[Eeglablist] ICA data length
arno at salk.edu
Wed Mar 31 16:10:08 PST 2004
> I have some queries on ICA which I hope you can help with.
> I have 12 channel EEG data and I am working on removing respiration
> and EKG
> artifacts from my data automatedly. One of the techniques that I'm
> exploring is
> ICA. I perform ICA on the data in the following manner:
> 1. seperate data into epochs of 5 sec (12 channel data sampling rate:
> 185 Hz)
> 2. remove bad epochs i.e transients , pop artifacts etc by time domain
> 3. perform ICA on remaining epochs individually
Do you mean you apply ICA on each epoch? You should apply ICA to the
collection of all concatenated epochs (but I guess this is what you are
> 4. Determine components containing respiration/EKG artifacts by
> matching their
> freq with freq of reference respiration/ekg channel
> 5. project remaining components to get the clean epoch.
> Do you think this method is correct?. In particular, is the data
> length (12 x 925 points) sufficient to give 12 components?. I have
> tried using longer epoch lengths
This method is theoretically correct. It then depends on which component
you are removing and if you have correctly identified them. It is better
to use longer epoch lengths (as long as they do not overlap of course)
to obtain more stable ICA decompositions.
> but it does not help much in giving cleaner more distinct components,
> because the artifacts are fleeting i.e the respiration/EKG artifacts
> are not
> present throughout the data but they come and go.
These artifacts (due to their nature) are probably present throughout
the whole data (with different amplitude modulations). However, if you
fail to find them (since you have very few channels, this is quite
possible), you should run ICA only on the portion of data where you can
detect them visually.
*Arnaud Delorme, Ph.D.*
Computational Neurobiology Lab, Salk Institute
10010 North Torrey Pines Road
La Jolla, CA 92037 USA
*Tel* : /(+1)-858-458-1927 ext 15/
*Fax* : /(+1)-858-458-1847/
*Web page *: www.sccn.ucsd.edu/~arno <http://www.sccn.ucsd.edu/%7Earno>
*To think upon*:
All firm belief is a sickness.
More information about the Eeglablist