Remove channels with potential montage points from maxwell_filter_prepare_emptyroom #13208
Conversation
for more information, see https://pre-commit.ci
…ython into maxwellpepare
mscheltienne
requested review from
larsoner,
drammock,
agramfort and
dengemann
as code owners
| # just in case of combine MEG/other modality, let's explicitly drop other channels | ||
| # that might have a digitization and emit a warning. | ||
| picks = [elt for elt in ("eeg", "ecog", "seeg", "dbs") if elt in raw_er_prepared] | ||
| if len(picks) != 0: | ||
| picks = _picks_to_idx(raw_er_prepared.info, picks=picks, exclude=()) | ||
| idx = np.setdiff1d(np.arange(raw_er_prepared.info["nchan"]), picks) | ||
| raw_er_prepared.pick(idx, exclude=()) | ||
| if len(raw_er.ch_names) != len(raw_er_prepared.ch_names): | ||
| warn( | ||
| "The empty-room recording contained EEG-like channels. These channels " | ||
| "were dropped from the empty-room recording, as they are not " | ||
| "compatible with Maxwell filtering. If you need to, add those channels " | ||
| "back after the execution of 'maxwell_filter_prepare_emptyroom' from your " | ||
| "original empty-room recording using 'raw.add_channels'." | ||
| ) |
There was a problem hiding this comment.
The more I think about it, the more I feel like it should simply be .pick("meg", exclude=()). I can't think of a reason to add those channels back.
There was a problem hiding this comment.
Not sure about adding the EEG channels back, but regarding .pick("meg", exclude=()) there are other channel types that you could reasonably want to have around (e.g., stim, CHPI status, active shield status, etc.) and it's hard to know every end-users use case / needs, hence the use of a blocklist approach (eeg-like) here rather than an allowlist (meg) approach.
There was a problem hiding this comment.
I'm not fully convinced this is the best fix, if anyone has a better proposal, please advise.
I think at UW we have recorded empty room data with EEG enabled simply because we were doing simultaneous M/EEG and didn't change the acq settings (can't remember if it was an error or standard practice anymore, but I know it happened at least sometimes!).
I am inclined to add a drop_eeg=True as a new kwarg as a bugfix since I think it's what people will want most of the time. But we could allow them to set it to False and then figure out how to fix the MF stuff if it crops up again for people who want drop_eeg=False for some reason.
Exactly what happened in GVA as well.
(2) is simpler, but I can look into (1) if you prefer, starting by adding a test and fixing the prepare ER function. |
|
I think the MF code should ideally not care about the presence or absence of EEG channels. For subject data they pass through unchanged so I don't see why they wouldn't for empty-room data as well. So (1) seems better/preferable |
|
Interesting that mne-tools/mne-bids-pipeline#1040 actually picked the MEG channels and removed the rest, taking the allowlist approach. |
|
MNE-BIDS-Pipeline has the luxury of being a bit more opinionated than MNE-Python because MNE-BIDS-Pipeline knows how it's internally going to use its outputs downstream to a larger extent than we can safely assume on behalf of our end users in MNE-Python 🙂 But it's also possible that we shouldn't have done that there either ... |
Following https://mne.discourse.group/t/montage-error-in-preparing-empty-room-for-maxfilter-for-simultaneous-meg-eeg/11089
In Geneva, an acquisition with combined MEG / EEG was done while preserving the EEG channels enabled. If you blindly apply
maxwell_filter_prepare_emptyroomyou get a failure during theget_montage(...)/set_montage(...)step. Beside the fiducial digitization, I can't think of an information from theDigMontagerequired by MaxWell filter to be set on the ER. Pointing in that direction, the existingtestare removing EEG channels from the test files prior to runningmaxwell_filter_prepare_emptyroom.This PR adds a tiny piece of documentation about it; and drops EEG (and ECOG, sEEG, DBS channels which might also not played nicely since they would contain some un-needed electrode location information) in
maxwell_filter_prepare_emptyroom. It also adds a warning informing that this channel drop happens; maybe you will need to add them back at a later stage to the ER, e.g. to run ICA if you fitted it on combined MEG / EEG?I'm not fully convinced this is the best fix, if anyone has a better proposal, please advise.