I am currently working on an EEG dataset with 5 EEG channels, 2 EOG channels, and 1 ECG. Shown in the image below is the data for one of the subjects.
For data looking like this, what is the source of noise (like line noise, channel noise)?
Secondly, after rejecting some ICA components, is that data useful, or should the whole data be discarded?
Thanks in advance.
Could you share the file?
This artifact is very weird, it kind of looks like it’s on the reference. ICA might not be possible on a dataset with only 5 channels… but maybe a simple common average reference might drop it if it is indeed on the reference…
Mathieu
Thank you for your reply. I have attached the link to the raw file here. It is saved in a pkl (pickle) file format; you just need to load it using pickle.load().
And about ICA, I have used it for the other subject’s data and it works pretty well for most cases.
Please let me know if you can access the data, and the reason for such noisy data.
Best,
Poorva
Hello,
No I can not access the data, you also need to set the read permission.
Also, try not to use pickle to store data on the long term. This is not what it is designed to do, and it will cause you headaches in the future. You can save a raw object with raw.save().
ICA with only 5 channels will fit very few components which will capture both noise and brain activity. Thus, you will likely remove both noise and brain information simultaneously when zeroing a component.
Mathieu
Hi,
Please try accessing the data now saved in fif format here. I have changed the sharing settings. And yes, saving in pkl is just temporary to share on the MNE forum. About ICA, yes, that is the problem with data with few channels. What can be the solution for artifact removal in such cases?
In some cases, ICA worked well, but in others, I lose brain activity data as well.
You can read the data using the code below:
from mne.io import Raw
raw = Raw(‘raw_n40.fif’)
print(raw)
print(raw.info)
raw.plot(scalings=0.03, n_channels=8);
Thanks,
Poorva
I’m not convince there is any brain data recorded in this file. It really doesn’t look like it, and it doesn’t look like it’s a common noise on the reference.
Also, please load your data with one of the reader, either mne.io.read_raw (chooses the correct reader automatically based on the file name) or mne.io.read_raw_fif (for FIFF format).
For the processing on recordings with little channels, you could try SSP or Regressors: Repairing artifacts with regression — MNE 1.4.0.dev65+g43979006d documentation
Mathieu
Okay. Thank you for the insight on the data! Will check this.
Thanks,
Poorva
Please check if the reference was connected well. I saw this kind of symmetrical artifact in one of my data when I forgot to connect the earlobe references. At that time I was using the OpenBCI Ultracortex.
Also, if you can share the details of your data acquisition hardware, that may help in understanding the origin of the issue.
P.S. Pardon if you already mentioned it.
Thanks,
Sagar.
Hi,
Thank you for your reply. Yes, that was one of the major reasons I thought as well for such data, could be a problem with the reference. To answer your question, AcqKnowledge 4.1 software (BIOPAC, Inc.) was used to collect the data.
Poorva
Hello,
I have a raw EEG data. How do I know that it is noisy or not
Hello @Ammara1 and welcome to the forum!
Whether the data is noisy can usually be assessed by visual inspection. It’s helpful to sit together with an experienced colleague who can show you common artifacts in your data.
For automated bad channel detection, you can use tools like pyprep. Manual visual inspection is still required even when using such tools.
Best wishes,
Richard
@richard
Thanks a lot
