Identify Noisy Channels

The noisy_channels step identifies and marks EEG channels as noisy based on whether their per-epoch signal standard deviation is consistently an outlier relative to other channels.

This step operates in two stages:

  1. For each epoch, channels whose standard deviation is identified as an outlier are flagged.

  2. A channel is marked as noisy if it is flagged in more than flag_crit proportion of epochs.

Examples

noisy_channels:
    flag_crit: 0.2
    outlier_method: quantile
    outliers_kwargs:
        k: 6
        lower: 0.25
        upper: 0.75

Parameters

  • flag_critfloat, default=0.2

    Proportion of epochs in which a channel must be flagged as an outlier to be marked as noisy.

  • outlier_methodstr, default=’quantile’

    Method used to identify per-epoch outlier channels. Options are 'quantile', 'trimmed', or 'fixed'.

    • 'quantile'

      Identifies outliers using asymmetric, quantile-based thresholds. For a given epoch, channels are flagged if their standard deviation lies outside

      μ - k * (μ - q_low) or μ + k * (q_high - μ)

      where μ is the mean standard deviation across channels, q_low and q_high are the lower or upper quantile specified by outliers_kwargs['lower'] and outliers_kwargs['upper'], and k is a scaling factor.

      This method is robust to skewed distributions and is suitable when channel variability is not symmetrically distributed.

    • 'trimmed'

      Identifies outliers using robust statistics by computing the trimmed mean and trimmed standard deviation across channels. For a given epoch, channels with standard deviation values outside

      trimmed_mean ± k * trimmed_std

      are flagged as noisy.

      This method reduces the influence of extreme artifacts while preserving sensitivity to moderately noisy channels.

    • 'fixed'

      Identifies outliers using fixed, absolute thresholds. For a given epoch, channels with standard deviation values outside the thresholds (in volts) specified by outliers_kwargs['lower'] and outliers_kwargs['upper'] are flagged.

  • outliers_kwargsdict

    Additional keyword arguments specific to the chosen outlier detection method.

    • For 'quantile':
      • kfloat, default=6

        Scaling factor for the quantile-based thresholds.

      • lowerfloat, default=0.25

        Lower quantile.

      • upperfloat, default=0.75

        Upper quantile.

    • For 'trimmed':
      • kfloat, default=3

        Scaling factor for trimmed standard deviation thresholds.

Notes

The trimmed and fixed methods are not robustly tested in the current codebase. Please use with caution and validate results carefully.