Automated REM vs NREM sleep staging using single overnight heart rate and accelerometer data.

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Wearable Technology

Background:

• Accurate sleep staging is essential for diagnosing sleep disorders like insomnia, REM behavior disorder, and sleep apnea.
• Polysomnography (PSG) is the gold standard but is costly, labor-intensive, and typically requires a clinical setting.
• There is a need for accessible, cost-effective sleep staging methods for widespread use.

Purpose of the Study:

• To develop and validate an algorithm for automated sleep staging using wearable device data.
• To assess the performance of a K-means/hidden Markov modeling approach for sleep stage classification.
• To evaluate the feasibility of using heart rate and triaxial acceleration data for low-cost sleep analysis.

Main Methods:

• A two-stage K-means/hidden Markov modeling (HMM) algorithm was developed.
• The algorithm utilized overnight heart rate and triaxial acceleration data from 13 participants' Apple Watches.
• Performance was benchmarked against polysomnography-scored sleep labels.

Main Results:

• The algorithm achieved an average accuracy of 81.5% in differentiating REM from NREM sleep.
• Macro F1-score was 0.73, and Matthews Correlation Coefficient (MCC) was 0.5.
• The study demonstrated strong performance in sleep staging using readily available consumer devices.

Conclusions:

• A Hidden Markov Model (HMM) approach using heart rate and accelerometer data offers a scalable and cost-effective alternative to polysomnography.
• This method has the potential to significantly improve at-home sleep tracking.
• The findings support the enhancement of personalized sleep health interventions through accessible technology.