Automatic Classification of Artifactual ICA-Components for Artifact Removal in EEG Signals

By Irene Winkler, Stefan Haufe, Michael W. Tangermann. Behavioral and Brain Functions, 2011

752 citations recorded by the source corpus.

BackgroundArtifacts contained in EEG recordings hamper both, the visual interpretation by experts as well as the algorithmic processing and analysis (e.g. for Brain-Computer Interfaces (BCI) or for Mental State Monitoring). While hand-optimized selection of source components derived from Independent Component Analysis (ICA) to clean EEG data is widespread, the field could greatly profit from automated solutions based on Machine Learning methods. Existing ICA-based removal strategies depend on explicit recordings of an individual's artifacts or have not been shown to reliably identify muscle artifacts.MethodsWe propose an automatic method for the classification of general artifactual source components. They are estimated by TDSEP, an ICA method that takes temporal correlations into account. The linear classifier is based on an optimized feature subset determined by a Linear Programming Machine (LPM). The subset is composed of features from the frequency-, the spatial- and temporal domain. A subject independent classifier was trained on 640 TDSEP components (reaction time (RT) study, n = 12) that were hand labeled by experts as artifactual or brain sources and tested on 1080 new components of RT data of the same study. Generalization was tested on new data from two studies (auditory Event Related Potential (ERP) paradigm, n = 18; motor imagery BCI paradigm, n = 80) that used data with different channel setups and from new subjects.ResultsBased on six features only, the optimized linear classifier performed on level with the inter-expert disagreement (< 10% Mean Squared Error (MSE)) on the RT data. On data of the auditory ERP study, the same pre-calculated classifier generalized well and achieved 15% MSE. On data of the motor imagery paradigm, we demonstrate that the discriminant information used for BCI is preserved when removing up to 60% of the most artifactual source components.ConclusionsWe propose a universal and efficient classifier of ICA components for the subject independent removal of artifacts from EEG data. Based on linear methods, it is applicable for different electrode placements and supports the introspection of results. Trained on expert ratings of large data sets, it is not restricted to the detection of eye- and muscle artifacts. Its performance and generalization ability is demonstrated on data of different EEG studies.

Subfields: Cognitive Neuroscience

View Paper: https://www.semanticscholar.org/paper/14ee30cabe44a516b888db4cf4736398f85ed464

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

Citations in the corpus

Robust artifactual independent component classification for BCI practitioners (2014, methodology): “In this paper, we take a step forward by analyzing the generalization ability of a state-of-the-art supervised IC classiﬁcation algorithm which we have recently proposed [34].”
The Harvard Automated Processing Pipeline for Electroencephalography (HAPPE): Standardized Processing Software for Developmental and High-Artifact Data (2018, methodology): “Automated component rejection is achieved through the Multiple Artifact Rejection Algorithm (MARA), a machine-learning algorithm that evaluates the ICA-derived components (Winkler et al., 2011, 2014).”
On the influence of high-pass filtering on ICA-based artifact reduction in EEG-ERP (2015, methodology): “For varying cutoff frequencies, we train ICA on highpass filtered data, automatically classify the resulting ICs with MARA [7], [8], and apply the obtained filter weights on the original non high-pass filtered data (see Fig.”
Visual and Auditory Brain–Computer Interfaces (2014, background): “ICA enhances the SNR of EEG signals by separating task-related EEG components from the task-irrelevant EEG components and the artifactual components [129, 130].”
ICLabel: An automated electroencephalographic independent component classifier, dataset, and website (2019, neutral): “The previous methods listed above used one or two experiment types with the exception of SASICA and MARA which used more.”
Reliable but Wrong? Automated Detection of Heart Rate Artifacts as a Source of Bias in Pediatric EEG Data (2026, unknown): “…common decision algorithm, Multiple Artifact Rejection Algorithm (MARA), evaluates ICA components on 6 features (mean local skewness, log alpha power, lambda, ﬁt error, range within pattern, and current density norm) and combines these metrics with manual training data to classify components [9].”
A human electrophysiological signature of Fragile X pathophysiology is shared in V1 of Fmr1-/y mice (2026, unknown): “…high-pass and 100 Hz low-pass ﬁ ltered, down-sampled to 250Hz, and then run through the HAPPE module for 60 Hz line noise removal, bad channel rejection and artifact removal using combined wavelet-enhanced independent component analysis (ICA) and Multiple Artifact Rejection Algorithm (MARA) 111 .”
Alterations in aperiodic and periodic EEG activity in young children with Down syndrome (2024, unknown): “…to 250 Hz and then run through the HAPPE module which included 60 Hz line noise removal, bad channel rejection, and artifact removal using combined wavelet-enhanced independent component analysis (ICA) and Multiple Artifact Rejection Algorithm (MARA (Winkler et al., 2011; Winkler et al., 2014)).”
Alterations in aperiodic and periodic EEG activity in young children with Down syndrome (2024, unknown): “…to 250 Hz and then run through the HAPPE module which included 60 Hz line noise removal, bad channel rejection, and artifact removal using combined wavelet-enhanced independent component analysis (ICA) and Multiple Artifact Rejection Algorithm (MARA (Winkler et al., 2011; Winkler et al., 2014)).”
Evaluating early EEG correlates of restricted and repetitive behaviors for toddlers with or without autism (2024, methodology): “Artifacts were rejected (e.g., eye blinks, movement, and muscle activity) through wavelet-enhanced ICA with automated component rejection via EEGLAB [54] and the Multiple Artifact Rejection Algorithm [55, 56].”
The developmental trajectory of functional excitation‐inhibition balance relates to language abilities in autistic and allistic children (2023, methodology): “Artifact components were automatically rejected using the multiple artifact rejection algorithm (MARA; M autistic = 43.88, SD = 9.41; M allistic = 46.32, SD = 10.09, t = (cid:1) 1.73, p = 0.086; Winkler et al., 2011).”