Sleep apnea diagnosis based on respiratory related movements
Maziar Hafezi (1), Nasim Montazeri (1, 2), Kayin Zhu (1,2), Azadeh Yadollahi (1, 2), Babak Taati (2)
1: KITE, Toronto Rehabilitation Institute-University Health Network, Toronto, ON Canada
2: Institute of Biomaterials and Biomedical Engineering, University of Toronto, Toronto, ON Canada
Sleep apnea is a common chronic respiratory disorder which occurs due to the repetitive complete or partial cessations of breathing during sleep. The gold standard assessment of sleep apnea requires full night polysomnography (PSG) in a sleep laboratory. However, PSG is expensive, time consuming, and inconvenient. Consequently, about 85% of the adults with sleep apnea remain undiagnosed. Hence, there is an urgent need for a convenient, robust and wearable monitoring device for screening of sleep apnea. Portable devices have developed to diagnose sleep apnea by monitoring different physiological signals. This study investigates the possibility of diagnosing sleep apnea severity by tracking the respiratory movements recorded over the suprasternal notch which is close to the pharyngeal airway and the potential site of airway collapse. Method: Participants who were referred to the sleep laboratory of the Toronto Rehabilitation Institute for sleep studies were included. Simultaneously with PSG, a 3D accelerometer attached to subjects’ trachea was used to record respiratory movements. Inclusion criteria were adults >18 years with no allergic reactions to clinical tape. Accelerometer signals were filtered and 7 features were extracted from each axis. A supervised deep learning classifier was developed to detect apneas and hypopneas by analyzing the extracted respiratory features. Apnea hypopnea index (AHI) was estimated by counting the number of detected events per hour of sleep. We evaluated the agreement between estimated AHI and PSG-derived AHI using Pearson correlation and Bland-Altman test. Also, the performance of the algorithm in estimating the severity of sleep apnea was validated based on AHI cut-offs of 10 and 15 events/hr. Results: 69 subjects (32 females, age: 52.1 ± 16.0 years, BMI: 29.5 ± 6.3kg/m2) participated in the study. A high correlation between the estimated AHI and the PSG AHI was observed (r = 0.92, p < 0.0001). The Bland-Altman limits of agreement were −19.71 to 20.64 (mean=0.46). For AHI cut-off of 10/hour, the sensitivity, specificity and accuracy were 87%, 77%, 83%, respectively. For AHI cut-off of 15/hour, the sensitivity, specificity and accuracy were 79%, 82%, and 81%, respectively. Conclusion: The AHI estimated using tracheal movements showed a high agreement with the PSG-AHI. Once validated in a larger clinical cohort, the proposed algorithm can be used to develop a wearable technology for assessing sleep apnea severity.