A domain adversarial network for small-sample obstructive sleep apnea detection using a single-lead piezoelectric...

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Wearable Technology

Background:

• Home monitoring for sleep health is advancing with wearable devices.
• Key challenges include unobtrusive sensing for long-term use and algorithms that generalize with limited data.

Purpose of the Study:

• To develop a comfortable, unobtrusive wearable system for home sleep monitoring.
• To create a robust algorithm for detecting sleep-disordered breathing events using limited data.

Main Methods:

• A single-lead bone-conduction monitoring system using a throat-placed piezoelectric sensor was developed.
• A Convolutional Memory Adaptive Prototypic Network (CMAP) model was designed for robust event recognition.
• The system was evaluated against polysomnography (PSG) in 95 participants.

Main Results:

• The system achieved high precision (0.90) and recall (0.93) for snore and breath sound event detection.
• The estimated Apnea-Hypopnea Index (AHI) strongly correlated with PSG (r=0.94).
• Accurate four-class severity classification for obstructive sleep apnea (OSA) was achieved with 83.2% accuracy.

Conclusions:

• A single-sensor, distribution-aware approach reliably estimates AHI and classifies OSA severity.
• The wearable system and CMAP model provide a practical, comfortable, and cost-effective solution for OSA screening and monitoring.
• This technology enables scalable home-based screening and longitudinal monitoring with low user burden.