<div dir="ltr">Dear all,<div><br></div><div>I have also looked at this in a different context and found that bemcp computation breaks down at distances smaller than mean triangle side length from the boundary (for meshes used in SPM that's about 6mm). So you can avoid numerical problems by making sure your sources are never closer than that to the inner skull boundary. This can be done with cfg.inwardshift (for grids) or cfg.moveinward (for meshes) arguments of ft_prepare_sourcemodel, In SPM12 this is done automatically for the forward models we generate.</div><div><br></div><div>Best,</div><div><br></div><div>Vladimir</div></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Nov 4, 2015 at 1:36 PM, RICHARDS, JOHN <span dir="ltr"><<a href="mailto:RICHARDS@mailbox.sc.edu" target="_blank">RICHARDS@mailbox.sc.edu</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">In addition to the answer given by Alexandre Gramfort.<br>
<br>
I have compared BEMCP, Dipoli, Sim-bio FEM, and spherical models. My tests use an individual head model with that individuals data, and then head models based on other participants, age-appropriate MRI templates, and age-inappropriate MRI templates; with infants. The BEMCP is the three-compartment model, the Dipoli is a four-compartment model, the Simbio is a full segmented head model with 10 different materials, the spherical is four spheres.<br>
<br>
1—I sometimes have issues getting the forward model / source analysis with BEMCP. It appears to have singularities on the borders of the compartments and gives the same answer as you report. Or does not even compute. Using the same (or similar) head model mesh with Dipole, it works.<br>
<br>
2—I have been using the Dipoli with four compartments. I am comparing the four methods with a set of empirical data and simulated data. I have found that the dipole solutions on a single individual(s) for the BEMCP and the spherical models are closer in solution, and the Dipoli and the SimBio-FEM are closer in the solution.<br>
<br>
3—I also have been using head models from different individuals to solve the forward model for the EEG from a single individual, comparing the solutions across head models. Both the SimBio-FEM and Dipoli show the greatest change in solution across individuals with different head sizes or ages (using age-appropriate and age-inappropriate heads), whereas the BEMCP and spherical model show the least change. Interesting, the larger change for the two models occurs because its theoretical fit when the head comes from the individual is better than the fit when the head comes from an inappropriate age, whereas the age-appropriate head model are not as important for BEMCP and spherical models because they fit poorer.<br>
<br>
The results from 2 and 3 are being analyzed now, eventually will reach publication.<br>
<br>
John<br>
<br>
<br>
>------------------------------<br>
><br>
>Message: 4<br>
>Date: Wed, 4 Nov 2015 00:00:35 +0100<br>
>From: Maris Skujevskis <<a href="mailto:icelandhouse@gmail.com">icelandhouse@gmail.com</a>><br>
>To: <a href="mailto:fieldtrip@science.ru.nl">fieldtrip@science.ru.nl</a><br>
>Subject: [FieldTrip] bemcp vs dipoli<br>
>Message-ID:<br>
> <CAKEvDicMF9uo1G-_TVjJvGXE9vvnbXmGVg7gLZQXG-=<a href="mailto:dQimv7Q@mail.gmail.com">dQimv7Q@mail.gmail.com</a>><br>
>Content-Type: text/plain; charset="utf-8"<br>
<span class="">><br>
>Hi Fieldtrip users,<br>
><br>
>When constructing a volume conduction model for EEG using the boundary<br>
>element method (BEM), there are two methods available: 'dipoli' and<br>
>'bemcp'. Besides technicalities (i.e., 'dipoli' only available on Linux),<br>
>are there any differences that you know of/have experienced that make one<br>
>method better (more reliable, more accurate, or anything else that makes<br>
>you prefer one rather) than the other?<br>
><br>
>For some of my EEG subjects, 'dipoli' succeeds where 'bemcp' fails. During<br>
>processing with 'bemcp' there is a "warning: Matrix is singular, close to<br>
>singular or badly scaled. Results may be inaccurate. RCOND = NaN", with the<br>
>end result being that vol.mat contains NaNs.<br>
>The input in both cases, besides the method chosen, is identical.<br>
><br>
>Taking one step back in the processing pipeline, I am aware that a poor<br>
>segmentation outcome might be a/the cause of the warnings and eventual<br>
>errors when constructing a volume conduction model. But in general this<br>
>should hold equally for both methods. What I am wondering about is why one<br>
>method might deal more successfully than the other with the same input.<br>
><br>
>Best,<br>
>Maris<br>
</span>>-------------- next part --------------<br>
<div class="HOEnZb"><div class="h5"><br>
_______________________________________________<br>
fieldtrip mailing list<br>
<a href="mailto:fieldtrip@donders.ru.nl">fieldtrip@donders.ru.nl</a><br>
<a href="http://mailman.science.ru.nl/mailman/listinfo/fieldtrip" rel="noreferrer" target="_blank">http://mailman.science.ru.nl/mailman/listinfo/fieldtrip</a></div></div></blockquote></div><br></div>