This subjects ICA components have left me with an uneasy feeling

đź’¬ Support & Discussions
1

MNE 1.3.0
MacOS BigSur 11.6.7

image
image1876Ă—1300 422 KB

Have you ever seen something so freaky? It’s like this subject’s brain activity is sending a message from another dimension. Here’s my code & data:

import mne
from mne.preprocessing import ICA
raw = mne.io.read_raw_eeglab('data/participant/sub-03_task-rsvp_continuous.set', preload=True,
                             montage_units="dm")
raw.set_montage("standard_1005")
raw.filter(0.01, 100.0, method='iir', n_jobs=-1)
filt_raw = raw.copy().filter(l_freq=1., h_freq=None)
ica = ICA(n_components=63, max_iter='auto', random_state=97)
ica.fit(filt_raw)
ica.plot_components()

Data from a freely available source:
https://openneuro.org/datasets/ds003825/versions/1.2.0
This is subject 3 in derivatives/eeglab.

1 Like
2

:smile: This looks scary indeed! I’d start with verifying that the raw data looks sane and doesn’t contain huge artifacts. I sometimes get such components when there is an extremely large spike or other weird stuff somewhere in the data. I also wonder why you have Cz in your topoplots when this is supposed to be the reference channel (according to the docs).

1 Like
3

Wow nice ones ahah
I would add that you could remove the reference from the inst used to fit the data, if it is flat. i.e. keep it if you referenced the data e.g. to a common average. And secondly, you could set n_components=None, you might be fitting too many components.

Mathieu

1 Like
5

Thanks! Yeah I tried with a small, random sub sample of the raw data and they then look completely different. I find that the raw.plot() is quite slow and sometimes crashes though, which means I can’t really scroll through the entire raw to look for components, do you maybe have any tips for finding particularly bad chunks of raw data?

6

Great thanks, the data was referenced to an average reference. I cut off the first and last 500 seconds from the raw and this gets rid of the disturbing plots, so now I can finally sleep at night :smiling_face:

2 Likes