Link Search Menu Expand Document

A quick tutorial on ICA artifact rejection

EEGLAB is a powerful tool for eliminating several important types of non-brain artifacts from EEG data. EEGLAB allows the user to reject many such artifacts in an efficient and user-friendly manner. This minimalist guide is for non-EEGLAB users to import their EEG data, reject artifacts, then export the data back to a software package of their choice. For more comprehensive documentation on using EEGLAB, refer to the main sections of the EEGLAB tutorial.

Table of contents

1. Start MATLAB and EEGLAB, then import your data

Type >> eeglab to start EEGLAB under MATLAB.

Select menu item File → Import data to import your data file in any of a variety of file formats. See the Import data section for more details.

Scroll and check data using menu item Plot → Channel data (scroll).

2. Import a channel location file

Importing a channel location file is critical for visualizing the independent components of your data. Select Edit → Channel locations menu item.

  • Solution 1. If channel labels are present in your dataset, EEGLAB will try to look up channel locations based on their labels. Simply press Ok to look up locations.

  • Solution 2. If channel labels are not present, press the button Read locations in the bottom right corner of the channel edit window. Press Ok after selecting the file and then press Ok to have EEGLAB recognize the file format automatically from the file extension.

Press Ok in the channel edit window to import the channel locations into EEGLAB.

To check that your channel locations have been imported correctly, use menu item Plot → Channel locations → By name

3. Reject artifact-laden data

The quality of the data is critical for obtaining a good ICA decomposition. ICA can separate out certain types of artifacts – only those associated with fixed scalp-map projections.

These include eye movements and eye blinks, temporal muscle activity, and line noise. ICA may not be used to efficiently reject other types of artifacts – those associated with a series of one-of-a-kind scalp maps.

For example, if the subject were to scratch their EEG cap for several seconds, the result would be a long series of slightly different scalp maps associated with the channel and wire movements, etc. Therefore, such types of “non-stereotyped” or “paroxysmal” noise need to be removed by the user before performing ICA decomposition.

You have two solutions to reject bad data:

  • Automated solution: Select menu item Tools → Reject data using Clean Rawdata and ASR. Press the first checkbox to high pass filter the data and press Ok. A new window pops up to ask for a name for the new dataset. Press Ok.

  • Manual solution:

    • To reject “noisy channels“ of either continuous or epoched data, select menu item Edit → select data.
  • To reject noisy portions of “continuous data”, select menu item Tools → Inspect/Reject data by eye. Then mark noisy portions of continuous data for rejection by dragging the mouse horizontally with the left button held down. Press Reject when done. A new window pops up to ask for a name for the new dataset. Press Ok.

4. Run ICA and reject artifactual components

Although optional, we advise re-referencing the data to average reference using the Tools → Re-reference the data menu item.

Use menu Tools → Decompose data by ICA to run the ICA algorithm. To accept the default options, press Ok.

You then have two solutions to reject bad ICA components:

  • Automated solution:
    • Label components using the Tools → Classify components using IClabel → Label components menu item.
    • Classify components using the Tools → Classify components using IClabel → Flag components as artifact menu item.
    • Select menu item Tools → Remove components to actually remove the selected component from the data.
  • Manual solution:
    • Use menu Tools → Reject data using ICA → reject component by maps to select artifactual components. See the Data analysis (running ICA) tutorial for more details.
    • Select menu item Tools → Remove components to actually remove the selected component from the data.

See the Data analysis (running ICA) tutorial for more details and some hints on how to select artifactual components.

5. Further processing of and/or exporting the cleaned data

You may also apply a similar procedure to groups of datasets from the EEGLAB GUI, as explained in this tutorial.

Your data has now hopefully been pruned of its major artifacts. You may now proceed with further EEGLAB processing. You may also choose to export your data to the format of your choice.