Estimating the Severity of Obstructive Sleep Apnea Using ECG, Respiratory Effort and Neural Networks

Area of Science:

• Cardiorespiratory signal analysis
• Artificial intelligence in medicine
• Sleep medicine

Background:

• Wearable sensor technology has advanced, but clinical use for obstructive sleep apnea (OSA) assessment is hindered by a lack of large, representative datasets synchronized with polysomnography (PSG).
• Current OSA assessment often relies on airflow and oxygen saturation (SpO2) measurements, limiting unobtrusive monitoring.
• Cardiorespiratory signals are fundamental to PSG and are readily measurable with wearable devices.

Purpose of the Study:

• To explore the feasibility of using common cardiorespiratory signals, measurable by wearable sensors, to estimate OSA severity.
• To develop and evaluate an artificial neural network for detecting sleep disordered breathing events and estimating OSA severity.

Main Methods:

• An artificial neural network was developed to detect sleep disordered breathing events using electrocardiogram (ECG) and respiratory effort signals.
• This network was integrated with a sleep staging algorithm and validated against PSG in 653 participants.
• Performance was assessed for sleep staging, apnea-hypopnea index (AHI) estimation, and OSA severity classification.

Main Results:

• Four-class sleep staging achieved a Cohen's kappa (κ) of 0.69 compared to PSG.
• Apnea-hypopnea index (AHI) estimation demonstrated a high intraclass correlation coefficient (ICC) of 0.91.
• The model exhibited strong diagnostic performance across various OSA severity thresholds.

Conclusions:

• Cardiorespiratory signals alone are sufficient for estimating OSA severity, without requiring airflow or SpO2 data.
• This approach shows promise for developing practical, unobtrusive wearable systems for long-term sleep disordered breathing monitoring.
• Existing large datasets can serve as a basis for future wearable OSA monitoring systems, enabling new clinical research avenues.