A Gated Artifact Management Pipeline for Low-Density Eyewear EEG

Area of Science:

• Biomedical Engineering
• Neuroscience
• Signal Processing

Background:

• Wearable eyewear platforms offer a promising avenue for unobtrusive electroencephalography (EEG) monitoring.
• Low-density, dry-electrode EEG systems present unique challenges for artifact management due to signal quality and potential for motion artifacts.
• Effective artifact detection, classification, and correction are crucial for reliable EEG data analysis in real-world, ambulatory settings.

Purpose of the Study:

• To develop and validate an artifact-management method specifically designed for low-density, dry-electrode EEG data acquired from wearable eyewear.
• To enable efficient artifact detection, classification, and correction while minimizing the computational load on resource-constrained wearable devices.
• To ensure that clean neural activity is processed minimally, preserving signal integrity.

Main Methods:

• A gated, two-stage processing pipeline was implemented, featuring a lightweight gate classifier to identify artifact-contaminated EEG windows.
• The pipeline integrates a time-domain feature-based detector, a convolutional-recurrent network for artifact classification (eye blinks, horizontal eye movements, facial movements), and a UNet-based denoising autoencoder.
• A leave-one-subject-out cross-validation scheme was employed for evaluating the system on a dedicated eyewear EEG dataset.

Main Results:

• The gate classifier achieved a balanced accuracy of 0.89, while the artifact classifier reached a balanced accuracy of 0.84 (median macro-F1 of 0.83).
• The denoising autoencoder demonstrated a median normalized RMSE below 0.20 and a spectral cosine similarity above 0.84.
• End-to-end evaluation showed a median normalized RMSE of 0.19 and spectral cosine similarity of 0.89, with comparable performance on an independent benchmark dataset.

Conclusions:

• The proposed artifact-management approach is effective for wearable eyewear EEG.
• The system's design is compatible with the constraints of wearable, resource-limited implementations.
• This method facilitates reliable EEG data acquisition and analysis in ambulatory settings using eyewear platforms.