Scribd
Upload
8 views

11. Gaussian Mixture Models for Detecting Sleep Apnea Events Using Single Oronasal Airflow Record

Uploaded by

hod.ece
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
8 views

11. Gaussian Mixture Models for Detecting Sleep Apnea Events Using Single Oronasal Airflow Record

Uploaded by

hod.ece
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOC, PDF, TXT or read online on Scribd
You are on page 1/ 2

Gaussian Mixture Models for Detecting Sleep Apnea Events Using Single Oronasal Airflow Record

Hisham Elmoaqet, Jungyoon Kim, Dawn Tilbury, Krishna Satya, Ramachandran, Mutaz
Ryalat, Chao-Hsien Chu
Applied Sciences
2020
18 citations
Semantic Scholar

Summary
A new algorithm is developed for automated detection of sleep apnea events using a single
oronasal airflow sensor, which leverages AASM recommendations and uses a probabilistic
GMM-based framework to achieve significantly improved performance over previous
methods.
Methodology
- A data-driven approach is used to derive the respiratory flow baseline from the airflow
signal. - A sliding window method is used, with a baseline window (W_b) to derive the local
respiratory baseline, and a detection window (W_m) to detect apneic events based on
relative changes in inter-breath intervals and breath amplitudes. - A probabilistic approach
using a Gaussian Mixture Model (GMM) is proposed for the classification of apneic events.
Limitations
- Class imbalance problem leading to false positives - Low airflow signal quality leading to
false positives - Irregular breathing patterns and artifacts leading to false positives - High
airflow peak amplitudes after apnea events contributing falsely to the respiratory baseline
and leading to false positives - Need for more advanced signal filtration algorithms to handle
artifacts and low SNR airflow segments
Summary of introduction
The introduction provides background on sleep apnea, stating that it is a highly prevalent
sleep disorder that can cause significant health issues, and motivates the need for
alternative, automated methods for detecting sleep apnea using fewer sensors than the gold
standard polysomnography (PSG).
Study objectives
- Develop a probabilistic classification approach for automated detection of sleep apnea
events using only the oronasal airflow record - Evaluate the performance of the proposed
approach on a large dataset of 96 patients with varying sleep apnea severity levels
Research question
How can we develop an automated system for detecting sleep apnea events that overcomes
the limitations of current rule-based threshold detectors?
Research gaps
- Limited diagnostic accuracy of existing device technologies for detecting apnea events,
even when using multiple physiological inputs - Limitations of classical rule-based threshold
detectors that overlook the statistical distributions of input features and output classes -
Lack of research on probabilistic approaches to classification for sleep apnea detection -
Limitations of single-respiration channel based apnea detection methods, which have lower
diagnostic accuracy compared to methods using three or more physiological parameters
Future research
The authors suggest future research could focus on: 1) Adding more advanced signal
filtration algorithms to handle artifacts and low SNR in the airflow signal; 2) Incorporating
the nasal pressure signal in addition to the airflow signal to enable detection of both apnea
and hypopnea events, improve respiratory baseline estimation, and provide diagnosis of
apnea type; 3) Adding the respiratory effort signal as an additional input to further improve
apnea type diagnosis.
Dataset
The dataset used in the study consisted of full polysomnography (PSG) data collected from
96 patients at the University of Michigan Sleep Disorders Center. The dataset included
various physiological signals, including oronasal airflow (FlowTH). 66 patients were used for
training the proposed modeling framework, and 30 distinct patients were used for testing,
including 5 none/minimal apnea, 5 mild, 5 moderate, and 15 severe sleep apnea patients.
Hypotheses tested
- A probabilistic classification approach using Gaussian Mixture Models can effectively detect
sleep apnea events using only the oronasal airflow record. - This probabilistic approach can
perform well across a large dataset of patients with varying sleep apnea severity levels.
Study design
The study design was a retrospective, non-controlled, non-randomized, observational study
with a training set of 66 patients and a test set of 30 distinct patients, stratified by sleep
apnea severity levels.
Algorithms
The two main algorithms introduced in the study are: 1. A rule-based classifier that uses a
two-step optimization method to learn classification thresholds for the input features I_c and
A_c. 2. A Gaussian Mixture Model (GMM) classifier that models the probability density
function of the input features [I_c, A_c] as a weighted sum of Gaussian distributions, and
uses the learned GMM parameters to compute the posterior probability of apnea.
Summary of discussion
The paper developed a probabilistic framework based on Gaussian Mixture Models to detect
sleep apnea events using a single oronasal airflow signal, which leverages AASM
recommendations and can better handle uncertainty compared to previous single-channel
methods.

You might also like