mne-tools
interactive sensor-level field patterns browser
Our interactive source estimate viewer is so awesome, it makes me wish for an equally awesome sensor level viewer. For this, the venerable proprietary "Xfit" program that ships with the MegIn scanner may serve as inspiration.
With MEG, the raw sensor timecourses are quite difficult to interpret: gradiometers and magnetometers measure the field in different directions. The choice made by Xfit to make the field patterns the central visualization is a good one. See a short screencast of Xfit in action, here:
https://gfycat.com/jointaromatichorsemouse
For MNE-Python, I envision an interface similar to the source viewer, but with the field patterns instead of the pysurfer brain. On the bottom, we have the GFP timecourse.
What would be super nice would be the ability to "drop a dipole". Similar to clicking a vertex in the source viewer, have the ability to fit a dipole at the currently selected time, and see the dipole timecourse.
As extra bonus, it would be great if we would select a subset of channels and view the field patterns generated by only those sensors, and fit a dipole using only those sensors.
maybe a challenge for @GuillaumeFavelier ?
really cool video. It helps a lot to demo.
really cool video. It helps a lot to demo.
It looks cool! :+1:
When I have the cycles this week, I will prototype something but I might need help to guide me though :sweat_smile:
Some aspects of this are a special / enhanced case of "showing sensor data with the brain" -- we discussed this before at some point but I can't remember where. Along those lines I would propose (probably in PR-by-PR order):
- A
brain.add_sensor_data, where MEG sensors and EEG sensors are shown as inplot_alignment, but get colors according to their activity level. This is basically "just" parts ofplot_alignmentwith some color mapping and time-point-updating built on top. - Some sort of surface mapping option, that allows you to map MEG sensor data to the MEG helmet (what @wmvanvliet is talking about) and EEG sensor data to the scalp, each with field map lines. This is "just" the functionality of
plot_evoked_fieldmapped into theBrainobject (plus interactivity). - GUI options for toggling these sensors and surface mappings on the fly.
Each of these I'd expect just to be 200-300 lines of code. If this makes sense to everyone, @GuillaumeFavelier feel free to move these points to #7162, and then we can discuss what I think is the trickier part:
- Some way to "place a dipole" and do
mne.dipole_fitwith it
What interaction allows us to define it, clicking? Do we constrain to a surface if a surface is shown? Is it allowed in volume mode? How do we add/delete, etc.? @wmvanvliet hopefully you have some ideas here. It's a great use case but not as straightforward to implement in my mind compared to the other things above.
The idea of using Brain to do some sort of interactive dipole fitting is cool, but it almost seems like a sufficientlry different use case to justify a separate class or GUI (that uses Brain), like we have for coreg. Doing a single (or multi?) interactive dipole fit on the data involves some different choices from visualizing already-localized activations, such as BEM/cov/trans/src (if you want one or not), etc., so it might make sense to separate out the functionality a bit. EDIT: same applies to sensor subselection for fitting, this is a great case for a different class and UI.
cc @adam2392 the above three points at least are relevant for sEEG + ECoG as well
Dipole fitting certainly is a tricky use case (and the main reason Xfit exists) that could certainly use a dedicated GUI.
I don't see how coloring sensors to their value is very useful. Sure it is "easy" to implement, but if it's not super useful it's not worth doing. The field maps are the really useful thing. Interesting idea to add this to the Brain object.
I don't see how coloring sensors to their value is very useful. Sure it is "easy" to implement, but if it's not super useful it's not worth doing.
It's been our ECoG example for quite some time