Artifact-reference multivariate backward regression (ARMBR): a novel method for EEG blink artifact removal with...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Ocular artifacts, particularly eye blinks, are a significant source of noise in electroencephalography (EEG) recordings.
• Accurate artifact removal is crucial for reliable EEG data analysis and interpretation.
• Existing methods for blink artifact removal vary in complexity, data requirements, and effectiveness.

Purpose of the Study:

• To introduce a novel, lightweight, and data-efficient method for removing ocular artifacts from EEG.
• To evaluate the performance of the proposed method against established artifact removal techniques.
• To assess the method's potential for online EEG processing and clinical translation.

Main Methods:

• A robust, cross-validated thresholding procedure automatically detects eye blinks.
• A simplified time-locked reference signal is regressed against multi-channel EEG to estimate scalp projection.
• Performance is compared against MNE's signal subspace projection and forward regression, EEGLab's ICA+ICLabel, and an ASR Python implementation.

Main Results:

• The proposed method demonstrated superior ground truth reconstruction on semi-synthetic data compared to MNE methods.
• Performance was comparable or better than ASR and ICA+ICLabel on semi-synthetic data.
• On real EEG data, the method showed targeted artifact removal with minimal impact on uncontaminated data and higher frequencies, similar to ICA+ICLabel.

Conclusions:

• The novel algorithm effectively removes ocular artifacts while preserving important neural signals, such as event-related potentials.
• The method offers comparable or superior performance to existing techniques, with the advantage of requiring minimal training data.
• Its lightweight nature and potential for online application position it as a valuable tool for advancing neural engineering technologies.