Real-time estimation of EEG-based engagement in different tasks

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCI) are increasingly used for passive monitoring of cognitive states like engagement.
• Engagement monitoring is crucial for adaptive learning systems, immersive entertainment, and personalized rehabilitation.
• Existing methods often lack real-time cross-task applicability for engagement estimation.

Purpose of the Study:

• To develop and validate a novel approach for real-time engagement estimation using electroencephalography (EEG).
• To assess the cross-task generalizability of the developed EEG-based engagement estimation model.
• To investigate the relationship between perceived and objectively measured engagement.

Main Methods:

• Twenty-three healthy subjects underwent EEG recording during a modified d2 test to elicit engagement.
• Within-subject classification models were trained using filter-bank common spatial patterns and linear discriminant analysis.
• Model performance was evaluated across diverse tasks: Tetris at varying speeds and watching different videos.

Main Results:

• The EEG engagement estimation model achieved an average classification accuracy of 90% on an independent dataset.
• Higher engagement was detected during advertisement videos versus landscape videos, and during faster Tetris gameplay.
• A significant linear correlation (r=0.44, p<0.001) was found between perceived and estimated engagement.

Conclusions:

• A task-specific EEG engagement estimation model with demonstrated cross-task capabilities was successfully developed.
• The findings suggest a viable framework for real-world applications requiring real-time cognitive state monitoring.
• Theta and alpha band power changes in EEG correlate with engagement levels, providing neural correlates.