Machine learning of brain-specific biomarkers from EEG.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalography (EEG) is a clinical tool for studying brain function with untapped potential for biomarker discovery.
• Machine learning (ML) can leverage complex EEG signals for innovation, but often overlooks physiological artifacts.
• Ignoring artifacts may hinder the development of central nervous system (CNS)-specific biomarkers from EEG data.

Purpose of the Study:

• To present a framework for conceptualizing ML from CNS versus peripheral signals in EEG.
• To investigate the impact of peripheral signals and artifact removal on ML models for age and sex prediction.
• To determine the contribution of brain versus peripheral signals to predictive performance.

Main Methods:

• Developed a signal representation using Morlet wavelets for EEG analysis.
• Utilized traditional brain activity features and covariance matrices for ML models.
• Analyzed over 2600 EEG recordings from public databases (TUAB, TDBRAIN).

Main Results:

• Basic artifact rejection improved ML model performance.
• Independent Component Analysis (ICA)-based removal of peripheral signals decreased performance.
• Peripheral signals contributed to age and sex prediction but less than brain signals.

Conclusions:

• Both brain and peripheral signals within EEG can predict personal characteristics like age and sex.
• Careful signal separation is essential when aiming for CNS-specific biomarker development using ML.
• Findings highlight the importance of distinguishing signal sources for accurate biomarker discovery.