MRG : source reconstruction on EEG using fsaverage brain #6146

agramfort · 2019-04-13T12:39:37Z

this PR aims to cover a long standing issue with MNE which is to do EEG source reconstruction using a template brain here fsaverage.

Limitation of the current PR:

very coarse narrative doc
has no cross-links with other tutorials
only tests with one montage file 'standard_1005'
data contains induced power so current evoked pipeline is not appropriate.
needs mne.datasets.fetch_fsaverage_3layer_bem or so

@drammock any thoughts on how to integrate this in the new in the making documentation?

@massich this is a good test for the new digitization refactoring

…ing fsaverage brain

tutorials/plot_eeg_no_mri.py

agramfort · 2019-04-15T07:38:43Z

we should put here a link to the forward tutorial rather than duplicating the narrative. I won't touch at this before the sprint next week

…

tutorials/plot_eeg_no_mri.py

agramfort · 2019-04-15T07:45:52Z

it's WIP so we don't loose this. If you have time to improve please push directly to this branch.

mmagnuski · 2019-04-15T07:48:11Z

Circle errors on the example:

Traceback (most recent call last):
  File "/home/circleci/.local/lib/python3.7/site-packages/sphinx_gallery/gen_rst.py", line 398, in _memory_usage
    multiprocess=True)
  File "/home/circleci/.local/lib/python3.7/site-packages/memory_profiler.py", line 336, in memory_usage
    returned = f(*args, **kw)
  File "/home/circleci/.local/lib/python3.7/site-packages/sphinx_gallery/gen_rst.py", line 389, in __call__
    exec(self.code, self.globals)
  File "/home/circleci/project/tutorials/plot_eeg_no_mri.py", line 58, in <module>
    subjects_dir=subjects_dir, trans=trans_fname)
  File "</home/circleci/project/mne/externals/decorator.py:decorator-gen-117>", line 2, in plot_alignment
  File "/home/circleci/project/mne/utils/_logging.py", line 89, in wrapper
    return function(*args, **kwargs)
  File "/home/circleci/project/mne/viz/_3d.py", line 815, in plot_alignment
    % (subject, '\n'.join(try_fnames)))
OSError: No head surface found for subject fsaverage after trying:
/home/circleci/mne_data/MNE-sample-data/subjects/fsaverage/bem/outer_skin.surf
/home/circleci/mne_data/MNE-sample-data/subjects/fsaverage/bem/flash/outer_skin.surf
/home/circleci/mne_data/MNE-sample-data/subjects/fsaverage/bem/fsaverage-head.fif

mmagnuski · 2019-04-15T07:54:58Z

@agramfort Sure, I can try, but I had a lot of problems before trying to push to other people's PRs. If you also have a tutorial on that, that would be helpful. :)

mmagnuski · 2019-04-15T07:57:22Z

Regarding the error: is it possible that mne's fsaverage ships without head surface? I remember I had to construct it when using fsaverage for source localization.

agramfort · 2019-04-15T11:38:38Z

yes a new fsaverage dataset is necessary or at least an update of sample dataset to add these missing surfaces. I had to compute them locally.

…

larsoner · 2019-04-15T11:56:31Z

If you upload the surfaces I can update the dataset. Or if it's as simple as mne watershed_bem I'll run it locally and update.

agramfort · 2019-04-15T12:44:03Z

yes it's just running mne watershed_bem we can add this to sample but I think it could be smart to have a tiny dataset with just fsaverage for those we want to do EEG with MRI templates

tutorials/plot_eeg_no_mri.py

larsoner · 2019-04-15T13:15:08Z

We already have fsaverage-inner_skull-bem.fif in mne/data/ but no 3-layer, and I don't think we should add it there because it bloats the repo. We could add a mne.datasets.fetch_fsaverage_3layer_bem or something. We can converge at the sprint

agramfort · 2019-04-15T13:35:56Z

+1 for a specific fetcher like mne.datasets.fetch_fsaverage_3layer_bem but I would not assume that sample is present to have it working

…

massich · 2019-04-15T16:13:35Z

This tuto does something similar to what we have in #5836 but with fsavarege rather than sample. I can take over and the missing files so that this PR renders in circle.

codecov · 2019-04-15T20:31:15Z

Codecov Report

Merging #6146 into master will increase coverage by <.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #6146      +/-   ##
==========================================
+ Coverage   87.21%   87.21%   +<.01%     
==========================================
  Files         412      412              
  Lines       73985    73985              
  Branches    12270    12270              
==========================================
+ Hits        64523    64525       +2     
+ Misses       6592     6588       -4     
- Partials     2870     2872       +2

massich · 2019-04-15T20:54:56Z

here is the rendering https://12255-1301584-gh.circle-artifacts.com/0/dev/auto_tutorials/plot_eeg_no_mri.html?highlight=plot_eeg_no_mri

massich · 2019-04-15T20:59:34Z

The current solution requires to download 1.8Gb we could either allow for:

Partial download of the sample
split sample and fsaverage and,
2a) ship them separately
2b) ship them twice

drammock · 2019-04-15T22:36:08Z

@agramfort for now, putting it in the root of tutorials (as you've done) is fine. This PR will almost surely get merged before the big tutorials overhaul, and I'll rebase against it and move this new tutorial as necessary at that time.

mmagnuski · 2019-04-16T13:09:17Z

examples/visualization/plot_montage.py

+# check all montages
+#
+
+fix_units = {'EGI_256':'cm', 'GSN-HydroCel-128':'cm', 'GSN-HydroCel-129':'cm',


EGI_256 is not in cm, take a look at this comment: the distance for each channel from the center is exactly 1. for EGI_256.

I missed your comment. The way you figured out the distance is smarter than the way I did it.

The code was pulled from an early stage of #5836 and yes EGI_256 it's broken. Now that we are at it, when load EGI_256 with eeg=['original', 'projected'] you can see that the original do have a 'brainish' shape but the projected does not.

Either way, we decided with @agramfort to add the example ploting all helmets i a different PR. (I'll ping you there)

larsoner · 2019-04-19T12:03:31Z

I'm okay with the sensitivity map, we should just link at the end to a standard MNE tutorial or something then

agramfort · 2019-04-19T12:13:44Z

yes +100

massich · 2019-04-19T13:03:19Z

does it mean that we make #6167 green (or cherry pick 32efc81) and call it a day? Or are we missing something else?

massich · 2019-04-19T13:04:56Z

does it mean that we make #6167 green (or cherry pick 32efc81) and call it a day? Or are we missing something else?

I'm referring to the example itself.

agramfort · 2019-04-19T13:20:01Z

Yes. Lets move on

larsoner · 2019-04-19T15:17:51Z

Feel free to push to my branch the commit to go back to eegbci

agramfort · 2019-04-19T19:25:46Z

mne/datasets/_fsaverage.py

+    return subjects_dir
+
+
+def set_montage_coreg_path(subjects_dir=None):


I would make this private

agramfort

@massich I am lost reviewing this PR. I see unrelated things and I still see the use of set_montage_coreg_path

agramfort · 2019-04-20T09:08:05Z

mne/bem.py

+        if incomplete == 'raise':
+            raise RuntimeError(msg)
+        else:
+            warn(msg)


why is this here???

got here likely by accident in 8d1023e

agramfort · 2019-04-20T09:08:42Z

mne/commands/tests/test_commands.py

+    fnames = glob.glob(op.join(tempdir, '*proj.fif'))
+    assert len(fnames) == 1
+    fnames = glob.glob(op.join(tempdir, '*-eve.fif'))
+    assert len(fnames) == 1


how did this happen in this PR?

massich · 2019-04-23T10:55:28Z

mne/commands/mne_watershed_bem.py

@@ -37,9 +37,6 @@ def run():
                      action="store_true")
    parser.add_option("-p", "--preflood", dest="preflood",
                      help="Change the preflood height", default=None)
-    parser.add_option("--copy", dest="copy",
-                      help="Use copies instead of symlinks for surfaces",
-                      action="store_true")


@larsoner all this bem related things that were added to the PR. Were they needed?? I mean, they are unrelated to the PR. But maybe you wanted them.

They were needed when I created the 3-layer BEM for fsaverage, so yes please keep

larsoner · 2019-04-23T12:29:19Z

mne/commands/tests/test_commands.py

    """Test mne flash_bem."""
    check_usage(mne_flash_bem, force_help=True)
    # Using the sample dataset
    subjects_dir = op.join(sample.data_path(download=False), 'subjects')
    # Copy necessary files to tempdir
-    tempdir = str(tmpdir)
+    tempdir = _TempDir()


We want almost everything from this commit. Feel free to push a new PR but we shouldn't lose them

no prob. I'll get it back and just ping you to make sure I did not forget anything important.

Actually I made a PR #6192 feel free to push in it.

jona-sassenhagen · 2019-04-23T14:24:41Z

Didn't see this ... very cool MNE is making an effort to support the "lazy and stupid" crowd finally!!!

drammock

I reviewed only the Tutorial at @massich's request; there are lots of small improvements that are possible, but nothing that is unclear enough to be a blocker.

drammock · 2019-04-23T14:32:45Z

tutorials/plot_eeg_no_mri.py

+# The files live in:
+subject = 'fsaverage'
+trans = op.join(fs_dir, 'bem', 'fsaverage-trans.fif')
+src = op.join(fs_dir, 'bem', 'fsaverage-5-src.fif')


I think this needs to be 'fsaverage-ico-5-src.fif?

larsoner

LGTM +1 for merge. Rendered example:

https://12457-1301584-gh.circle-artifacts.com/0/dev/auto_tutorials/plot_eeg_no_mri.html

massich · 2019-04-23T16:11:27Z

the failing error is unrelated. Merging

WIP : adding a demo example that runs source reconstruction on EEG us…

629ae1a

…ing fsaverage brain

agramfort mentioned this pull request Apr 13, 2019

plot_alignment not showing EEG electrode locations on scalp #6138

Closed

mmagnuski reviewed Apr 15, 2019

View reviewed changes