Detect and Correct EDA Artifacts

Identifies artifacts in electrodermal activity (EDA) signals using one or more detection methods (threshold, gradient, flatline, and epoch-based wavelet / machine-learning methods) and optionally corrects them via interpolation or NA replacement. An accelerometer channel can additionally gate out high-motion epochs (Taylor et al., 2015).

Usage

edaArtifact(
  x,
  methods = c("threshold", "gradient", "flatline"),
  threshold_range = c(0.001, 60),
  gradient_max = NULL,
  flatline_sec = 5,
  epoch_sec = 5,
  wavelet_k = 5,
  ml_threshold = 0.5,
  acc_channel = NULL,
  acc_threshold = NULL,
  correct = "interpolate",
  assay_name = NULL,
  output_assay = "cleaned"
)

Arguments

x

A PhysioExperiment object containing EDA data.

methods

Character vector of detection methods to apply. Any combination of "threshold", "gradient", "flatline", "wavelet" (per-5 s-epoch stationary-wavelet fine-scale energy outliers), and "ml" (a shipped ridge-logistic classifier on wavelet/derivative epoch features). Default is the first three.

threshold_range

Numeric vector of length 2 giving the acceptable range of EDA values in microsiemens (default: c(0.001, 60)). Values outside this range are flagged as artifacts.

gradient_max

Maximum allowable absolute gradient in microsiemens per sample. If NULL (default), computed as 10 / samplingRate(x) (equivalent to 10 uS/sec).

flatline_sec

Minimum duration in seconds of a constant-value segment to be flagged as a flatline artifact (default: 5).

epoch_sec

Epoch length in seconds for the "wavelet" / "ml" methods and accelerometer gating (default: 5).

wavelet_k

Robust-outlier multiplier for the "wavelet" method; an epoch is flagged when its peak level-1 detail coefficient exceeds median + wavelet_k * MAD across epochs (default: 5).

ml_threshold

Probability threshold for the "ml" method's binary label (default: 0.5).

acc_channel

Optional channel index or label (or several) identifying accelerometer channel(s). When supplied, per-epoch motion energy (variance of the accelerometer magnitude) gates the EDA channels: epochs with energy above acc_threshold are flagged. The accelerometer channels are excluded from EDA artifact detection.

acc_threshold

Motion-energy threshold for accelerometer gating. If NULL (default), computed robustly as median + 3 * MAD of the per-epoch motion energies.

correct

Correction strategy: "interpolate" (linear interpolation across artifact regions), "na" (replace with NA), or "none" (detection only, no correction). Default is "interpolate".

assay_name

Name of the input assay. If NULL, uses defaultAssay(x).

output_assay

Name for the corrected output assay (default: "cleaned"). Only used when correct != "none".

Value

A modified PhysioExperiment with artifact information stored in metadata(x)$eda_artifacts, a list containing:

mask

Logical matrix (time x channels) where TRUE indicates an artifact sample.

summary

A data.frame with columns channel, method, n_artifacts, and pct giving artifact counts per channel per detection method.

epochs

(only when "wavelet"/"ml" or accelerometer gating is used) A data.frame, one row per EDA channel per epoch, with the epoch bounds and, as applicable, wavelet_flag, ml_prob, ml_flag, motion_energy, and acc_flag.

If correct != "none", the corrected signal is stored in the output_assay.

References

Taylor, S., et al. (2015). "Automatic identification of artifacts in electrodermal activity data." IEEE EMBC, 1934-1937. doi:10.1109/EMBC.2015.7318762

Kleckner, I.R., et al. (2018). "Simple, transparent, and flexible automated quality assessment procedures for ambulatory electrodermal activity data." IEEE Transactions on Biomedical Engineering, 65(7), 1460-1467. doi:10.1109/TBME.2017.2758643

Boucsein, W. (2012). Electrodermal Activity. 2nd ed. Springer. doi:10.1007/978-1-4614-1126-0

See also

edaQuality for signal quality assessment, edaFilter for frequency-domain filtering, edaDecompose for tonic/phasic decomposition