Deep-Learning based Sleep Apnea Detection using SpO2 and Pulse Rate.

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Sleep apnea is a common disorder affecting millions.
• Current diagnostic methods can be invasive and costly.
• Non-invasive monitoring using pulse oximetry holds potential.

Purpose of the Study:

• To develop and validate an automated deep-learning algorithm for sleep apnea event detection.
• To assess the algorithm's performance using blood oxygen saturation and pulse rate data.
• To evaluate the algorithm's effectiveness across diverse patient populations and apnea subtypes.

Main Methods:

• Utilized a large dataset (n=8068) from the Sleep Heart Health Study.
• Trained a deep-learning model to detect apnea in 30-second epochs.
• Validated performance on independent test cohorts using ROC and precision-recall curves.

Main Results:

• Achieved 90.4% area under the ROC curve for epoch-based apnea detection.
• Demonstrated high sensitivity for obstructive apnea (93.4%), central apnea (90.5%), and hypopnea (89.1%).
• The algorithm showed robust performance across diverse patient groups and comorbidities.

Conclusions:

• The proposed algorithm provides a sensitive and robust method for automated sleep apnea detection.
• Utilizing non-invasive pulse oximetry makes this a cost-effective approach.
• This technology can facilitate convenient long-term sleep and health monitoring.