Bad electrode interpolation

[ashish gupta]

Hi Mike, 

Independent Component Analysis of EEG data is faced with two challenges:

(a)  Running ICA on EEG data with bad channel reduce the quality of ICA by introducing noise in the data.

(b)  Bad electrodes interpolated before ICA introduces non-linearity (due to spherical interpolation) in the EEG data which also reduces the quality of ICA solution.

 

We ran ICA on the good channel after removing the bad electrodes for it safeguards the ICA decompositions from both noise (due to bad electrode) and non-linearity (due to interpolation of bad electrodes). 

question 1> Is the above reason of running ICA on good channel and interpolating bad channel afterwards sufficient.... are they more reasons to support it ?

question 2> you have also recommended at several place  (1) remove bad electrodes, (2) run ICA, (3) interpolate missing electrodes. this sequence.. I would like to know if they are some paper  of yours where this sequence of processing, you have used .....as reference for me to look at it.

question 3> My purpose of ICA is to remove artifiacts so I feel the order should not matter much...is this right ?  

Question 4> Is there somebody who have done some kind of comparative study ?

question 5> I have in one of my paper used the sequence as   (1) remove bad electrodes, (2) run ICA, (3) interpolate missing electrode.. but generally many protocol recommend the interpolate bad electrodes and then run ICA like 

Makoto's preprocessing pipeline

 so I want to know the above reason which I gave is sufficient to justify the sequence I have used .....to any researcher...has somebody already used it....and testified it... 

waiting eagerly for your fast reply

Ashish

[Mike X Cohen]

Hi Ashish. See below.

On Thu, Jul 19, 2018 at 4:40 PM, ashish gupta <ashish.gu...@gmail.com> wrote:

Hi Mike, 

Independent Component Analysis of EEG data is faced with two challenges:

(a)  Running ICA on EEG data with bad channel reduce the quality of ICA by introducing noise in the data.

(b)  Bad electrodes interpolated before ICA introduces non-linearity (due to spherical interpolation) in the EEG data which also reduces the quality of ICA solution.

 

We ran ICA on the good channel after removing the bad electrodes for it safeguards the ICA decompositions from both noise (due to bad electrode) and non-linearity (due to interpolation of bad electrodes). 

question 1> Is the above reason of running ICA on good channel and interpolating bad channel afterwards sufficient.... are they more reasons to support it ?

Yes, that's the standard approach (remove channels, ICA, interpolate).

 

question 2> you have also recommended at several place  (1) remove bad electrodes, (2) run ICA, (3) interpolate missing electrodes. this sequence.. I would like to know if they are some paper  of yours where this sequence of processing, you have used .....as reference for me to look at it.

No, I don't have anything written. Possibly it's in the book. But this is a standard recommendation, which you can also find on the eeglab email discussion list.

 

question 3> My purpose of ICA is to remove artifiacts so I feel the order should not matter much...is this right ?  

How much the above order matters depends on what you are doing with ICA. I've also run ICA with bad channels included and it will still isolate a good blink component. If you plan on analyzing and interpreting the components, then it's good to be careful with how the ICA is run, and follow whatever best-practice guidelines the eeglab list recommends. 

If you are only using ICA to remove some artifacts (what most people do), then the exact order doesn't really matter. I don't mean to suggest that preprocessing or ICA should be done in a sloppy way, but the thing is that the blinks are so prominent and non-Gaussian that ICA will easily identify a blink component regardless of the state of the data.

 

Question 4> Is there somebody who have done some kind of comparative study ?

Not that I know of. But the eeglab list would know the answer.

 

question 5> I have in one of my paper used the sequence as   (1) remove bad electrodes, (2) run ICA, (3) interpolate missing electrode.. but generally many protocol recommend the interpolate bad electrodes and then run ICA like 

Makoto's preprocessing pipeline

 so I want to know the above reason which I gave is sufficient to justify the sequence I have used .....to any researcher...has somebody already used it....and testified it... 

See my comments above. 

Mike

 

waiting eagerly for your fast reply

Ashish

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.
To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimeseriesdata+unsub...@googlegroups.com.
Visit this group at https://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
Fresh look: mikexcohen.com