Rejecting Bad Channels/Epochs

Initialize NeuroAnalyzer 
using NeuroAnalyzer
eeg = load("files/eeg.hdf")
e10 = epoch(eeg; ep_len = 10)

Removing bad channels and bad epochs is a critical step in EEG preprocessing. Bad channels/epochs can introduce artifacts, noise, or biases, affecting the accuracy of analyses (e.g., ERP extraction, power spectral density, source localization).

Why Remove Bad Channels/Epochs?

Reason Description
Improve Signal Quality Removing noisy channels/epochs enhances the signal-to-noise ratio (SNR).
Avoid Artifacts Bad channels/epochs can introduce spikes, drifts, or muscle artifacts.
Ensure Reliable Analysis Analyses like ERP, connectivity, or source localization require clean data.
Automate Preprocessing Automated detection reduces manual effort and subjectivity.

NeuroAnalyzer provides automated methods to detect bad channels and epochs based on:

  • :flat: channels with a large proportion of flat windows
  • :rmse: channels whose RMSE vs the median reference is outside the 95 % CI
  • :rmsd: same as :rmse using RMSD
  • :euclid: same using Euclidean distance
  • :var: channels with IQR-outlier variance
  • :p2p: channels with excessive peak-to-peak variation
  • :tkeo: channels where z-scored TKEO diverges from z-scored signal
  • :kurt: channels with z-scored kurtosis exceeding z
  • :z: channels with a large proportion of samples above the z-score threshold
  • :ransac: channels poorly correlated with their nearest spatial neighbor
  • :amp: channels exceeding ± amplitude rejection threshold

Detecting bad channels:

bad_chs = channel_reject(e10;
                         ch = "all",
                         method = [:flat, :p2p, :var])

Output: bad_chs vector of bad channels.

Header of each NEURO object contains a field for marking channels as bad (obj.header.recording[:bad_channel]). To modify this vector in-place, use the mutator variant:

channel_reject!(e10;
                ch = "all",
                method = [:flat, :p2p, :var])

Tip: Bad channels are drawn in less intense color. You may also manually mark channels as bad when using plot() by right-clicking on the channel label.

Tip: Channels marked as bad will be be automatically excluded from analysis if exclude_bads preference is set to true.

Similarly, epoch can be rejected using the same methods:

bad_eps = epoch_reject(e10;
                       ch = "all",
                       method = [:flat, :p2p, :var])