A fast approximation method for principal component analysis applied to ECG derived respiration for OSA detection

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Informatics

Background:

• Obstructive sleep apnea (OSA) diagnosis often relies on polysomnography.
• Electrocardiogram (ECG) signals offer a potential non-invasive alternative for respiration monitoring.
• Principal Component Analysis (PCA) is a technique that can be used to derive respiration signals from ECG.

Purpose of the Study:

• To present and evaluate an approximation method for PCA for respiration estimation from overnight ECG.
• To compare the performance of the approximated PCA method against a full PCA method for OSA detection.
• To assess the computational efficiency and memory requirements of the approximated PCA method.

Main Methods:

• An approximation method for PCA was developed and applied to overnight ECG signals.
• ECG-derived respiration (EDR) signals were generated using both approximated and segmented full PCA.
• Extreme Learning Machine (ELM) and Linear Discriminant Analysis (LDA) classifiers were trained to detect OSA using EDR features.
• Leave-one-record-out cross-validation was performed on the MIT PhysioNet Apnea-ECG database (35 recordings).

Main Results:

• The approximated PCA method achieved the highest accuracy of 78.4% with LDA and 76.4% with ELM.
• The segmented full PCA method yielded lower accuracies (76.6% with LDA, 75.9% with ELM).
• The approximated PCA method demonstrated computational speed and low memory usage.

Conclusions:

• The approximated PCA method is a computationally efficient and effective approach for deriving respiration signals from overnight ECG.
• This method shows promise for accurate obstructive sleep apnea detection using ECG data.
• The findings suggest that approximated PCA can be a viable alternative to traditional methods for sleep apnea screening.