[Sleep apnea automatic detection method based on convolutional neural network].

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Traditional machine learning for sleep apnea (SA) detection requires extensive feature engineering and classifier design.
• Existing methods may lack the efficiency and accuracy needed for widespread clinical application.

Purpose of the Study:

• To develop and validate a deep learning model for automated sleep apnea detection using single-channel ECG signals.
• To evaluate the performance of a 1D CNN model in extracting and classifying features for SA detection.

Main Methods:

• A 1D convolutional neural network (CNN) model was constructed with four convolution, four pooling, and two fully connected layers.
• The model was trained and validated using the Apnea-ECG dataset, comprising whole-night single-channel ECG signals from 70 subjects.
• Performance was assessed using various input signal types: ECG, RR interval (RRI) sequence, R peak sequence, and combined RRI + R peak sequences.

Main Results:

• The CNN model achieved per-segment SA detection accuracies ranging from 80.1% to 88.0% depending on the input signal.
• The highest performance was observed using RRI sequence + R peak sequence, yielding 88.0% accuracy, 85.1% sensitivity, and 89.9% specificity for per-segment detection.
• An impressive 100% accuracy was achieved for per-recording SA diagnosis.

Conclusions:

• The proposed 1D CNN model effectively automates feature extraction and classification for sleep apnea detection from ECG signals.
• The method demonstrates high accuracy and robustness, outperforming recent approaches and showing potential for portable screening devices.