A Cross-Subject Band-Power Complexity Metric for Detecting Mental Fatigue Through EEG

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) shows promise for fatigue detection due to its direct reflection of neural states.
• Current EEG fatigue detection methods face challenges with subject-specific calibration and unstable labeling.
• Classical EEG features are susceptible to brain rhythm variations, leading to performance degradation across sessions and subjects.

Purpose of the Study:

• To develop a robust, cross-subject EEG metric for fatigue detection.
• To overcome the limitations of subject-specific calibration and domain shifts in EEG-based fatigue monitoring.
• To introduce a fatigue indicator that does not require additional model training.

Main Methods:

• Proposed a novel metric, Short-Term Second-Order Differential Entropy (ST-SODE), inspired by biological fatigue rebound.
• ST-SODE is designed to suppress background brain rhythm interference, improving robustness to cross-domain drift.
• The metric yields a one-dimensional output for fatigue state indication.

Main Results:

• ST-SODE achieved a correlation coefficient of 0.56 on the SEED-VIG driving fatigue dataset, outperforming differential entropy (DE) (0.4).
• On a private Vigilance N-Back task dataset, ST-SODE reached a binary classification accuracy of 93.75%.
• The proposed metric demonstrated superior performance compared to other EEG-based fatigue detection methods.

Conclusions:

• ST-SODE provides a reliable and robust solution for fatigue detection using EEG.
• The metric's cross-subject applicability and reduced need for calibration make it suitable for real-world deployment.
• Potential applications include enhancing safety in driving, manufacturing, and healthcare settings by mitigating fatigue-related incidents.