Sleep/wakefulness detection using the tracheal respiratory sound and movements

Nasim Montazeri Ghahjaverestan (1,2), Bojan Gavrilovic (1,2), Shumit Saha (1,2), Babak Taati (1,2,3), Azadeh Yadollahi (1,2)

1: Institute of Biomaterials and Biomedical Engineering, University of Toronto, Toronto, ON Canada

2: KITE, Toronto Rehabilitation Institute-University Health Network, Toronto, ON Canada

3: Computer science, University of Toronto, Toronto, ON Canada

Introduction: The gold-standard to assess sleep/wakefulness in polysomnography (PSG) is electroencephalogram, which is inconvenient. Thus, portable sleep diagnosis devices cannot detect sleep time. Compared to wakefulness, sleep is associated with reduction in the activity of pharyngeal dilator muscles and ventilation resulting in lower respiratory drive and more regular breathing pattern. Also, it has been shown that the respiratory activity can be assessed conveniently using tracheal sound and movement. Hence, the goal of this study was to detect sleep/wakefulness by analyzing tracheal sounds and movements compared to the PSG-scores. Materials and Methods: Participants with suspected sleep apnea (SA) who were referred to the sleep laboratory of Toronto Rehabilitation Institute were included in this study. Simultaneously with full PSG, tracheal sounds and movements were recorded with a small wearable device attached to the suprasternal notch. Dominant frequency and variance of autocorrelation peaks of tracheal sound were extracted. Also, movement spikes, zero-crossing rate and absolute velocity of tracheal movements were calculated. The features were extracted from 30-second epochs. Principal component analysis was used to extract the components representing the maximum variability in features. Principal components of each epoch were classified into sleep or wakefulness using nonlinear support vector machine. Pearson or Spearman’s correlation was used to compare the estimated total sleep time (TST) and sleep efficiency (SE) with PSG. Results: Sixty three subjects (31 Females, age: 51±16 yrs, BMI: 29.6±6.4 kg/m2 and number of apnea/hypopnea per hour: 12.6 (0.6-146)) were included in this study. The accuracy of sleep/wakefulness detection was 82.3±8.66% with sensitivity of 87.6 ± 8.0 % (sleep), specificity of 68.4 ± 18.2% (awake), F1 of 88.2±7.9% and Cohen's kappa of 0.49. The correlation between the estimated and gold standard PSG based measures for TST and SE were 0.75* and 0.55*, respectively. (*p<0.001) Conclusion: This study shows the extracted features from tracheal sound and movement are reliable to detect sleep/wakefulness status. Previously, we have shown that tracheal sound analysis can be used reliably to diagnose apnea/hypopnea during sleep. The results of this study combined with our previous studies provide strong evidence that respiratory sounds analysis can be used to develop convenient and cost-effective portable devices for SA monitoring.