Dig points disappear under transparent skin surface using mne coreg GUI

[nordme]

Description of the problem

Hi! Looks like there's something quirky going on with mne coreg.

Additional points in a head digitization are disappearing from view when they sink beneath the skin surface, regardless of the opacity of the surface. While cHPI coils and EEG sensor locations persist (red & pink dots), additional dig points (white dots) disappear.

Steps to reproduce

1. Set your subjects dir path to point to the "subjects" directory of the MNE sample data set. 
2. In the command line, open `mne coreg`
3. In the "path to info" field, set "sample_audvis_raw.fif" from the mne-python MEG sample dataset as your digitization source. 
4. Using the translation buttons, decrease the y-translation value (hit the down arrow by "tY") until the digitization points in the back of the head cross the head surface.

Link to data

No response

Expected results

White dots should remain visible underneath the skin surface.

Actual results

Additional information

`>>> mne.sys_info()
Platform Linux-6.2.0-34-generic-x86_64-with-glibc2.35
Python 3.11.5 | packaged by conda-forge | (main, Aug 27 2023, 03:34:09) [GCC 12.3.0]
Executable /home/erica/miniconda3/envs/sprint_2023/bin/python
CPU x86_64 (24 cores)
Memory 62.7 GB

Core
├☑ mne 1.6.0.dev88+gb9e907667
├☑ numpy 1.26.0 (unknown linalg bindings (threadpoolctl module not found: No module named 'threadpoolctl'))
├☑ scipy 1.11.2
├☑ matplotlib 3.8.0 (backend=QtAgg)
├☑ pooch 1.7.0
└☑ jinja2 3.1.2

Numerical (optional)
├☑ nibabel 5.1.0
├☑ dipy 1.7.0
├☑ pandas 2.1.1
└☐ unavailable sklearn, numba, nilearn, openmeeg, cupy

Visualization (optional)
├☑ pyvista 0.42.2 (OpenGL 4.6 (Core Profile) Mesa 23.0.4-0ubuntu1~22.04.1 via AMD Radeon RX 570 Series (polaris10, LLVM 15.0.7, DRM 3.49, 6.2.0-34-generic))
├☑ pyvistaqt 0.11.0
├☑ vtk 9.2.6
Warning: Ignoring XDG_SESSION_TYPE=wayland on Gnome. Use QT_QPA_PLATFORM=wayland to run on Wayland anyway.
├☑ qtpy 2.4.0 (PyQt5=5.15.2)
└☐ unavailable ipympl, pyqtgraph, mne-qt-browser, ipywidgets, trame_client, trame_server, trame_vtk, trame_vuetify

Ecosystem (optional)
├☑ mne-bids 0.14.dev0
├☑ mne-connectivity 0.6.0dev0
└☐ unavailable mne-nirs, mne-features, mne-icalabel, mne-bids-pipeline
`

[larsoner]

Running this to try to rule some stuff out

MNE_COREG_HEAD_OPACITY=0.25 MNE_COREG_ADVANCED_RENDERING=false MNE_COREG_MARK_INSIDE=false mne coreg --subjects-dir=~/mne_data/MNE-sample-data/subjects -s sample -f ~/mne_data/MNE-sample-data/MEG/sample/sample_audvis_raw.fif --trans ~/mne_data/MNE-sample-data/MEG/sample/sample_audvis_raw-trans.fif

and then clicking "lock fiducials" and setting the EEG weight to zero (so the EEG electrodes disappear) it's still clear that many extra points are missing:

Will have to look into why this is the case... this is where the plotting happens if you want to look into it before I get a chance to in the coming days @nordme

mne-python/mne/viz/_3d.py

Lines 1280 to 1332 in c54d229

	if surf is not None and len(loc) > 0:
	defaults = DEFAULTS["coreg"]
	scalars, vectors, proj_pts = _orient_glyphs(
	loc, surf, project_points, mark_inside, check_inside, nearest
	)
	if mark_inside:
	from matplotlib.colors import ListedColormap
	color = np.append(color, 1)
	colormap = ListedColormap(
	np.array(
	[
	(
	0,
	0,
	0,
	1,
	),
	color,
	]
	)
	)
	color = None
	clim = [0, 1]
	else:
	scalars = None
	colormap = None
	clim = None
	mode = "cylinder" if orient_glyphs else "sphere"
	scale_mode = "vector" if scale_by_distance else "none"
	x, y, z = proj_pts.T if project_points else loc.T
	u, v, w = vectors.T
	return renderer.quiver3d(
	x,
	y,
	z,
	u,
	v,
	w,
	color=color,
	scale=scale,
	mode=mode,
	glyph_height=defaults["eegp_height"],
	glyph_center=(0.0, -defaults["eegp_height"], 0),
	resolution=16,
	glyph_resolution=16,
	glyph_radius=None,
	opacity=opacity,
	scale_mode=scale_mode,
	scalars=scalars,
	colormap=colormap,
	clim=clim,
	)

[nordme]

Hi!! I have pulled the fix, but I'm still getting the disappearing points.

Here's my mne sys_info output again:
$ mne sys_info
Platform Linux-6.2.0-35-generic-x86_64-with-glibc2.35
Python 3.11.5 | packaged by conda-forge | (main, Aug 27 2023, 03:34:09) [GCC 12.3.0]
Executable /home/erica/miniconda3/envs/sprint_2023/bin/python3.1
CPU x86_64 (24 cores)
Memory 62.7 GB

Core
├☑ mne 1.6.0.dev88+gb9e907667
├☑ numpy 1.26.0 (unknown linalg bindings (threadpoolctl module not found: No module named 'threadpoolctl'))
├☑ scipy 1.11.2
├☑ matplotlib 3.8.0 (backend=QtAgg)
├☑ pooch 1.7.0
└☑ jinja2 3.1.2

Numerical (optional)
├☑ nibabel 5.1.0
├☑ dipy 1.7.0
├☑ pandas 2.1.1
└☐ unavailable sklearn, numba, nilearn, openmeeg, cupy

Visualization (optional)
├☑ pyvista 0.42.2 (OpenGL 4.6 (Core Profile) Mesa 23.0.4-0ubuntu1~22.04.1 via AMD Radeon RX 570 Series (polaris10, LLVM 15.0.7, DRM 3.49, 6.2.0-35-generic))
├☑ pyvistaqt 0.11.0
├☑ vtk 9.2.6
Warning: Ignoring XDG_SESSION_TYPE=wayland on Gnome. Use QT_QPA_PLATFORM=wayland to run on Wayland anyway.
├☑ qtpy 2.4.0 (PyQt5=5.15.2)
└☐ unavailable ipympl, pyqtgraph, mne-qt-browser, ipywidgets, trame_client, trame_server, trame_vtk, trame_vuetify

Ecosystem (optional)
├☑ mne-bids 0.14.dev0
├☑ mne-connectivity 0.6.0dev0
└☐ unavailable mne-nirs, mne-features, mne-icalabel, mne-bids-pipeline

[larsoner]

Can you try with MNE_COREG_MARK_INSIDE=false in your env? That should at least show them

[drammock]

unrelated note:

Python 3.11.5 | packaged by conda-forge | (main, Aug 27 2023, 03:34:09) [GCC 12.3.0]
Executable /home/erica/miniconda3/envs/sprint_2023/bin/python3.1

executable is reported as python 3.1. I think this is not actually wrong (on my system too, there's a python3.1 in the env bin folder that is a symlink to python3.10; guessing that it's a shim put in by conda/mamba to deal with short-sighted minor version parsing code). But ideally the symlink should get followed before writing the path, i.e. Path(sys.executable).resolve().

[larsoner]

@nordme never mind, I can replicate on my macOS machine. It's an issue with the scalars affecting not just the color but also the glyph scaling even though we set scale=False somewhere. So when we use 0 and 1 to set things to use color0 and color1, we end up scaling the color0 points (inside the head) to zero. I'll investigate further... unclear why this doesn't happen on my Linux machine but does on macOS.

[nordme]

@larsoner Yeah, sorry I forgot to mention that it happens regardless of what MNE_COREG_MARK_INSIDE is set to.