Generalized Deep Learning EEG Models for Cross-Participant and Cross-Task Detection of the Vigilance Decrement in...

Area of Science:

• Cognitive Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Sustained attention tasks are crucial in many professions, and cognitive fatigue significantly impacts performance.
• Physiological markers of mental fatigue are increasingly understood, suggesting potential for objective detection.
• Existing research highlights the need for a generalized model to detect vigilance decrements across diverse tasks and participants.

Purpose of the Study:

• To develop and test a task-generic, cross-participant electroencephalogram (EEG) model for detecting vigilance decrements.
• To investigate the feasibility of a generalized EEG model that works across unseen individuals and tasks.
• To compare the efficacy of different machine learning models in identifying vigilance decrements from EEG data.

Main Methods:

• Utilized three distinct machine learning models: a multi-layer perceptron neural network (MLPNN) using spectral EEG features, and two temporal convolutional network (TCN) models (TCN and TCN autoencoder) using raw EEG time-series data.
• Extracted spectral features from five traditional EEG frequency bands for the MLPNN.
• Applied a 7-fold cross-validation approach to evaluate model performance.

Main Results:

• Both the MLPNN and TCN models demonstrated accuracy significantly above random chance (50%).
• The MLPNN achieved the highest balanced accuracy of 64% (95% CI: 0.59, 0.69).
• The MLPNN showed validation accuracies exceeding random chance for 9 out of 14 participants.

Conclusions:

• It is feasible to classify vigilance decrements using EEG data, even from individuals and tasks not included in the model's training set.
• The developed task-generic EEG model shows promise for objective monitoring of cognitive fatigue in real-world scenarios.
• Further research can refine these models for broader application in performance monitoring and safety-critical domains.