[FieldTrip] Identifying bad channels

Poppy Watson popwatson at hotmail.com
Mon Jan 14 10:03:59 CET 2019

Dear fieldtrippers,

I’m following our lab protocol for pre-processing of EEG data – but implementing it in Fieldtrip rather than Brain Vision Analyzer. So far things are great. However, I’m going back to preprocessing stage because ideally I would like to stick as closely to the procedure we use for automated artifact/channel rejection in BVA but am wondering about the most efficient way to implement this in fieldtrip.

What I have done in fieldtrip is:

1.      Remove channels that were marked as bad during recording and interpolated them.

2.      After lots of playing around, find zvalue cutoff settings that I’m comfortable with and use the same settings on every participant for jump artifacts, muscle artifacts and eye movements and remove these using ft_rejectartifact (conservative but easily replicable).

3.      I then quickly scrolled through each trial using ft_rejectvisual to make sure all channels looked ok – any really messy noisy channels I then made a note of and repeated steps 1 and 2 above.
I’m not happy with this step 3 however, because it is quite subjective.

The way that we identify bad channels in BVA which I would ultimately like to implement is:

1.      Mark trials as bad where a channel is present that has:

a.      Low activity of less than 0.5 microvolts in any 100ms interval

b.      Noisy activity where the absolute max-min voltage in any 200ms interval is greater than 200 microvolts (sometimes this is 100Uv depends on expected motor activity in paradigm).

2.      Identify, for every channel, the percentage of data that is being marked as bad – if this is higher than a given threshold (i.e. 25%), then remove the channel and/or interpolate.

3.      Rerun step 1 and remove bad trials.

What I am planning to do in Fieldtrip, but would appreciate some input from those more experienced is:

1.   Remove trials with eyeblinks etc using the Fieldtrip methods described above

2.  Split each trial segment into 100ms or 200ms intervals

3. Calculate the max-min per channel in order to identify intervals that have low voltage or noisy activity etc as per a and b above

4. Keep track of the proportion of trials/intervals that any given channel is identified as bad so that I can remove those channels that are above a threshold (i.e. 25%).

Is there a more efficient way to implement something like this? Am I missing out on a great Fieldtrip tool?

Please advise


Poppy Watson

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20190114/a4f8c508/attachment-0001.html>

More information about the fieldtrip mailing list