Consumer Smartwatches As a Portable PSG: LSTM Based Neural Networks for a Sleep-Related Physiological Parameters...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Wearable Technology

Background:

• Mobile and wearable devices offer non-invasive health monitoring but face limitations like intermittent data and power constraints.
• Continuous monitoring of sleep-related physiological parameters is crucial for diagnosing various health conditions.

Purpose of the Study:

• To develop a deep recurrent neural network system for automatic, continuous, non-invasive monitoring of sleep stages, respiratory events, snoring, and blood oxygen saturation (SpO2).
• To evaluate the accuracy and efficacy of a smartwatch-based algorithm for sleep disorder screening.

Main Methods:

• A deep recurrent neural network model was developed for analyzing biosignals from wearable monitoring systems.
• The algorithm was trained and validated for classifying sleep stages, respiratory events (sleep apnea/hypopnea), snore events, and SpO2 levels.

Main Results:

• The model achieved 77% accuracy in sleep stage prediction.
• Respiratory event classification accuracy exceeded 80% on an epoch-by-epoch basis.
• Snore event classification accuracy was above 60%, and SpO2 level classification accuracy was above 70% (two-class problem, 95% threshold).

Conclusions:

• The developed deep recurrent neural network system provides a non-invasive, cost-effective screening tool for sleep-related physiological parameters and pathological states.
• Smartwatch-based biosignal monitoring integrated with deep learning shows significant potential for remote and continuous health assessment.