Scribd
Upload
3 views

2. Obstructive Sleep Apnea Detection Based on Sleep Sounds via Deep Learning

Uploaded by

hod.ece
Copyright
© © All Rights Reserved
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
3 views

2. Obstructive Sleep Apnea Detection Based on Sleep Sounds via Deep Learning

Uploaded by

hod.ece
Copyright
© © All Rights Reserved
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 2

Obstructive Sleep Apnea Detection Based on Sleep Sounds via Deep Learning

Bochun Wang, Xianwen Tang, Hao Ai, Yanru Li, Wen Xu, Xingjun Wang, Demin Han
Nature and Science of Sleep
2022
6 citations
Semantic Scholar

Summary
This paper presents a novel deep learning method called OSAnet that can detect sleep
apneic events and identify obstructive sleep apnea (OSA) based solely on sleep sounds
recorded by a non-contact audio recorder, which shows promise for OSA screening and
assessment in a community setting in a comfortable and low-cost manner.
Limitations
- The approach is dependent on sleep sounds, and a small percentage of OSA patients may
produce breathing sounds weaker than background noise, making the algorithm unsuitable
for them. - Night-to-night variability in sleep sounds may introduce inaccuracies in the
model. - The study was conducted at a single center, and the algorithm requires further
external validation in community populations and home-based settings.
Summary of introduction
The introduction provides background on obstructive sleep apnea (OSA), a prevalent sleep
disorder that can lead to various medical conditions, and highlights the need for alternative,
more accessible methods for OSA screening and diagnosis beyond the current gold standard
of polysomnography (PSG).
Summary of discussion
The novel deep learning method OSAnet, which detects obstructive sleep apnea events
based on sleep sounds recorded by a non-contact voice recorder, could be a useful tool for
home-based OSA screening and assessment in the community.
Study objectives
- Propose a novel deep-learning method for automatic sleep apneic event detection -
Estimate the apnea hypopnea index (AHI) based on sleep sounds - Identify obstructive sleep
apnea (OSA) in an event-by-event manner based on sleep sounds obtained by a noncontact
audio recorder
Research question
How can we develop a novel deep-learning method to detect sleep apneic events and
identify obstructive sleep apnea (OSA) based solely on sleep sounds recorded by a non-
contact audio recorder?
Research gaps
- The best combination of acoustic features for OSA detection remains to be exploited. -
There is a lack of a universal rule to compare the performance of different OSA detection
techniques. - The algorithm may not work well for a small percentage of OSA patients with
weak breathing sounds, and may be affected by night-to-night variability in sleep sounds. -
The algorithm requires further external validation in community populations and home-
based settings.
Hypotheses tested
The main hypothesis of this study is that a deep learning algorithm can accurately detect
sleep apnea events and estimate the apnea-hypopnea index (AHI) based solely on sleep
sounds recorded by a non-contact audio recorder.
Future research
Further studies to develop a tool for OSA detection based on a home-based setting, and
external validation of the algorithm in community populations in a home-based setting.
Dataset
The dataset consists of sleep sound recordings and polysomnography data from 194
participants referred for evaluation of habitual snoring or heavy breathing during sleep,
divided into a training group (n=116), a validation group (n=19), and a test group (n=59).
Algorithms
OSAnet, a deep convolutional neural network for automatic apneic event detection from
sleep sounds

You might also like