Detection of Unfocused EEG Epochs by the Application of Machine Learning Algorithm

Area of Science:

• Neuroscience
• Cognitive Science
• Biomarker Development

Background:

• Electroencephalography (EEG) records human brain activity, with event-related potentials (ERPs) serving as biomarkers for cognitive processes.
• Real-world ERP research is limited by challenges in controlling experimental variables and subject attentiveness.
• Developing methods to ensure ERP reliability outside strict laboratory settings is crucial.

Purpose of the Study:

• To evaluate unsupervised machine learning algorithms (MLAs) for identifying artifact-affected ERP epochs.
• To compare MLA performance against human inspection and EEGLab for artifact detection.
• To enable the use of ERPs as active biomarkers in less controlled environments.

Main Methods:

• Collected EEG data using an auditory oddball paradigm under varied experimental conditions.
• Analyzed ERP epochs to detect unfocused data influenced by artifacts and external distortions.
• Applied four unsupervised MLAs to identify unfocused epochs and compared their accuracy to human analysis and EEGLab.

Main Results:

• All four unsupervised MLAs achieved 95-100% accuracy in identifying unfocused ERP epochs.
• MLAs demonstrated superior ability to detect subtle deviations in ERP patterns compared to human observers.
• Unsupervised MLAs outperformed both human inspection and EEGLab in detecting artifactual epochs.

Conclusions:

• Unsupervised MLAs are highly effective for identifying unfocused ERP epochs, surpassing traditional methods.
• This approach enhances the potential of ERPs as reliable biomarkers in real-world applications.
• Machine learning offers a robust solution for improving the quality and consistency of ERP data analysis.