Sleep Posture Detection Using an Accelerometer Placed on the Neck.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Sleep Science

Background:

• Sleep position monitoring is crucial for managing posture-related sleep disorders.
• Existing methods often use optimal body locations or non-contact sensing.
• Detecting sleep position from suboptimal locations, like the neck, can enable multi-modal sensing.

Purpose of the Study:

• To investigate the feasibility of detecting four main sleep positions (supine, prone, lateral right, lateral left) using neck-worn accelerometry.
• To develop and compare two distinct models for sleep position classification: a lightweight threshold-based model and an Extra-Trees classifier.
• To assess the performance of these models for potential multi-modal sleep monitoring applications.

Main Methods:

• Utilized accelerometry data from a single wearable device positioned on the neck.
• Developed an ultra-lightweight threshold-based model for sleep position detection.
• Implemented an Extra-Trees classifier with 25 base estimators and a maximum depth of 20.

Main Results:

• The threshold-based model achieved 95% average accuracy and a 0.89 F1-score on out-of-sample data.
• The Extra-Trees classifier demonstrated superior performance with 99% average accuracy and a 0.99 average F1-score.
• Both models exhibited high accuracy in detecting sleep posture from neck-based accelerometer signals.

Conclusions:

• Neck-worn accelerometers can effectively detect sleep positions with high accuracy.
• Simple, rule-based models can achieve moderate performance, while more complex classifiers yield excellent results.
• This approach facilitates multi-modal sleep monitoring by leveraging a suboptimal sensor location for posture detection.