A computationally efficient algorithm for wearable sleep staging in clinical populations.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Sleep Medicine

Background:

• Accurate sleep staging is crucial for diagnosing sleep disorders.
• Traditional polysomnography is resource-intensive.
• Developing efficient, non-invasive sleep monitoring tools is essential.

Purpose of the Study:

• To develop and validate a computationally efficient algorithm for 4-class sleep staging.
• To assess the performance and execution time compared to existing methods.
• To explore the potential of neural networks in discovering sleep stage patterns from physiological signals.

Main Methods:

• Utilized an accelerometer for body movements and a photoplethysmographic (PPG) sensor for cardiac activity (interbeat intervals, heart rate).
• Trained a neural network to classify sleep stages (wake, N1/N2, N3, REM) in 30-second epochs.
• Validated the algorithm against polysomnography (PSG) and compared execution time with a heart rate variability (HRV) based algorithm.

Main Results:

• Achieved a median epoch-per-epoch kappa (κ) of 0.638 and accuracy of 77.8%.
• Demonstrated equivalent performance to a previously developed HRV-based algorithm.
• Showcased a 50-times faster execution time compared to the HRV-based approach.

Conclusions:

• The neural network algorithm effectively maps cardiac and movement data to sleep stages without prior domain knowledge.
• The algorithm's efficiency and performance make it suitable for practical implementation in sleep diagnostics.
• This approach offers new possibilities for accessible and automated sleep analysis, even in patients with sleep pathologies.