Interbeat interval-based sleep staging: work in progress toward real-time implementation.

Area of Science:

• Cardiology
• Sleep Medicine
• Artificial Intelligence

Background:

• Sleep staging is crucial for diagnosing sleep disorders.
• Cardiac activity, specifically interbeat intervals (IBIs), offers a non-invasive method for sleep monitoring.
• Current methods for sleep staging can be cumbersome and require specialized equipment.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) algorithm for real-time sleep staging using IBIs.
• To assess the algorithm's performance across different sleep stages and in subjects with sleep disorders.
• To evaluate the algorithm's computational efficiency for potential real-time applications.

Main Methods:

• IBIs were extracted from electrocardiogram (ECG) signals and resampled.
• The IBI data was segmented into 150-second windows for DNN analysis.
• A DNN model comprising convolution and long short-term memory layers was trained and tested on multiple datasets.
• Performance was evaluated using Cohen's Kappa and accuracy, with optimization for stage-specific probability thresholds.

Main Results:

• The DNN achieved a mean Kappa of 0.44 and accuracy of 0.65 for 4 sleep stages in healthy subjects.
• For 3 sleep stages (light+deep, REM, wake), mean Kappa was 0.52 and accuracy was 0.76.
• Performance was significantly lower in subjects with REM behavior disorder (Kappa 0.24) and periodic limb movement disorder (Kappa 0.36).
• The algorithm demonstrated a processing time of 19.2 ms per window on an ARM microprocessor.

Conclusions:

• IBI-based sleep staging using DNN is feasible and offers a computationally efficient solution.
• The algorithm shows promise for real-time sleep staging with high specificity for deep and REM sleep.
• Further validation is needed for subjects with specific sleep disorders, but the algorithm's small footprint makes it suitable for various platforms and interventions.