Bad channels

Why do I need this step?

One bad channel harms all the data. In an average reference its noise enters every channel. In ICA it takes components for itself instead of the real sources.

So you find bad channels now, record their names, and remove them. In the next lesson you rebuild them by interpolation.

How do I do it in EEGLAB?

Start with the main dataset D1 hp0.1 notch60 loaded, as in the Starting point.

First: look at the signal

1. Open Plot › then Channel data (scroll).
2. Scroll through the whole recording, not just the first few seconds. Some channels go bad for a short time and then recover.
3. Look for three signs: a flat line, an amplitude much larger than the neighboring channels, or repeated jumps in a single channel.
4. In D1 you will see large peaks in EOG1, EOG2, and the frontal channels. These are eye blinks, a normal artifact that ICA removes later. A channel is not bad because it records blinks.

Second: look at the spectrum

1. Open Plot › then Channel spectra and maps.
2. Change Plotting frequency range [lo_Hz hi_Hz]: to 1 64, then press OK.
3. A bad channel shows up as a line far from the group: much higher, or much lower if it is flat.
4. Click any line in the plot, and its channel number appears in the command window.
5. In D1 the lines stay close together, and no line clearly stands apart.
6. To look closely at one suspect channel, open Plot › then Channel properties and type its number. The window shows the channel’s location on the head, its activity over time, and its spectrum together.

Third: check with the clean_rawdata channel criteria only

1. Open Tools › then Reject data using Clean Rawdata and ASR.
2. Leave Remove channel drift (data not already high-pass filtered) unchecked, because the data are already filtered.
3. Keep Process/remove channels checked.
4. Check Ignore these channels (ECG, EMG, …) and type EOG1 EOG2 in the field next to it.
5. Leave the three criteria at their defaults: Remove channel if it is flat for more than (seconds) at 5, Max acceptable high-frequency noise std dev at 4, and Min acceptable correlation with nearby chans [0-1] at 0.8.
6. Uncheck Perform Artifact Subspace Reconstruction bad burst correction/rejection, and uncheck Additional removal of bad data periods. These two remove time segments, which is the topic of another lesson.
7. Press OK. The command window prints Scanning for bad channels... and then its progress over 5-second blocks.
8. A window comparing the data before and after appears. A removed channel shows in red. In D1 no channel is red.
9. Name the resulting dataset D1 badch and leave Overwrite it in memory (set=yes; unset=create a new dataset) unchecked.

Fourth: record the names and remove what you found by eye

1. Write down the names of all the bad channels you found, from all three methods. You will need them for interpolation and for the Methods section of your paper.
2. clean_rawdata removes what it detects by itself. A channel you found only by eye, remove it with Edit › then Select data: type its name in Channel(s), check the box under on->remove these on the same row, then press OK.
3. In D1 the list is empty, so you remove nothing. Save D1 badch with File › then Save current dataset as. The interpolation lesson starts from it.

How do I write it as code?

% Detection with clean_rawdata channel criteria only:
% no drift removal, no ASR, no window rejection, EOG channels ignored
EEG_test = pop_clean_rawdata(EEG, 'FlatlineCriterion', 5, 'ChannelCriterion', 0.8, ...
    'LineNoiseCriterion', 4, 'Highpass', 'off', 'BurstCriterion', 'off', ...
    'WindowCriterion', 'off', 'BurstRejection', 'off', 'Distance', 'Euclidian', ...
    'channels_ignore', {'EOG1', 'EOG2'});

% Names of the channels it removed
badChans = setdiff({EEG.chanlocs.labels}, {EEG_test.chanlocs.labels}, 'stable');

% Add channels you found by eye, e.g. badChans = [badChans {'T7'}];
fprintf('Bad channels: %s\n', strjoin(badChans, ', '));

% Remove them by name from the main dataset
if ~isempty(badChans)
    EEG = pop_select(EEG, 'rmchannel', badChans);
end
EEG.setname = 'D1 badch';
EEG = pop_saveset(EEG, 'filename', 'D1_badch.set', 'filepath', pwd);

On D1 the script prints Bad channels: with an empty list, and the channel count stays 32. Copy badChans into the Starting point script of the following lessons, which start with the same variable.

Which value should I choose?

How do I check it worked?

1. Channel count: in the main window, Channels per frame equals the original number minus the removed channels. In D1 it stays 32.
2. Your list is written down: you have the names of the removed channels, or “none” as in D1.
3. Spectrum after removal: open Plot › then Channel spectra and maps with the range 1 64. No line clearly stands apart from the rest.
4. The eye channels are still there: EOG1 and EOG2 are still in the channel list.

What mistakes do people make?

• Not excluding the eye channels. Without excluding EOG1 and EOG2, clean_rawdata may remove EOG1 in D1, because its signal is unlike the neighboring channels. It is not bad: it is a non-brain channel by nature. The clean_rawdata documentation itself says that channels carrying non-brain signals are usually removed this way. That is why the official clean_rawdata page may show red channels on the same data.
• Treating a blink channel as bad. Large peaks in the frontal channels are eye blinks. ICA removes them later, and the channel is not removed because of them.
• Looking only at the first few seconds. Some channels go bad in the middle of a recording. Scroll through all of it.
• Removing a channel without recording its name. You need the names for interpolation, for the Methods section, and to process the other participants the same way.
• Removing many channels without asking why. If many channels look bad, first check the electrode locations and the whole recording. clean_rawdata itself stops with an error if it would remove more than 75% of the channels, and says the cause is usually wrong electrode locations.

How do I write it in my paper?

A template: edit the data, settings and versions to match what you ran.

Bad channels were identified by visual inspection of the continuous data and channel spectra and with clean_rawdata (flatline criterion 5 s, channel correlation criterion 0.8, line noise criterion 4 SD; EOG channels excluded; clean_rawdata plugin 2.11, EEGLAB 2026.1) and removed ([0] channels).

Going further

• EEGLAB Tutorial: https://eeglab.org/tutorials/06_RejectArtifacts/Channel_rejection.html (6. Reject artifacts > a. Remove bad channels by visual inspection)
• EEGLAB Tutorial: https://eeglab.org/tutorials/06_RejectArtifacts/cleanrawdata.html (6. Reject artifacts > Automated rejection using clean_rawdata)
• Luck, Applied Event-Related Potential Data Analysis (1st edition): https://socialsci.libretexts.org/Bookshelves/Psychology/Biological_Psychology/Applied_Event-Related_Potential_Data_Analysis_(Luck) (Chapter 7: Inspecting the EEG and Interpolating Bad Channels)
• EEGLAB preprocessing #6: Removing bad channels: https://www.youtube.com/watch?v=5BTT-L2Ab04 (Full video)