7. Automatic Detection of Obstructive Sleep Apnea Events Using a Deep CNN

Uploaded by

hod.ece

Available Formats

Download as DOC, PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

3 views

7. Automatic Detection of Obstructive Sleep Apnea Events Using a Deep CNN

Uploaded by

hod.ece

Available Formats

Download as DOC, PDF, TXT or read online on Scribd

You are on page 1/ 2

Automatic Detection of Obstructive Sleep Apnea Events Using a Deep CNN-LSTM Model

Junming Zhang, Zhen Tang, Jinfeng Gao, Li Lin, Zhiliang Liu, Haitao Wu, Fang Liu, Ruxian Yao
Computational Intelligence and Neuroscience
2021
27 citations
Semantic Scholar

Summary
The paper proposes a deep learning model based on CNN and LSTM to automatically detect
obstructive sleep apnea (OSA) events from single-channel ECG signals, which achieves high
performance on the Apnea-ECG dataset.
Methodology
- Preprocessing the ECG data by applying a Chebyshev type-II band-pass filter (5-11 Hz) to
remove noise - Segmenting the ECG data into 10-second epochs - Using a 1D convolutional
neural network (CNN) to classify the ECG epochs - Applying batch normalization to improve
the training efficiency of the CNN - Using a long short-term memory (LSTM) network to learn
the long-term dependencies and transition rules in the ECG data - Experimenting with
different filter sizes in the CNN to extract features at different scales
Limitations
- The proposed model can only detect OSA and normal events, but not hypopnea events - It
is difficult for the method to score transition epochs
Summary of introduction
The introduction provides background on obstructive sleep apnea (OSA), a major sleep
disorder with serious health consequences, and highlights the limitations of the current gold
standard for OSA detection, polysomnography (PSG), motivating the need to develop new
methods for OSA detection that can be used at home with fewer signals.
Summary of discussion
The paper discusses the limitations of the proposed CNN-LSTM model, including its inability
to detect hypopnea events, difficulty in classifying transition epochs, and challenges in
detecting OSA events in noisy ECG data, and states that the authors will work on improving
the model's performance in the future.
Study objectives
- Develop a CNN-based model for detecting obstructive sleep apnea (OSA) events using
single-channel ECG signals, for use in portable OSA monitoring devices - Detect complete
OSA events by segmenting the raw ECG signals using a 10-second overlapping sliding
window and evaluating the model's performance
Research question
How can we develop a reliable method for detecting obstructive sleep apnea (OSA) using
fewer signals that can be used at home?
Research gaps
- The proposed model can only detect OSA and normal events, but not hypopnea events - It
is difficult for the proposed model to score transition epochs
Hypotheses tested
The specific hypothesis tested in this study is to develop an automated method for detecting
obstructive sleep apnea (OSA) events using single-channel electrocardiogram (ECG) signals,
without the need for manual feature extraction and selection.
Future research
The authors suggest the following future research directions: 1. Expand the model to detect
hypopnea events in addition to OSA and normal events. 2. Improve the model's ability to
accurately score transition epochs between normal and apnea events. 3. Develop
automated anomaly detection of ECG signals using CNN to rapidly assess ECG data quality.
Dataset
The Apnea-ECG dataset, downloaded from
https://www.physionet.org/content/apnea-ecg/1.0.0, which contains 70 PSG recordings (35
for training, 35 for testing) with ECG signals, apnea annotations, and QRS annotations