Scribd
Upload
6 views

1. Detection of sleep apnea using deep neural networks and single

Uploaded by

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

1. Detection of sleep apnea using deep neural networks and single

Uploaded by

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

Detection of sleep apnea using deep neural networks and single-lead ECG signals

Asghar Zarei, Hossein Beheshti, Babak Mohammadzadeh Asl


Biomedical Signal Processing and Control
2022
35 citations
Semantic Scholar

Summary
The paper presents a novel deep learning-based method for automatic detection of
obstructive sleep apnea using single-lead ECG signals, which outperforms previous state-of-
the-art approaches.
Methodology
- Preprocessing the ECG signals using a Chebyshev bandpass filter - Dividing the ECG signals
into 1-minute or 30-second segments - Eliminating noisy segments using an automatic
weight calculation algorithm - Using a deep learning architecture with: - 4 2D CNN layers to
extract spatial features - 3 LSTM layers to extract temporal features - Fully connected layers
to classify apnea vs. normal segments
Limitations
- The effect of using transfer learning to improve the performance of the deep learning
models should be investigated in future studies. - The performance of the OSA detection
algorithm in the presence of arrhythmias should be investigated, as both OSA and
arrhythmias can cause changes in the ECG signal.
Summary of discussion
The paper presents a new deep learning-based algorithm for classifying apnea events and
normal segments, which outperforms conventional feature engineering and classification
methods and can help physicians make more accurate decisions in diagnosing apnea
patients.
Summary of introduction
The introduction summarizes the key problems with existing sleep apnea detection methods
and proposes an automatic feature extraction method using a combination of CNN and LSTM
networks to address these issues.
Study objectives
- Develop an automatic feature extraction and classification technique for sleep apnea
detection using deep neural networks - Develop a fully automatic, end-to-end deep learning-
based method for obstructive sleep apnea (OSA) detection
Research question
How can we develop an automatic and robust method for detecting sleep apnea using
single-lead ECG signals?
Research gaps
- The performance of previous OSA recognition methods depends heavily on the feature
extraction approaches, which rely on the experience and prior knowledge of specialists. -
Previous studies have not focused on improving the performance of the classification models
themselves.
Hypotheses tested
- An automatic feature extraction method can be developed by combining CNN and LSTM
networks to distinguish apnea events from normal segments. - The AHI can be used to
discriminate apnea subjects from healthy ones. - A fully automatic, end-to-end deep
learning-based OSA detection method can be developed, where all feature extraction and
classification is done using deep learning layers.
Future research
1. Investigate the effect of using transfer learning to improve the performance of deep
learning models for OSA detection. 2. Develop an OSA detection algorithm that can work in
the presence of arrhythmias.
Dataset
The datasets used in the study are: 1. Apnea-ECG dataset: Contains 35 recordings in a
released set and 35 recordings in a withheld set, with one-minute ECG segments labeled as
normal or apnea. The total number of segments is 17,045 in the released set and 6,550 in
the withheld set. The subjects are 57 men and 13 women, aged 27-63 years and weighing
53-unknown kg. 2. UCDDB dataset: Contains 25 full overnight polysomnograms with
simultaneous three-channel Holter ECG, recorded from adult subjects with suspected sleep-
disordered breathing. The subject weights are 59.8-128.6 kg, ages are 28-68 years, and AHI
values are 1.7-90.9. The ECG recordings have a sampling frequency of 128 Hz.

You might also like