[Eeglablist] Apply ICA matrix on other datasets

Malte Anders malteanders at gmail.com
Thu Aug 8 09:53:22 PDT 2019


Dear Community,

First, thanks to everybody who is taking his or her time to help people
like me. Much appreciated.

I have a really quick question:
I have datasets that are full of noise in higher frequency bands (>50 Hz)
because of the equipment we use besides the EEG (unfortunately, I can't
tell the hardware to stop doing that...). I have tried much, but that noise
can't be filtered out by no means except bandpass. Thus, without applying
lowpass filtering e.g. at 45 Hz, ICA/AMICA and MARA produce horrible
results because of that. With those bandpass settings, results are neat and
clean.

However, I still want to evaluate gamma frequencies around 40-80 Hz without
giving up on Artifact rejection.
My plan is to run Makotos preprocessing pipeline with aggressive settings
for highpass 1Hz/lowpass 40 Hz filtered data (e.g. ASR factor k=10, then
AMICA) and then extract the ICA matrix.

I then want to apply that ICA matrix to the same datasets filtered at 0.1
Hz and 80 Hz, as suggested. 2 questions:
-Is that a "good" plan?
-Should I still run ASR on that 0.1 Hz/80 Hz filtered datasets? I think
that if I run ASR on that, the "clean data" evaluated by ASR isn't really
clean and I have not found a solution to "export" ASR "settings" from 1
Hz/40 Hz filtered data to 0.1 Hz/80 Hz filtered data. Thus, when running
ASR on 0.1 Hz/80 Hz filtered data I am afraid I will loose too much
brain-related information.

Thank you all!
Malte


More information about the eeglablist mailing list