Comparison of a Pragmatic and Regression Approach for Wearable EEG Signal Quality Assessment

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Wearable electroencephalogram (EEG) systems enable portable, real-time brain activity monitoring in diverse environments.
• Accurate identification of EEG data artefacts is crucial for reliable signal interpretation, especially in uncontrolled settings.

Purpose of the Study:

• To develop and validate novel data quality indicator approaches for wearable EEG.
• To compare the performance of proposed methods against existing signal quality estimation techniques.

Main Methods:

• Two data quality indicator approaches were developed: a pragmatic method using statistical features and data-driven thresholding, and a regression-based method predicting data quality.
• Performance was validated using EEG data from uncontrolled laboratory and free-living conditions, compared against the FORCe signal quality estimation method.

Main Results:

• The proposed pragmatic and regression-based approaches achieved an average accuracy exceeding [Formula: see text] in detecting artefactual EEG data, outperforming the FORCe method ([Formula: see text]).
• A significant increase in specificity over the state-of-the-art was observed.
• Models trained on free-living data demonstrated better generalization across different recording conditions.

Conclusions:

• The pragmatic and regression-based approaches provide robust and accurate artefact detection for wearable EEG systems.
• These methods offer improved specificity and generalization capabilities, facilitating real-time implementation on wearable devices.