Loading…
Detection of REM/NREM snores in obstructive sleep apnoea patients using a machine learning technique
Obstructive sleep apnoea (OSA) is a serious sleep disorder in which patients suffer from frequent upper airway (UA) collapse during sleep. UA muscle tone in OSA patients is known to vary with rapid eye movement (REM) and non-REM (NREM) sleep states. Information on sleep-state specific UA collapse ha...
Saved in:
Published in: | Biomedical physics & engineering express 2016-10, Vol.2 (5), p.55022 |
---|---|
Main Authors: | , |
Format: | Article |
Language: | English |
Subjects: | |
Citations: | Items that this one cites Items that cite this one |
Online Access: | Get full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | Obstructive sleep apnoea (OSA) is a serious sleep disorder in which patients suffer from frequent upper airway (UA) collapse during sleep. UA muscle tone in OSA patients is known to vary with rapid eye movement (REM) and non-REM (NREM) sleep states. Information on sleep-state specific UA collapse has clinical importance in sleep studies and treatment. This paper proposes a machine learning technique to label snoring sounds as belonging to REM or NREM sleep in OSA patients. Our method is based on analysing snore and breathing sound recordings from OSA patients acquired with non-contact bedside microphones. The reference standard to diagnose OSA and sleep states was the laboratory-based clinical polysomnography (PSG). We trained multilevel artificial neural networks (hierarchical neural networks) to label sleep into REM and NREM classes using snoring and breathing sounds around a given snoring episode of interest. A total of 41 062 snoring episodes were used for training and testing of the model (training to testing sample ratio was 3:1). The training data set was obtained from 12 subjects and the testing data set was from 7. The two data sets were mutually exclusive. Our method achieved a substantial (Cohen's kappa k = 0.62) agreement with the reference standard, that is, the PSG-based clinical classification. Overall, our proposed method achieved 90% testing accuracy of labelling individual snores as REM and NREM. |
---|---|
ISSN: | 2057-1976 2057-1976 |
DOI: | 10.1088/2057-1976/2/5/055022 |