Order of vertices in morphed source spaces

Hi,

When morphing the source space of fsaverage to ‘sample’, the order of the morphed vertices is not preserved. The first 10 vertices of the morphed source space do not correspond to those of fsaverage:

fsaverage:

Morphed fsaverage -> sample:

import mne
import os

from mne.datasets.sample import data_path
from mayavi import mlab  # noqa
from surfer import Brain  # noqa

data_path = data_path()
subjects_dir = data_path + "/subjects/"
os.environ['SUBJECTS_DIR'] = subjects_dir

fname_fs = subjects_dir + "fsaverage/bem/fsaverage-ico-5-src.fif"
src_fs = mne.read_source_spaces(fname_fs)
src_sample = mne.morph_source_spaces(src_fs, subject_to="sample",
                                     subjects_dir=subjects_dir)
sources = np.arange(10)

for subject, src in zip(["fsaverage", "sample"], [src_fs, src_sample]):
    brain = Brain(subject, 'lh', 'inflated', subjects_dir=subjects_dir)
    surf = brain.geo['lh']
    vertices = src[0]["vertno"][sources]
    mlab.points3d(surf.x[vertices], surf.y[vertices],
                  surf.z[vertices], color=(1, 0, 0), scale_factor=6.)
    mlab.savefig("%s-sources.png" % subject)

This can be fixed by removing np.sort in https://github.com/mne-tools/mne-python/blob/master/mne/source_space.py#L2639

   ` to['vertno'] = np.sort(best[fro['vertno']])`

Then, the first 10 vertices of the morphed source space do match those of fsaverage:

Morphed fsaverage -> sample (after removing np.sort):

When sorting the vertices one loses their correspondance which is the purpose of morphing. The resulting forward operators based on these morphed source spaces would have completely mismatched locations.

Written with @agramfort

[agramfort]

I think it’s more complicated than this. I talked to @larsoner

MNE-C also outputs sorted vertices and I don’t know why. We need @mshamalainen wisdom on this. For me it defeats the purpose of source space morphing as with this sorting you still have to do stc.morph for group studies.

If we remove the np.sort then we might break stuff like backward compat with MNE-C and also some of our functions require stc.vertno to be sorted. Or at least if it’s not you get a ValueError…

[agramfort]

no but I am not sure we can fix it in a reasonable way.