[Eeglablist] ICA pipeline questions

[Samran]

Dear all,

I have 64 channel EEG data sampled at 2048Hz (SEPs and resting state EEG).
I am going to apply ICA to remove artifacts and planning to make a pipeline
for data processing. I have read from multiple sources about how to use
ICA, but I am a bit confused, and would like to verify if I have understood
the usage of ICA correctly. Following are the 2 pipelines I have come up
with after reading about ICA from different tutorials.

Pipeline 1:
1. Load data (e.g. after running PREP pipeline)
2. Highpass filter 1Hz
3. Epoch
4. Reject Epochs
5. Downsample to 256Hz
6. Run ICA (with PCA, if any channels were interpolated)
7. Determine bad ICs
8. Load data (e.g. after running PREP pipeline) again
9. Highpass & lowpass filteration (if required)
10. Epoch
11. Apply ICA matrix calculated in step 6.
12. Remove bad ICs determined in step 7.
13. Epoch rejection (Is this required?)
14. Average epochs

Pipeline 2:
1. Load data (e.g. after running PREP pipeline)
2. Highpass filter 1Hz
3. Downsample to 256Hz
4. Run ICA (with PCA, if any channels were interpolated)
5. Epoch
6. Reject Epochs based on IC activations
7. Run ICA again on the pruned data (with PCA, if any channels were
interpolated)
8. Determine bad ICs
9. Load data (e.g. after running PREP pipeline) again
10. Highpass & lowpass filteration (if required)
11. Epoch
12. Apply ICA matrix calculated in step 7.
13. Remove bad ICs determined in step 8.
14. Epoch rejection (Is this required?)
15. Average epochs

I would like to know if any of the above the 2 pipelines is correct? If
not, which step(s) in the pipeline is to be modified or get its position
changed. Have I mixed up things, and these pipelines are to be merged
together (e.g. Pipeline1 steps 1 to 6 and Pipeline2 steps 6 to 15)?

I would also like to know:
1. Is running ICA after rejection of epochs correct (step 4 in Pipeline1,
and step 6 in Pipeline2), as it will introduce discontinuity in the data.
Although it has been mentioned in tutorials, e.g. in Makoto's preprocessing
pipeline. I am missing some concept behind this as why this discontinuity
will not affect ICA results.
2. Is it required to reject epochs after removing bad ICs from the data
(step 13 in Pipeline1, and step 14 in Pipeline 2) because the epochs
rejected in step 4 of Pipeline1, or step 6 of Pipeline2 will still be
present in the data.
3. I would like to confirm if I understood fixing the rank of the data
correctly when running ICA.
3a. If I have 64 channel average referenced data, will I give the PCA
parameter = 63 in the runica function?
3b. If I have 64 channel data with 2 channels interpolated and is average
referenced, will I give the PCA parameter = 61 in the runica function?

Looking forward to the suggestions and comments. :)

Thanks.

-- 
Regards,


M. Samran Navid.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20170910/98364595/attachment.html>