Detection of Cortical Arousals in Sleep Using Multimodal Wearable Sensors and Machine Learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Sleep Medicine

Background:

• Cortical arousals disrupt sleep continuity, leading to various health impairments.
• Polysomnography, the standard for arousal detection, is costly and complex for widespread use.
• Wearable technology offers a potential solution for accessible sleep monitoring.

Purpose of the Study:

• To develop and validate a noninvasive, machine learning-based framework for detecting cortical arousals using wearable physiological signals.
• To assess the framework's performance in a pediatric cohort with attention-deficit/hyperactivity disorder (ADHD) and sleep concerns.

Main Methods:

• Utilized the RestEaze™ system, a leg-worn wearable recording accelerometry, gyroscope, photoplethysmography (PPG), and temperature.
• Applied machine learning classifiers including logistic regression, XGBoost, and Random Forest.
• Evaluated model performance using metrics like ROC AUC, precision, recall, and F1-score.

Main Results:

• Movement intensity features were most effective for arousal detection, outperforming heart rate variability.
• The Random Forest model demonstrated superior performance with a ROC AUC of 0.94.
• For the arousal class, the Random Forest model achieved a precision of 0.57, recall of 0.78, and F1-score of 0.65.

Conclusions:

• Wearable-based machine learning is feasible for real-world cortical arousal detection.
• The developed framework shows promise for monitoring sleep disruption in pediatric populations, specifically those with ADHD.
• This approach offers a more accessible alternative to traditional polysomnography for long-term sleep studies.