SNR estimate for single trials

✉️ Mailing List Archive (read-only)
1

Hello everyone,

I want to ask what would be the recommended SNR amplitude estimate when
computing single trial source estimates with
mne-python(apply_inverse/apply_inverse_epochs)?

I looked at this tutorial
http://martinos.org/mne/stable/auto_examples/inverse/plot_compute_mne_inverse_epochs_in_label.html

and for single trial SNR is kept the same as for the evoked data -
default 3. Would it not make sense to reduce it to less (e.g. 1, I have
been using previously with mne-c) if we know that single trials are much
noisier?

Thanks for your help,
Ana

2

hi Ana,

yes 1 is typically what is used if you estimate things on single trials.

I think we should update this example to clarify / change this.

See https://github.com/mne-tools/mne-python/issues/4131

Alex

3

Hi Alex,

Thanks for the clarification.

Ana

4

Hi Ana and Alex,

I am not so sure about specifying a low SNR for single trials... Yes, it's true of course that single trials are noisier, but often single trial processing in the source space is followed by combining across trials.. So in many (but not all) instances, it is useful to have a consistent inverse operator (specifying a fixed SNR... and for dSPM scaling the noise cov by a fixed number of trials).

For instance, if you did inverse first on single trials and then averaging in source space vs. average across trials in sensor space first and then do inverse, you would get different answers if you specified different SNRs for the two cases.

This issue came up before, and the 3.0 was used to make them consistent:

https://mail.nmr.mgh.harvard.edu/pipermail//mne_analysis/2014-April/002066.html

https://github.com/mne-tools/mne-python/pull/1237

Just something to keep in mind depending on how you are using single trial inverses.

- Hari

5

Hi All,

I second Hari, I always used 3.0 in my old Matlab stream for processing
single trial data for event related connectivity.

Thanks

Sheraz

6

ok fair enough. Can we just add some comments in the examples then?

Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20170402/fb357c87/attachment.html