument. Basically, If a
                          voxel contains two fibre bundles, the signal
                          you measure is the sum of the signals for each
                          bundle individually (at least, it&#39;s modelled
                          as such). This however does not necessarily
                          hold for the signal <i>attenuation</i>, since
                          the b=0 signal is not uniform throughout the
                          brain.  </div>
                        <div><br>
                        </div>
                        <div>Consider for example voxels containing
                          mixtures of WM &amp; CSF. The b=0 signal for
                          CSF is typically very high relative to WM (due
                          to its long T2). If half the voxel contains
                          CSF, the other half WM, the b=0 signal for
                          that voxel would be essentially double what it
                          would be for pure WM (assuming CSF b=0 signal
                          is ~3x that of WM). On the other hand, the DW
                          signal for CSF is small, and to all intents
                          and purposes negligible at high b-values. This
                          means the <i>raw</i> DW signal would be what
                          you would expect to measure for a voxel
                          containing half the volume of WM, but the
                          signal attenuation would be halved again
                          (since the b=0 signal is double). So the
                          apparent relative volume fraction (fibre
                          density) derived using signal <i>attenuation</i>
                          would be ~a quarter that of pure WM, while
                          using the<i> raw</i> DW signal would give you
                          the correct answer: half that of pure WM. If
                          you care about being able to compare apparent
                          fibre densities across voxels in the presence
                          of large variations in the b=0 signal between
                          different tissue types (i.e. as you would
                          expect in the brain), you shouldn&#39;t normalise
                          to the b=0 image. </div>
                        <div><br>
                        </div>
                        <div>Note this isn&#39;t just about voxel-based
                          analysis of apparent fibre density or SIFT:
                          this is also important for example during the
                          tractography itself, since the termination
                          criteria are applied on the FOD amplitude
                          directly. It is also important for anything
                          that involves consistent scaling of the noise
                          (e.g. bootstrap analysis), since normalising
                          to the b=0 will also introduce large and rapid
                          spatial variations in the noise
                          characteristics of the data. There are many
                          facets to this issue, and I won&#39;t bother going
                          into them in any more detail here - I&#39;ll leave
                          that for the future paper Rob mentioned. But
                          in a nutshell, this is the reason MRtrix has
                          always operated on the raw DW signal, not its
                          attenuated version. </div>
                        <div><br>
                        </div>
                        <div>Hope this all makes sense.</div>
                        <div>Cheers,</div>
                        <div><br>
                        </div>
                        <div>Donald.</div>
                        <div><br>
                        </div>
                      </div>
                      <div class="gmail_extra"><br>
                        <div class="gmail_quote">
                          <div>
                            <div>On 15 December 2014 at 00:32, Robert
                              Smith <span dir="ltr">&lt;<a href="mailto:robert.smith@florey.edu.au" target="_blank">robert.smith@florey.edu.au</a>&gt;</span>
                              wrote:</div>
                          </div>
                          <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">
                            <div>
                              <div>
                                <div dir="ltr">Hi Jan,
                                  <div><br>
                                  </div>
                                  <div>This is an important point, and
                                    one that we sometimes forget that we
                                    (as in, the MRtrix dev team) think
                                    about quite differently to others in
                                    Diffusion MR.</div>
                                  <div>We will draw attention to this
                                    issue in an upcoming publication,
                                    but I&#39;ll try to give a succinct
                                    explanation here.</div>
                                  <div><br>
                                  </div>
                                  <div>Conventionally, the log-transform
                                    with respect to the b=0 image
                                    converts a signal amplitude to an
                                    apparent diffusion coefficient;
                                    nothing controversial here. However
                                    if you were to then apply a
                                    spherical deconvolution transform,
                                    the FOD amplitude along a particular
                                    direction would be proportional to
                                    the ADC of the fibre population
                                    oriented along that direction. This
                                    isn&#39;t particularly useful
                                    information; it doesn&#39;t tell us much
                                    about differences between fibre
                                    populations throughout the image, or
                                    indeed within a voxel.</div>
                                  <div><br>
                                  </div>
                                  <div>Ideally what we actually want for
                                    a number of applications is the
                                    volume of each fibre population
                                    element, in all voxels throughout
                                    the image. Based on David Raffelt&#39;s
                                    early <a href="http://www.sciencedirect.com/science/article/pii/S1053811911012092" target="_blank">simulations</a>,
                                    it turns out that (under certain
                                    conditions) the radial component of
                                    the DWI signal amplitude is actually
                                    a pretty decent marker for
                                    intra-cellular volume. Therefore, by
                                    ignoring the b=0 images completely
                                    and just running SD on the raw DWI
                                    intensities, we get pretty useful
                                    biological information and
                                    interpretation from the FOD; we also
                                    conveniently bypass the issue of
                                    Gibbs ringing in the b=0 images.
                                    Caveat is that you need a uniform B1
                                    field (i.e. intensity bias field
                                    correction); for applications like
                                    AFD you also need inter-subject
                                    intensity normalisation, but that&#39;s
                                    not necessarily a problem for SIFT
                                    depending on how you&#39;re using it.</div>
                                  <div><br>
                                  </div>
                                  <div>That&#39;s all for now. Hope that
                                    clarifies why we choose to apply SD
                                    in this way; in fact, this approach
                                    dates all the way back to the
                                    original SD paper.</div>
                                  <div>Rob</div>
                                </div>
                                <div class="gmail_extra"><br clear="all">
                                  <div>
                                    <div>
                                      <div dir="ltr"><br>
                                        --<br>
                                        <br>
                                        <span style="color:rgb(255,102,0)"><b>Robert
                                            Smith, Ph.D</b><br>
                                          Research Officer, Imaging
                                          Division</span><br>
                                        <br>
                                        The Florey Institute of
                                        Neuroscience and Mental Health<br>
                                        Melbourne Brain Centre - Austin
                                        Campus<br>
                                        245 Burgundy Street<br>
                                        Heidelberg Vic 3084<br>
                                        Ph: <a href="tel:%2B61%203%209035%207128" value="+61390357128" target="_blank">+61 3 9035
                                          7128</a><br>
                                        Fax: <a href="tel:%2B61%203%209035%207301" value="+61390357301" target="_blank">+61
                                          3 9035 7301</a><br>
                                        <a href="http://www.florey.edu.au/" target="_blank">www.florey.edu.au</a><br>
                                      </div>
                                    </div>
                                  </div>
                                  <div>
                                    <div>
                                      <br>
                                      <div class="gmail_quote">On Sat,
                                        Dec 13, 2014 at 1:37 AM, Jan
                                        Schreiber <span dir="ltr">&lt;<a href="mailto:schreiber@cbs.mpg.de" target="_blank">schreiber@cbs.mpg.de</a>&gt;</span>
                                        wrote:
                                        <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-color:rgb(204,204,204);border-left-style:solid;padding-left:1ex">Dear
                                          MRtrix Team,<br>
                                          <br>
                                          thank you very much for this
                                          great software and for making
                                          it freely<br>
                                          available!<br>
                                          <br>
                                          In your publication &quot;SIFT:
                                          Spherical-deconvolution
                                          informed filtering of<br>
                                          tractograms&quot; you state<br>
                                          <br>
                                          &quot;The diffusion signal must not
                                          be normalised to the b = 0
                                          image<br>
                                          intensity. This preserves the
                                          linearity of the spherical
                                          deconvolution<br>
                                          transform between the measured
                                          DW signal and the resulting
                                          FOD.&quot;<br>
                                          <br>
                                          Shouldn&#39;t we preserve the
                                          linearity of the spherical
                                          deconvolution<br>
                                          transform between the FOD and
                                          the DW _signal attenuation_
                                          rather than<br>
                                          the DW _signal_?<br>
                                          <br>
                                          Thanks,<br>
                                          Jan<br>
                                          <br>
                                        </blockquote>
                                      </div>
                                    </div>
                                  </div>
                                </div>
                                <br>
                              </div>
                            </div>
_______________________________________________<br>
                            Mrtrix-discussion mailing list<br>
                            <a href="mailto:Mrtrix-discussion@www.nitrc.org" target="_blank">Mrtrix-discussion@www.nitrc.org</a><br>
                            <a href="http://www.nitrc.org/mailman/listinfo/mrtrix-discussion" target="_blank">http://www.nitrc.org/mailman/listinfo/mrtrix-discussion</a><br>
                            <br>
                          </blockquote>
                        </div>
                        <span><font color="#888888"><br clear="all">
                            <div><br>
                            </div>
                            -- <br>
                            <div>
                              <div dir="ltr"><b><font color="#990000">Dr
                                    J-Donald Tournier (PhD)</font></b><br>
                                <div><font color="#990000"><br>
                                  </font></div>
                                <i><font color="#990000">Senior
                                    Lecturer, </font></i><i><font color="#990000">Biomedical
                                    Engineering</font></i>
                                <div><i><font color="#990000">Division
                                      of Imaging Sciences &amp;
                                      Biomedical Engineering<br>
                                      King&#39;s College London</font></i>
                                  <div><i><font color="#990000"><br>
                                      </font></i></div>
                                  <div><i><font color="#990000"><b style="font-family:Calibri,sans-serif;font-size:15px"><span style="font-size:10pt">A:</span></b><span style="font-family:Calibri,sans-serif;font-size:10pt"> Department of
                                          Perinatal Imaging &amp;
                                          Health, 1<sup>st</sup> Floor
                                          South Wing, St Thomas&#39;
                                          Hospital, London. SE1 7EH</span><br>
                                      </font></i></div>
                                  <div><i><font color="#990000"><b>T:</b>
                                        <a href="tel:%2B44%20%280%2920%207188%207118%20ext%2053613" value="+442071887118" target="_blank">+44 (0)20 7188
                                          7118 ext 53613</a></font></i></div>
                                </div>
                                <div><i><font color="#990000"><b>W:</b> <a href="http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering" target="_blank">http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering</a></font></i><br>
                                </div>
                              </div>
                            </div>
                          </font></span></div>
                      <br>
                      _______________________________________________<br>
                      Mrtrix-discussion mailing list<br>
                      <a href="mailto:Mrtrix-discussion@www.nitrc.org" target="_blank">Mrtrix-discussion@www.nitrc.org</a><br>
                      <a href="http://www.nitrc.org/mailman/listinfo/mrtrix-discussion" target="_blank">http://www.nitrc.org/mailman/listinfo/mrtrix-discussion</a><br>
                      <br>
                    </blockquote>
                  </div>
                  <br>
                </div>
              </div>
            </div>
            <br>
            _______________________________________________<br>
            Mrtrix-discussion mailing list<br>
            <a href="mailto:Mrtrix-discussion@www.nitrc.org" target="_blank">Mrtrix-discussion@www.nitrc.org</a><br>
            <a href="http://www.nitrc.org/mailman/listinfo/mrtrix-discussion" target="_blank">http://www.nitrc.org/mailman/listinfo/mrtrix-discussion</a><br>
            <br>
          </blockquote>
        </div>
        <br clear="all">
        <div><br>
        </div>
        -- <br>
        <div>
          <div dir="ltr"><b><font color="#990000">Dr J-Donald Tournier
                (PhD)</font></b><br>
            <div><font color="#990000"><br>
              </font></div>
            <i><font color="#990000">Senior Lecturer, </font></i><i><font color="#990000">Biomedical Engineering</font></i>
            <div><i><font color="#990000">Division of Imaging Sciences
                  &amp; Biomedical Engineering<br>
                  King&#39;s College London</font></i>
              <div><i><font color="#990000"><br>
                  </font></i></div>
              <div><i><font color="#990000"><b style="font-family:Calibri,sans-serif;font-size:15px"><span style="font-size:10pt">A:</span></b><span style="font-family:Calibri,sans-serif;font-size:10pt"> Department
                      of Perinatal Imaging &amp; Health, 1<sup>st</sup> Floor
                      South Wing, St Thomas&#39; Hospital, London. SE1 7EH</span><br>
                  </font></i></div>
              <div><i><font color="#990000"><b>T:</b> +44 (0)20 7188
                    7118 ext 53613</font></i></div>
            </div>
            <div><i><font color="#990000"><b>W:</b> <a href="http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering" target="_blank">http://www.kcl.ac.uk/medicine/research/divisions/imaging/departments/biomedengineering</a></font></i><br>
            </div>
          </div>
        </div>
      </div>
      <br>
      <fieldset></fieldset>
      <br>
      <pre>_______________________________________________
Mrtrix-discussion mailing list
<a href="mailto:Mrtrix-discussion@www.nitrc.org" target="_blank">Mrtrix-discussion@www.nitrc.org</a>
<a href="http://www.nitrc.org/mailman/listinfo/mrtrix-discussion" target="_blank">http://www.nitrc.org/mailman/listinfo/mrtrix-discussion</a>
</pre>
    </blockquote>
    <br>
  </div></div></div>

<br>_______________________________________________<br>
Mrtrix-discussion mailing list<br>
<a href="mailto:Mrtrix-discussion@www.nitrc.org" target="_blank">Mrtrix-discussion@www.nitrc.org</a><br>
<a href="http://www.nitrc.org/mailman/listinfo/mrtrix-discussion" target="_blank">http://www.nitrc.org/mailman/listinfo/mrtrix-discussion</a><br>
<br></blockquote></div></div></div></div>
<br>_______________________________________________<br>
Mrtrix-discussion mailing list<br>
<a href="mailto:Mrtrix-discussion@www.nitrc.org">Mrtrix-discussion@www.nitrc.org</a><br>
<a href="http://www.nitrc.org/mailman/listinfo/mrtrix-discussion" target="_blank">http://www.nitrc.org/mailman/listinfo/mrtrix-discussion</a><br>
<br></blockquote></div><br clear="all"><div><br></div>-- <br><div class="gmail_signature"><div dir="ltr"><div><b><font color="#ff6600">David Raffelt (PhD)</font></b></div><div><font color="#ff6600">Post Doctoral Fellow</font></div><div><br></div><div>The Florey Institute of Neuroscience and Mental Health</div><div>Melbourne Brain Centre - Austin Campus</div><div>245 Burgundy Street</div><div>Heidelberg Vic 3084<div>Ph: <a value="+61390357024">+61 3 9035 7024</a></div></div><div><a value="+61390357024">www.florey.edu.au</a></div></div></div>
</div></div></div>