BEM model: watershed vs. flash

✉️ Mailing List Archive (read-only)
1

External Email - Use Caution

Hi MNE-experts,

I'm trying to build a 3-compartment BEM model to source localize combined
MEG and EEG data.

At the beginning, I used the watershed algorithm and MPRAGE images to build
the three-layer model. The plot (watershed:
https://www.dropbox.com/s/ceyucikgvcpjka9/watershed.png?dl=0) shows that
the outer-skull surface is closer to the outer-skin surface, which is very
different from what was shown on the MNE documentation webpage here:
https://mne-tools.github.io/dev/auto_tutorials/plot_forward.html#sphx-glr-auto-tutorials-plot-forward-py

Later, I used flash algorithm and FLASH images to have a better estimation
of the three layers. This way, the plot ( flash:
https://www.dropbox.com/s/8c6l5fiosl7p19k/flash.png?dl=0) shows similar
segmentation results as shown in the documentation webpage, i.e. the
outer-skull surface is closer to the inner-skull surface.

My first question is: which algorithm was used for the plot on the
documentation webpage? Should I use flash for the 3-layer model?

My second question is: is it possible to have a relatively accurate 3-layer
model using watershed? We have 9 subjects missing the FLASH images.
Although I can still build the 3-layer head model using watershed for 8
subjects, I run into this error for one subject: "the surface inner skull
is not completely inside surface outer skull". Is there any easy fix?

Thanks a lot for you help!

Lin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190416/3f736515/attachment.html

2

External Email - Use Caution

At the beginning, I used the watershed algorithm and MPRAGE images to
build the three-layer model. The plot (watershed:
Dropbox - watershed.png - Simplify your life) shows that
the outer-skull surface is closer to the outer-skin surface, which is very
different from what was shown on the MNE documentation webpage here:
https://mne-tools.github.io/dev/auto_tutorials/plot_forward.html#sphx-glr-auto-tutorials-plot-forward-py

Freesurfer's watershed bem strategy
<mri_watershed - Free Surfer Wiki;
is roughly:

1. Compute outer skin surface
2. Shrink outer skin inward make the "outer skull"
3. Compute brain surface (IIRC this is extracted from brain.mgz, which gets
masked by brainmask.mgz)
4. Expand brain surface outward to make the "inner skull"

So it's not 100% correct that the outer skull is this way, but it is
working as designed.

My first question is: which algorithm was used for the plot on the

documentation webpage? Should I use flash for the 3-layer model?

The `sample` dataset has FLASH images, and the BEM surfaces were computed
from those.

My second question is: is it possible to have a relatively accurate 3-layer

model using watershed?

FLASH should be better than watershed. As to the errors that the watershed
creates, I'm not sure. For your subjects that have usable FLASH sequences,
you could compute head models using FLASH and using watershed, and compare
them, for example how much worse the watershed peak localization is
compared to the FLASH ones. You can adapt some of our basic minimum norm
benchmarking code
<https://github.com/mne-tools/mne-python/blob/master/mne/minimum_norm/tests/test_inverse.py#L296-L313&gt;
if you want. It's also likely someone has already done this, it's probably
worth a literature search (I don't know offhand).

We have 9 subjects missing the FLASH images. Although I can still build the

3-layer head model using watershed for 8 subjects, I run into this error
for one subject: "the surface inner skull is not completely inside surface
outer skull". Is there any easy fix?

In these cases you can try changing the --preflood / -p
<Command line tools using Python — MNE 1.7.0.dev17+g20174f448 documentation;
parameter
in mne watershed_bem. If that does not do it, you can manually run
mri_watershed
<mri_watershed - Free Surfer Wiki;
(which
mne watershed_bem calls under the hood) with various parameters to fix the
output. Or you might have to go farther back in your Freesurfer pipeline to
fix the problem. Inner skull not within outer skull can be caused by the
brain masking step (part of autorecon1
<https://surfer.nmr.mgh.harvard.edu/fswiki/ReconAllDevTable&gt;\) getting some
dura/skull within the brain mask. Refer to the Freesurfer listserv for
ideas on how to fix this problem, for example tweaking the mri_normalize
parameters via xopts
<https://www.mail-archive.com/freesurfer at nmr.mgh.harvard.edu/msg20991.html>
can sometimes help.

Best,
Eric
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190417/9b540eea/attachment.html

3

External Email - Use Caution

Hi Eric,

Thank you very much for your clear explanations. I'll try out your suggestions.

Best,
Lin