Somnography using unobtrusive motion sensors and Android-based mobile phones

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Wearable Technology

Background:

• Sleep disorders affect up to 50% of the population, necessitating accessible diagnostic tools.
• Current gold standard polysomnography (PSG) is obtrusive and requires a sleep laboratory.
• There is a need for convenient, at-home sleep monitoring solutions.

Purpose of the Study:

• To develop an easy-to-use, unobtrusive method for home somnography using motion sensors.
• To classify sleep stages (wakefulness, REM, non-REM) using accelerometer data.
• To implement the sleep analysis algorithm on a mobile phone for real-time application.

Main Methods:

• Utilized Shimmer platform motion sensors placed in bed to record accelerometer data.
• Extracted motion features from 30-second epochs of accelerometer signals.
• Trained a Naive Bayes classifier using motion features and reference hypnograms from a SOMNOwatch system.

Main Results:

• The Naive Bayes classifier achieved a mean accuracy of 79.0% (SD 9.2%) in distinguishing wakefulness, REM, and non-REM sleep.
• The algorithm was successfully implemented on an Android mobile phone for real-time analysis.
• The system demonstrated feasibility for unobtrusive, at-home sleep stage assessment.

Conclusions:

• Developed a feasible and accurate method for home-based sleep stage classification using unobtrusive motion sensors.
• Mobile phone implementation enables real-time, accessible somnography.
• This approach offers a potential future for widespread, convenient sleep monitoring.