Check for ICA (toward Epochs Rejection)

[richard (Richard Höchenberger)]

@BaggioMarco

Regarding the ICA, I tried to replace the “fastica” with “picard” in my code but I received the error:
ModuleNotFoundError: No module named 'picard'

If you’re using conda (as suggested in the MNE installation instructions), you can install picard via

conda install -c conda-forge python-picard

Alternatively, you can install it via pip:

pip install python-picard

@BaggioMarco

By the ICA I try to reconstruct the signal in occurrence of eye blinks, while with the criteria added in the epoch section I just can obtain back the “seconds with no artifacts”.

To remove eye blinks, you’ll want to follow the following approach:

• epoch your data and run ica.fit() on your epochs to get the components
• go back to your raw data, identify eye blinks and create epochs around the blink events. MNE-Python can do this automatically for you using mne.preprocessing.create_eog_epochs()
• using the ICA fit you created in the first step, run ica.find_bads_eog() on the EOG epochs. This will return a list of ICs that appear to be related to ocular activity
• You can review the effects of excluding those components by adding them to ica.exclude and running ica.plot_overlay() on the averaged EOG epochs, i.e., ica.plot_overlay(eog_epochs.average())
• Run ica.apply() on your epochs to remove ocular artifacts.

If you’re brave & curious, you can have a look at how we’re handling it in the MNE Study Template; but be warned, the code is slightly more complex than what I’ve explained above.