The Combination of Topological Data Analysis and Mathematical Modeling Improves Sleep Stage Prediction From...

Area of Science:

• Biomedical Engineering
• Data Science
• Sleep Medicine

Background:

• Wearable devices collect vast physiological data, but noise and complexity hinder clinical sleep analysis.
• Existing wearable-based sleep scoring methods have limitations in accuracy and generalizability.

Purpose of the Study:

• To develop a novel neural network algorithm for accurate sleep stage prediction using wearable data.
• To enhance sleep analysis by incorporating topological features (TFs) and clock proxies (CPs).

Main Methods:

• A neural network was developed to analyze topological features and clock proxies from wearable motion and heart rate data.
• The algorithm was validated against polysomnography (PSG) in young and elderly cohorts.
• Performance was compared against existing state-of-the-art wearable sleep scoring algorithms.

Main Results:

• The algorithm incorporating TFs and CPs significantly improved Wake/REM/NREM sleep classification accuracy (>12%) compared to using raw data alone.
• Heart rate TFs were identified as a key contributor to performance enhancement.
• The algorithm demonstrated superior performance across different populations and outperformed existing methods.

Conclusions:

• Combining topological data analysis and mathematical modeling with wearable data offers a powerful approach for robust sleep stage prediction.
• This method improves the clinical utility of wearable sensor data for sleep research and applications.
• The developed algorithm represents a significant advancement in wearable-based sleep scoring technology.