Comparison of sleep-wake classification using electroencephalogram and wrist-worn multi-modal sensor data

Area of Science:

• Biomedical Engineering
• Sleep Science
• Wearable Technology

Background:

• Accurate sleep-wake classification is crucial for diagnosing sleep disorders and monitoring overall health.
• Traditional methods like polysomnography (PSG) are resource-intensive and not suitable for continuous monitoring.
• Wearable sensors offer a promising alternative for unobtrusive, long-term sleep tracking.

Purpose of the Study:

• To compare the efficacy of electroencephalogram (EEG) versus multi-modal wrist wearable sensor data for sleep-wake classification.
• To identify key physiological features from wearable sensors that contribute most to accurate sleep-wake detection.
• To evaluate both intra-subject and inter-subject classification performance.

Main Methods:

• Collected simultaneous electroencephalogram (EEG), skin conductance (SC), skin temperature (ST), and acceleration (ACC) data from 15 college students during sleep.
• Extracted relevant features from both EEG and wrist wearable sensor data.
• Performed intra- and inter-subject classification analysis to assess sleep-wake states.

Main Results:

• EEG-based features achieved a classification accuracy of 83% for sleep-wake states.
• Wrist wearable sensor data yielded a classification accuracy of 74%.
• The combination of acceleration (ACC) and skin temperature (ST) data from the wrist sensor proved most influential for classification.

Conclusions:

• Electroencephalogram (EEG) remains a highly accurate method for sleep-wake classification.
• Wrist wearable sensors show potential for sleep-wake classification, though with lower accuracy than EEG.
• Specific features like acceleration and skin temperature are critical for improving the performance of wearable-based sleep monitoring systems.