A Deep Transfer Learning Approach for Sleep Stage Classification and Sleep Apnea Detection Using Wrist-Worn Consumer...

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Obstructive sleep apnea (OSA) is a prevalent yet frequently undiagnosed sleep disorder.
• OSA poses significant risks to overall health and well-being.
• Current diagnostic methods can be resource-intensive.

Purpose of the Study:

• To develop and validate a deep transfer learning model for sleep apnea detection using consumer sleep technologies (CSTs).
• To assess the efficacy of using accelerometer and photoplethysmography data from wrist-worn devices for sleep apnea screening.
• To evaluate the model's performance across different data sources (clinical vs. CST).

Main Methods:

• A deep convolutional neural network (DNN) was employed for sleep stage classification and sleep apnea detection.
• The DNN utilized raw accelerometer and photoplethysmography signals from nocturnal recordings.
• Model training and internal testing used clinical and CST datasets, with external validation on a separate CST dataset.

Main Results:

• Training on clinical data significantly enhanced model performance.
• Direct raw data input outperformed feature-based input when using CST datasets.
• The model demonstrated good generalization, with slightly better performance on clinical data compared to CST data.
• The system showed particular strength in detecting events during REM sleep, correlating with arousal and oxygen desaturation.

Conclusions:

• Consumer sleep technologies hold potential as a viable screening tool for identifying undiagnosed obstructive sleep apnea cases.
• The developed deep transfer learning approach offers a promising method for improving sleep apnea detection across various wearable devices.
• Further validation and integration of CSTs could expand accessibility for OSA screening.