Use this URL to cite or link to this record in EThOS:
Title: Video and audio analysis for the detection of obstructive sleep apnoea
Author: Gederi, Elnaz
ISNI:       0000 0004 7232 3060
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Obstructive sleep apnoea (OSA) is a common sleep disorder characterised by serious sleep fragmentation due to repeated breathing pauses (OSA events) followed by brief awakenings. OSA is under-diagnosed with many consequences that may be life threatening. The standard screening test, polysomnography (PSG), requires an overnight stay in a clinic and the attachment of several on-body sensors which may be uncomfortable for some patients. Large-scale screening for OSA is limited to the availability of equipment and sleep specialists. The manual review of hours of PSG recordings is also cumbersome and costly. The increasing prevalence of OSA and the long waiting times for screening and treatment highlights a need for contactless home monitoring solutions. The research presented in this thesis aims to (i) investigate whether automatic processing of video recordings of sleep can be used to differentiate patients with OSA who more urgently require treatment and are more likely to respond to the treatment, and (ii) investigate the possibility of deriving respiratory information (respiratory signal, respiratory rate) from video recordings of sleep and scoring individual OSA events using the derived respiratory information. This thesis presents the analysis of PSG recordings from 259 patients with suspected OSA and 30 normal volunteers. First, PSG video recordings are used to identify moderate-to-severe OSA patients based on the pattern of their overnight gross-body movements. Then, PSG audio recordings are used to improve the identification of moderate-to-severe OSA patients based on their pattern of snoring, cessation of snoring, and choking sounds. Using the features derived from patients' gross-body movements and audio recordings, moderate-to-severe OSA patients are differentiated from mild-to-no OSA patients with an accuracy of 85.5%, sensitivity of 76.4%, and specificity of 91.1%. Next, a respiratory signal (VDR signal) is derived from the subtle respiratory movements in the PSG video recordings. Patients' respiratory rate (RR) per minute is estimated from the VDR signals. The mean absolute error for the derived RR is 0.82 breaths per minute (bpm) for mild-to-no OSA patients and 0.98 bpm for moderate-to-severe OSA patients. Finally, individual OSA events were scored in the VDR signals and an OSA severity index was calculated. The scored OSA events had an F1score of 76.0%, sensitivity of 81.0%, and positive predictive value of 73.8% when compared to the reference OSA events.
Supervisor: Tarassenko, Lionel ; Clifford, Gari Sponsor: RCUK
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available