Adaptable and Robust EEG Bad Channel Detection Using Local Outlier Factor (LOF).

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) data quality is often compromised by artifacts, necessitating effective bad channel detection.
• Existing methods may struggle with diverse EEG datasets due to variations in data quality, artifact types, and experimental designs.

Purpose of the Study:

• To introduce a robust and adaptable method for detecting bad channels in EEG data.
• To address the limitations of current methods in handling diverse EEG acquisition scenarios.

Main Methods:

• Proposed a novel bad channel detection technique utilizing the Local Outlier Factor (LOF) algorithm.
• Validated the LOF method on EEG data from newborns, infants, and adults across event-related and frequency-tagging paradigms.
• Benchmarked LOF performance against state-of-the-art (SoA) bad channel detection techniques.

Main Results:

• The LOF algorithm demonstrated adaptability to various EEG data types after hyperparameter calibration (LOF threshold).
• LOF significantly outperformed existing SoA methods in bad channel detection.
• Achieved substantial improvements in F1 Score: approximately 40% for newborns/infants and 87.5% for adults.

Conclusions:

• The LOF-based approach offers a versatile and superior solution for EEG bad channel detection.
• This method enhances the reliability of EEG data preprocessing across different subject groups and experimental setups.
• The findings suggest LOF as a valuable tool for improving the accuracy of EEG analysis in clinical and research settings.