Automatic bad channel detection in intracranial electroencephalographic recordings using ensemble machine learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Intracranial electroencephalographic (iEEG) recordings are crucial for understanding brain activity but often contain "bad channels" with non-neuronal signals.
• Accurate identification of these bad channels is essential for reliable data analysis and interpretation.

Purpose of the Study:

• To develop and validate a novel machine learning method for the automatic detection of bad channels in iEEG data.
• To enhance the quality control process for iEEG signal review.

Main Methods:

• A machine learning approach was developed using seven signal features, including variance, spatial-temporal correlation, and nonlinear properties.
• An ensemble bagging classifier was employed to handle the imbalanced class distribution (fewer bad channels than good channels).
• The method was applied to stereo-electroencephalographic (SEEG) signals from 206 patients across five clinical centers.

Main Results:

• The classification accuracy reached 99.77% with 110 subjects, demonstrating excellent performance.
• The accuracy plateaued with an increasing number of subjects, indicating robust model training.
• The multicentric nature of the data did not negatively impact the classification performance.

Conclusions:

• The developed method provides a reliable and automated solution for detecting bad channels in iEEG data.
• This approach can be scaled for automatic quality control in large iEEG datasets.
• The method represents a significant advancement in iEEG data selection and review.