Sleep Apnea Diagnosis using Tracheal Respiratory Sounds and Movement

Shumit Saha (1,2), Muammar Kabir (1,2), Nasim Montazeri (1,2),

Hisham Alshaer (2.3), Azadeh Yadollahi (1.2)

1. Institute of Biomaterials and Biomedical Engineering, University of Toronto

2. KITE, Toronto Rehabilitation Institute, University Health Network

3. BresoTec Inc.

Background: Sleep apnea is a chronic respiratory disorder due to intermittent partial (hypopnea) or complete (apnea) collapse of the pharyngeal airway during sleep. 26% of the Canadian adults are at high risk of sleep apnea. However, due to the complexity and limited access to polysomnography (PSG), 84% of Canadians who are at high risk of sleep apnea are not diagnosed. To address this problem, a robust and cost-effective home based technology to assess sleep apnea severity is required. Thus, we aimed to develop a new algorithm for sleep apnea diagnosis using respiratory sounds and respiratory related movement recorded over trachea.

Methods: Adults referred to the sleep lab of Toronto Rehabilitations Institute for suspected sleep apnea were recruited for this study. Simultaneously with PSG, respiratory sounds and respiratory movement were recorded over the suprasternal notch using a wearable device developed by our group, which includes a microphone and an accelerometer. We developed an algorithm to differentiate breathing and snoring segments from the respiratory sounds. The accelerometer signal was low-pass filtered to extract respiratory related movements. Energy and duration of breathing and snoring segments as well as the magnitude of respiratory related movements were extracted. Extracted features were normalized between 0 and 1 and the weighted averages of the features were compared with an adaptive threshold to detect the events. We increased the threshold in the breathing segments around the time that subject was upright which shows high probability of wakefulness, and lowered the threshold for the supine position. The number of apneas and hypopneas per hour of recording time (apnea-hypopnea index, AHI) was estimated. Estimated AHI was compared to the AHI obtained from PSG (PSG-AHI) scored by technicians according to standard criteria.

Results: Data from 59 subjects, age: 50.2±16.2 years, BMI: 29.6±5.4 kg/m2 were investigated. A high correlation was found between the estimated AHI and PSG-AHI (r = 0.84, p<0.01). Considering AHI cut-off of 15, sensitivity and specificity of diagnosing sleep apnea were 84.0% and 85.3%, respectively.
Conclusion: Utilizing a microphone and an accelerometer embedded in a small wearable device, we could achieve very high accuracies in diagnosing sleep apnea. Introduction of small, cost-effective and easily accessible wearable devices will significantly increase the diagnostic rate of sleep apnea.