Sleep stage classification from heart-rate variability using long short-term memory neural networks.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Sleep Medicine

Background:

• Heart rate variability (HRV) analysis offers a non-invasive method for sleep monitoring.
• Current automated sleep stage classification methods struggle with long-term sleep architecture.
• Polysomnography, the gold standard, is costly and not suitable for home use.

Purpose of the Study:

• To develop and validate a long short-term memory (LSTM) network for automated sleep stage classification using HRV.
• To assess the model's ability to capture long-term cardiac sleep architecture.
• To compare the LSTM model's performance against existing state-of-the-art methods.

Main Methods:

• A comprehensive dataset of 584 nights from 292 participants was used.
• Sleep stages were annotated using the Rechtschaffen and Kales (R&K) standard.
• A deep learning LSTM network was employed to model temporal dependencies in HRV data.

Main Results:

• The LSTM model achieved a Cohen's kappa of 0.61 ± 0.15 and accuracy of 77.00 ± 8.90%.
• The proposed model outperformed existing methods limited to non-temporal or short-term recurrent classifiers.
• Performance showed a potential decline in individuals aged 50 and older.

Conclusions:

• Deep temporal modeling using LSTM networks advances HRV-based sleep stage classification.
• The model demonstrates potential for ergonomic, low-cost, home-based sleep monitoring.
• Further investigation is needed for older populations (50+) due to observed performance variations.