Skip to contents

Detect Artifact ICA Components

Source: R/eeg-ica.R
eegICAdetect.Rd

Automatically identifies artifact components using one of four methods: correlation with frontal channels, kurtosis, spatial weight pattern, or the ICLabel-style seven-class classifier (eegICLabel()). The "iclabel" method delegates to eegICLabel() and collapses its result to the two-class contract (any non-brain argmax is an "artifact"), with the score being one minus the brain probability.

Usage

eegICAdetect(
  x,
  method = c("correlation", "kurtosis", "spatial", "iclabel"),
  threshold = 0.3,
  ica_assay = "ica_components"
)

Arguments

x

A PhysioExperiment object with ICA results (from eegICA).

method

Detection method: "correlation" (frontal channel correlation), "kurtosis" (excess kurtosis), "spatial" (spatial weight pattern), or "iclabel" (seven-class classifier).

threshold

Threshold for artifact detection. For "correlation", absolute correlation > threshold marks artifact (default: 0.3).

ica_assay

Assay name containing ICA activations (default: "ica").

Value

A data.frame with columns: component (integer index), type ("artifact" or "neural"), method (detection method used), and score (numeric detection score).

References

Hyvarinen, A., & Oja, E. (2000). Independent component analysis: algorithms and applications. Neural Networks, 13(4-5), 411-430.

Bell, A. J., & Sejnowski, T. J. (1995). An information-maximization approach to blind separation and blind deconvolution. Neural Computation, 7(6), 1129-1159.

See also

eegICA(), eegICAremove(), eegICAmix()

Examples

if (FALSE) { # \dontrun{
pe <- make_eeg(n_time = 5000, sr = 500)
pe <- eegICA(pe, n_components = 4, method = "fastica")
artifacts <- eegICAdetect(pe, method = "kurtosis")
} # }