[Eeglablist] Slow cortical potential analysis-preprocessing pipeline

[Ahmad, Jumana]

I read that:

§  You need minimum of (number of channel)^2 x 20 to 30 data points to perform ICA. The number ’20 to 30’ should (exponentially) increase as the number of channels increases. This is the recommended length of the data by past publications from SCCN. This is purely an empirical values, and we have not performed a systematic investigation on it.
Thanks again to Makoto for his helpful pre-processing pipeline!
Jumana

From: Rafał Jończyk [mailto:rafal.jonczyk at gmail.com]
Sent: 10 February 2017 07:55
To: mmiyakoshi at ucsd.edu; Ahmad, Jumana <jumana.ahmad at kcl.ac.uk>
Cc: eeglablist at sccn.ucsd.edu
Subject: Re: [Eeglablist] Slow cortical potential analysis-preprocessing pipeline

thank you very much Jumana and Makoto for your feedback! This is very helpful. I think I will also downsample to 250Hz to make EEG files lighter and potentially further improve ICA (as suggested by Makoto in his pipeline). Out of curiosity, is there any risk of making ICA worse rather than better by downsampling the data (given that we don't go lower than 250Hz)?

Best,
Rafal

2017-02-10 4:30 GMT+01:00 Makoto Miyakoshi <mmiyakoshi at ucsd.edu<mailto:mmiyakoshi at ucsd.edu>>:
Dear Rafal,

At least, you can highpass filter the data for ICA purpose only. You can copy the weight matrix to the unfiltered data. Please consider doing it.

> The problem I have, however, is that ICA is not advised to be run on interpolated channels. Is there a possible compromise/solution to this issue?

In most cases, ICA will pick up the rank deficiency and automatically adjust the rank. So you'll be probably ok.

Makoto



On Tue, Feb 7, 2017 at 6:48 AM, Rafał Jończyk <rafal.jonczyk at gmail.com<mailto:rafal.jonczyk at gmail.com>> wrote:

Dear eeglab(list) users,

I’m currently analysing a 64-channel dataset looking at a slow cortical potential – Stimulus Preceding Negativity (each epoch lasts 3800 ms). I'm using eeglab and erplab. I’ve been thinking which processing steps would be best to get the a reliable dataset, i.e., representative of what’s really going on without unintentionally getting an effect due to incorrect preprocessing analysis.

Below I present the pipeline I’ve been experimenting with so far. Steps marked with „*” are done in erplab. I would really appreciate your feedback, because I’m still relatively new to EEG analysis. Also, I shortly discuss the issue of channel interpolation/ICA below the pipeline.

1. Re-sample from 1000Hz to 500Hz

2. Edit channel locations + append Cz + re-reference to Cz

3. Filter: lowpass, cut-off at 20Hz, order 2 (12 dB/Octave)*

4. Epoch [-0.105 3.805]

5. Automatic Epoch Rejection (based on probability) + rejection based on visual inspection

6. ICA (runica)

7. Removing ICA components: only components related to eye movements + visual inspection

8. Convert epochs to a continuous dataset*

9. Creating event lists & assigning bins*

10. Epoching: [-100 3800]*

11. Baseline removal [-100 0]*

12. Artefact rejection (peak-to-peak moving window)*

13. Averaging + visual inspection*

14. Global average*

##Comment concerning Step 2: I assume it would be better to remove bad channels after re-sampling, then re-reference to Cz with bad channels removed, and then interpolate bad channels from previous dataset (one with bad channels included) so as to avoid puting noise into the average. This will be followed by filtering in step 3. The problem I have, however, is that ICA is not advised to be run on interpolated channels. Is there a possible compromise/solution to this issue?

I would be very eager to follow the preprocessing and data cleaning advice suggested in Mokoto’s pipeline, PREP pipeline, or CleanLine plugin more closely, but their approach is largely based on high-pass filtering which would eliminate the slow drifts I’m looking at.

Thank you for your insights!
Best,
Rafal

--
Rafał Jończyk
Assistant Professor
Faculty of English
Adam Mickiewicz University, Poznań | Poland

_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html<https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fsccn.ucsd.edu%2Feeglab%2Feeglabmail.html&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7C0bd2a89d7ed54f34d74308d4518a0f2e%7C8370cf1416f34c16b83c724071654356%7C0&sdata=Nw3C4e2WwKdG%2FcZ1mvCu9lhgrxGuOKASx%2BIdKR2bLro%3D&reserved=0>
To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu<mailto:eeglablist-unsubscribe at sccn.ucsd.edu>
For digest mode, send an email with the subject "set digest mime" to eeglablist-request at sccn.ucsd.edu<mailto:eeglablist-request at sccn.ucsd.edu>



--
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego



--
Rafał Jończyk
Assistant Professor
Faculty of English
Adam Mickiewicz University, Poznań | Poland
[https://docs.google.com/uc?export=download&id=0B-bMuTu3fxATSnhVb0dFTUV0MVU&revid=0B-bMuTu3fxATeElUV05hNmRjVEJZSHBxNzV6bExNemhRYWFzPQ]<https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.springer.com%2Fus%2Fbook%2F9783319476346&data=01%7C01%7Cjumana.ahmad%40kcl.ac.uk%7C0bd2a89d7ed54f34d74308d4518a0f2e%7C8370cf1416f34c16b83c724071654356%7C0&sdata=N2ovkM2D1MY2DmjJWqCSEGk8XcPL0yzeUJqpYaRrzrE%3D&reserved=0>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20170210/3f84da4f/attachment.html>