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

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Cardiology

Background:

• Obstructive sleep apnea (OSA) is a prevalent respiratory disorder with underdiagnosis due to monitor limitations.
• Existing monitoring equipment often restricts widespread detection of OSA.
• There is a need for accessible and portable OSA monitoring solutions.

Purpose of the Study:

• To develop and evaluate a novel sleep-monitoring model for detecting obstructive sleep apnea (OSA).
• To utilize single-channel electrocardiogram (ECG) data for OSA detection via a convolutional neural network (CNN) and long short-term memory (LSTM) model.
• To enable the use of portable devices for OSA monitoring.

Main Methods:

• A CNN-LSTM model was designed, incorporating multi-scale filters in the initial convolution layer and LSTM for temporal dependencies.
• Raw ECG signals were segmented using a 10-second overlapping sliding window for model training and testing.
• The model employed a softmax function for final OSA event classification.

Main Results:

• The proposed model achieved high performance on the Apnea-ECG dataset, with Cohen's kappa of 0.92.
• Excellent detection metrics were reported: 96.1% sensitivity, 96.2% specificity, and 96.1% accuracy.
• The model significantly outperformed baseline methods in OSA event detection.

Conclusions:

• The developed CNN-LSTM model demonstrates a highly effective and accurate approach for detecting obstructive sleep apnea using single-lead ECG.
• This method offers a promising tool for portable and accessible OSA monitoring.
• The findings support the utility of ECG-based analysis for diagnosing sleep-related respiratory disorders.