[spatiotemporal cluster tests sensors]

✉️ Mailing List Archive (read-only)
1

External Email - Use Caution

Hi mne peeps,

this is quick inquiry regarding the timeline and/or efforts needed to
enable such stastical approaches in mne-python. Is there python code that
gets close to it on which we could contribute (I think it is available in
fieldtrip) or does it require more intense integration?

Thank you for the info! (I launch this email because 3 studies would
currently benefit from this)

Cheers,
Virginie van Wassenhove
*https://brainthemind.com/*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190603/2fcfb69f/attachment.html

2

External Email - Use Caution

In terms of spatio-temporal cluster tests across sensors -- I'm not sure
what in particular you are looking for, but here is an example of a
spatio-temporal EEG sensor cluster test contrasting two conditions:

http://mne-tools.github.io/mne-biomag-group-demo/auto_examples/statistics/plot_sensor_spatio_temporal_cluster_stats.html

Or one with ANOVA:

https://mne-tools.github.io/stable/auto_tutorials/stats-sensor-space/plot_stats_spatio_temporal_cluster_sensors.html

We have some plans to simplify these sorts of things in MNE, since right
now they do require juggling NumPy arrays a bit. Nobody has started it as
far as I know but here is the issue for discussion:

https://github.com/mne-tools/mne-python/issues/4859

Christian Brodbeck maintains a separate repository/project called Eelbrain
that has some more generalized statistical functions that might also make
things easier (I have not personally tried it yet):

https://eelbrain.readthedocs.io/en/stable/reference.html#module-testnd

Eric

3

External Email - Use Caution

Hi Eric,
What would be really useful for our work is a 3d cluster test (frequencies,
time, sensors), as already discussed here
<https://github.com/mne-tools/mne-python/issues/5144>.
Currently, we are forced to a-priori divide and average the data in
pre-defined frequency bands, which, as we know from the literature, are
somewhat arbitrary.
Letting the data speak as to where the effects are would be the better
approach.

To add to this discussion
<https://github.com/mne-tools/mne-python/issues/4859>, I personally think
mne's linear regression function used with a flexibly defined design matrix
(SPM-style or LIMO-toolbox) is the most versatile approach.

Let us know if our input could be of help!
Best,
Sophie