Detecting arousals and sleep from respiratory inductance plethysmography

Area of Science:

• Sleep Medicine
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate sleep state and arousal identification are crucial for diagnosing sleep disorders.
• Polysomnography (PSG) is the gold standard but is expensive and lab-based.
• Home sleep testing (HST) is more accessible but limited in sleep and arousal assessment without electroencephalography.

Purpose of the Study:

• To evaluate a deep learning algorithm for determining sleep states (REM, NREM, Wake) and arousals from breathing signals.
• To assess the algorithm's performance in comparison to traditional sleep study metrics.

Main Methods:

• A novel deep learning algorithm was developed using respiratory inductance plethysmography signals.
• Sleep states were classified in 30-second epochs, and arousal probabilities were calculated at 1-second resolution.
• Validation was performed on a clinical dataset of 1,299 adults, analyzing sensitivity, specificity, arousal index (ArI), and total sleep time (TST).

Main Results:

• The algorithm demonstrated high accuracy in classifying sleep states (e.g., NREM: 93.9% sensitivity, 80.4% specificity).
• Arousal detection achieved 66.1% sensitivity and 86.7% specificity.
• Bland-Altman analysis showed acceptable agreement for ArI and TST, with intraclass correlations of 0.74 for ArI and 0.91 for TST.

Conclusions:

• The deep learning algorithm reliably identifies sleep states and arousals from breathing signals.
• Performance is comparable to the inherent variability in manual sleep scoring.
• This technology has the potential to enhance HST, making sleep diagnostics more accessible, cost-effective, and dependable.