A Connectivity-Aware Graph Neural Network for Real-Time Drowsiness Classification.

Area of Science:

• Neuroscience
• Machine Learning
• Transportation Safety

Background:

• Drowsy driving is a major cause of road fatalities.
• Electroencephalography (EEG) detects drowsiness via brain activity.
• Brain connectivity graphs show potential for drowsiness prediction.

Purpose of the Study:

• To develop a task-relevant brain connectivity network for drowsy driving detection.
• To improve real-time drowsiness prediction accuracy.
• To identify specific brain connectivity patterns associated with drowsiness.

Main Methods:

• Proposed a connectivity-aware graph neural network (CAGNN) with a self-attention mechanism.
• Utilized end-to-end training to generate task-relevant connectivity networks.
• Incorporated a squeeze-and-excitation (SE) block for feature importance analysis.

Main Results:

• Achieved 72.6% accuracy in drowsy driving detection, outperforming existing CNNs and graph generation methods.
• Identified reduced occipital and interregional connectivity in drowsy states.
• Demonstrated that SE attention scores highlight crucial feature bands and identified alpha spindles as a drowsiness indicator.

Conclusions:

• CAGNN effectively generates task-relevant brain connectivity networks for improved drowsy driving detection.
• The method provides insights into brain connectivity changes during drowsiness.
• The approach offers a promising tool for enhancing driver safety through real-time drowsiness monitoring.