Skip to main content

MlIcaLabel

Namespace: DECODINGLIB  ·  Library: Decoding Library

Python equivalent

mne_icalabel in MNE-Python.

#include <decoding/decoding_ica_label.h>

class DECODINGLIB::MlIcaLabel

Static utility that labels ICA components against EOG/ECG references and a muscle spectral heuristic.

classify walks the rows of the source matrix and applies, in order, the EOG correlation test, the ECG correlation test and the high-frequency muscle test; the first one whose score exceeds its threshold wins, otherwise the component is labelled Brain. The thresholds default to the canonical 0.3 used by the MNE-Python helpers and can be tightened or loosened per call. findArtifactComponents is the convenience that returns just the indices of the rejected components in the order expected by the ICA reconstruction code, making it a one-liner to wire automatic cleaning into a real-time pipeline.

The two scoring primitives — maxAbsCorrelation and muscleScore — are exposed publicly so an interactive viewer can display the per-component evidence behind each decision and let the user override the label without having to recompute the ICA itself. The class has no state, deleted constructor, and no virtual methods; it is purely a namespaced collection of pure functions.

QList<IcaLabelResult> labels = MlIcaLabel::classify(
    icaSources, eogData, ecgData, sFreq);
QVector<int> artIdx = MlIcaLabel::findArtifactComponents(labels);

Static Methods​

classify(matSources, matEog, matEcg, dSFreq, dEogThresh, dEcgThresh)​

Classify each ICA component using reference signals and spectral heuristics.

Parameters:

  • matSources : const Eigen::MatrixXd & ICA sources (n_components x n_samples).

  • matEog : const Eigen::MatrixXd & EOG reference channels (n_eog x n_samples). Can be empty.

  • matEcg : const Eigen::MatrixXd & ECG reference channels (n_ecg x n_samples). Can be empty.

  • dSFreq : double Sampling frequency in Hz.

  • dEogThresh : double Correlation threshold for EOG classification (default 0.3).

  • dEcgThresh : double Correlation threshold for ECG classification (default 0.3).

Returns:

  • QList< IcaLabelResult > — Label results for each component.

findArtifactComponents(labels)​

Extract indices of artifact components (EOG, ECG, muscle).

Parameters:

  • labels : const QList< IcaLabelResult > & Label results from classify().

Returns:

  • QVector< int > — 0-based indices of artifact components.

maxAbsCorrelation(source, matRef)​

Compute the maximum absolute Pearson correlation between a source and reference channels.

Parameters:

  • source : const Eigen::VectorXd & Source time course (n_samples).

  • matRef : const Eigen::MatrixXd & Reference channels (n_ref x n_samples).

Returns:

  • double — Maximum |r| across reference channels.

muscleScore(source, dSFreq)​

Compute a simple spectral power ratio heuristic for muscle artifact detection.

Muscle artifacts typically have high power above 30 Hz.

Parameters:

  • source : const Eigen::VectorXd & Source time course (n_samples).

  • dSFreq : double Sampling frequency in Hz.

Returns:

  • double — Ratio of high-frequency to total power (0 = no muscle, 1 = all muscle).

Authors of this file​