A Deep Learning Framework for Automatic Sleep Apnea Classification Based on Empirical Mode Decomposition Derived from...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Polysomnography (PSG) is the gold standard for sleep apnea (SA) diagnosis but is cumbersome and requires expert interpretation.
• Electrocardiogram (ECG)-derived respiration (EDR) and heart rate variability (HRV) offer potential for automated SA detection.
• Existing machine learning approaches often require extensive feature engineering and expert knowledge.

Purpose of the Study:

• To develop a novel deep learning (DL) algorithm for automated sleep apnea (SA) detection.
• To differentiate between normal breathing and apnea events using ECG signals.
• To implement a computationally efficient and low-cost SA detection system.

Main Methods:

• Utilized a deep learning framework combining 1D and 2D deep Convolutional Neural Networks (CNNs).
• Employed Empirical Mode Decomposition (EMD) to preprocess and extract features from ECG signals.
• Validated the algorithm on overnight ECG recordings from 33 subjects using segment-level and subject-level cross-validation.

Main Results:

• Achieved 93.8% accuracy, 94.9% sensitivity, and 92.7% specificity at the segment level (5-fold CV).
• Attained 83.5% accuracy, 75.9% sensitivity, and 88.7% specificity at the subject level (LOSO-CV).
• Demonstrated the robustness of the algorithm in classifying normal and apnea events.

Conclusions:

• A novel and robust sleep apnea detection algorithm was successfully developed.
• The algorithm leverages ECG signal decomposition via EMD and deep CNNs.
• This approach offers a promising alternative for convenient and accurate SA diagnosis.