Sleep Apnea Severity Estimation from Respiratory Related Movements Using Deep Learning.

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Wearable Technology

Background:

• Sleep apnea is a prevalent respiratory disorder characterized by breathing cessations during sleep.
• Current diagnosis relies on polysomnography, which is costly, time-consuming, and inconvenient for patients.
• There is a significant need for accessible, portable, and reliable sleep apnea screening tools.

Purpose of the Study:

• To develop and evaluate a wearable, accelerometer-based system for estimating sleep apnea severity.
• To assess the system's performance against the gold standard polysomnography.
• To provide a convenient and robust solution for home-based sleep apnea screening.

Main Methods:

• A 3D accelerometer was used to record respiratory-related tracheal movements at the suprasternal notch.
• Twenty-one physiological features were extracted from the accelerometer data.
• Three deep learning models (CNN, RNN, and combined) were employed to estimate the apnea-hypopnea index (AHI).

Main Results:

• The system achieved a high correlation coefficient (r=0.84) between estimated and gold standard AHI values in 20 participants.
• The deep learning models demonstrated effectiveness in analyzing tracheal movements for sleep apnea assessment.
• The portable system showed accuracy comparable to traditional diagnostic methods.

Conclusions:

• The proposed accelerometer-based system is an accurate, convenient, and portable device for sleep apnea screening.
• This technology can facilitate widespread home-based monitoring and early detection of sleep apnea.
• The findings support the potential of wearable sensors and AI in managing chronic respiratory conditions.