Selection of valid and reliable EEG features for predicting auditory and visual alertness levels

Area of Science:

• Neuroscience
• Cognitive Science
• Signal Processing

Background:

• Behavioral alertness is crucial for performance in vigilance tasks.
• Electroencephalogram (EEG) signals offer insights into brain states related to alertness.
• Developing objective measures for alertness prediction is essential.

Purpose of the Study:

• To establish an efficient selection procedure for electroencephalogram (EEG) features to predict behavioral alertness.
• To identify optimal EEG features for predicting performance in auditory and visual vigilance tasks.
• To enhance the robustness of EEG analysis against movement artifacts.

Main Methods:

• A three-rule selection procedure (high efficiency, low individual variability, low redundancy) was applied to 24 derived EEG features.
• Behavioral alertness was quantified using correct performance rates in auditory and visual vigilance tasks.
• An averaging subwindow procedure within a moving time window was employed for EEG analysis.

Main Results:

• For auditory tasks, a combination of alpha and theta relative spectral amplitudes and mean spectrum frequency (MF) best predicted alertness.
• For visual tasks, the mean frequency of the beta band (Fbeta) was the sole predictive EEG feature.
• The averaging subwindow procedure significantly improved EEG feature predictive power and reduced movement artifact interference.

Conclusions:

• Specific EEG features, including spectral amplitudes and frequencies, can effectively predict behavioral alertness in different sensory modalities.
• The developed EEG feature selection procedure offers an efficient and reliable method for alertness monitoring.
• Advanced EEG processing techniques, like averaging subwindows, enhance data quality and predictive accuracy for alertness assessment.