[Eeglablist] Calculate ICA on short 'cleaned' epoch, then apply to long epoch

[Xiaoming Du]

Thanks Makoto. Your feedback is very helpful!

The main reason I want to use short epoch is that we only have 13 channels data. My concern is that if we feed data with various source of artifacts, 13 independent components might not be enough to capture the artifact we want for each subject. 

I think for now, I will have to use ICA on both long and short epochs, then compare the output. 

Thanks again!

Xiaoming


>>> Makoto Miyakoshi <mmiyakoshi at ucsd.edu> 6/8/2016 10:17 PM >>>

Dear Xiaoming,

> (1) I should feed ICA with 'cleaned' epochs (less non-stereotypic noise) and with enough data points (> k*N^2). 

It does not have to be epochs. You can feed long continuous data if you want.
Just in case, (EEG.nbchan^2) * k, where k>30.

> (2), I can calculate ICA on short epochs, then apply the ICA (weights and sphere matrices) to long epochs from the same dateset.

Sure.

> I want ICA focus on this short window to identify the artifact-components.

My impression is that feeding all data point is good after all. If the artifact is bad enough, ICA will pick it up. It's generally hard to imagine the situation (just from my experience though, no guarantee) that the artifact is so subtle that it can be ONLY decomposed in your suggested approach... It is because what we see and ICA sees in data are different, and our intuitive understanding of data is often wrong.


Makoto









On Tue, Jun 7, 2016 at 12:52 PM, Xiaoming Du <XDu at mprc.umaryland.edu> wrote:

Hi all,


I have a rookie question about using ICA to remove specific artifacts. Please correct me if I was wrong.


(1) I should feed ICA with 'cleaned' epochs (less non-stereotypic noise) and with enough data points (> k*N^2). 

(2), I can calculate ICA on short epochs, then apply the ICA (weights and sphere matrices) to long epochs from the same dateset.


(reference link: http://sccn.ucsd.edu/wiki/Chapter_09:_Decomposing_Data_Using_ICA#Studying_and_removing_ICA_components)


For my data set, I want to remove the known artifacts that occurs right after the events (within 100 ms). Also, there are only 13 channels (including 2 eye-channels) for each subject.

My goal is to remove this special artifacts using ICA on long epochs. However, because I already know the artifacts occurs shortly after events, I want ICA focus on this short window to identify the artifact-components. I want to minimize the effects of other artifacts or noise on ICA. 



My current thoughts are: 1, cut continuous file into long epochs (4 seconds). 

2, remove epochs with non-stereotypic noise.

3, 1-hz high-pass filter the remaining long epochs, then cut short epochs (-30 to 100 ms). In this way, the short epochs mainly contains the artifacts I want to remove. 

4, apply ICA on the 0.13-second short epochs. The sampling rate of our data is 1000 Hz. There are more than 50 trials. Therefore, The data point for ICA are around 1000*0.13*50 = 6500. This is larger than 5070 (30*13^2), so I should have enough data for ICA on those short epochs.

5, identify the artifact-components from step 4.

6, apply the ICA weights and sphere to long epochs (before 1-hz highpass filter), then reject artifact-components identified from step 5. 

7, for now, I should have long epochs that are artifact-corrected and not filtered. 


Please let me know if those steps are reasonable. Any suggestions or comments are appreciated!


Thanks. 

Xiaoming

_______________________________________________
Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
To unsubscribe, send an empty email to 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






-- 

Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20160609/199d97b3/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/gif
Size: 43 bytes
Desc: not available
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20160609/199d97b3/attachment.gif>