[Eeglablist] gamma high pass filter before ICA for connectivity analysis

[Amin Fi]

 Hi,

I'm somehow new to EEG processing, and I'm trying to pre-process my data
with eeglab which is collected from an experiment that subjects are free to
move their head. The setup is a 32-channel wired EEG cap. So the raw data
has many EMG artifacts, mostly in beta and gamma bands. I'm using ICA to
remove the artifacts, and since I have only 32 ICs per subject, the
cleaning is not optimal and much of the information will be removed
especially in lower frequency bands along with removed ICs. The collected
data for each subject is about 3-4 minutes, and I'm trying to do some
task-related time-frequency and non-directed connectivity analysis in scalp
space.
After some trial and error in preprocessing steps and reading other
people's advice in eeglab mailing list or other sources, I concluded that
if I split the data with band-pass filter into two parts (1-32hz and
28-80hz) and then run ICA for each of them independently and analyze gamma
bands (30-80hz) from the second part, the fewer data will remove in lower
frequencies and the gamma frequency will be more artifact-free. So my steps
are:

- High pass data to 1hz
- For the first part, low pass the data to 32hz, and for the second part
band pass to 28-80hz with default eeglab FIR filter
- Notch filter for the second part (48-52hz) for removing line-noise
- Remove bad segments of the data
- Running ICA with AMICA for each part and remove bad ICs
- Surface Laplacian

I want to be sure if my procedure is correct and if this method doesn't
make any problem in further analysis (do you think the 28hz high pass
filter makes any problem with time-related or phase-related connectivity
analysis?) since I understood it's not very conventional, but I got better
results in some rudimentary power-spectrum analysis. I would be appreciated
your comments.

Thank you so much.