Sleep apnea screening by autoregressive models from a single ECG lead

Area of Science:

• Biomedical Engineering
• Cardiology
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) is a prevalent sleep disorder linked to cardiovascular system irregularities.
• Electrocardiogram (ECG) signals offer a noninvasive window into these cardiovascular changes.
• Analyzing ECG characteristics like RR intervals and QRS complex areas can aid in OSA detection.

Purpose of the Study:

• To develop and validate a method for screening obstructive sleep apnea (OSA) using electrocardiogram (ECG) recordings.
• To assess the efficacy of machine learning classifiers in identifying apneic events from ECG data.

Main Methods:

• Utilized a bivariate time-varying autoregressive model (TVAM) for beat-by-beat spectral analysis of ECG features (RR intervals, QRS complex areas).
• Employed K-nearest neighbor (KNN) and neural network (NN) supervised learning classifiers for minute-by-minute classification of apneic and non-apneic sleep.
• Applied sequential forward selection for optimal feature subset selection.

Main Results:

• KNN achieved 88% accuracy, 85% sensitivity, and 90% specificity with ten selected features.
• NN achieved 88% accuracy, 89% sensitivity, and 86% specificity.
• Both classifiers demonstrated 100% accuracy in distinguishing entirely normal from apneic recordings in initial and additional datasets.

Conclusions:

• ECG-based analysis provides a highly accurate and noninvasive method for obstructive sleep apnea screening.
• Machine learning models, particularly KNN and NN, are effective tools for classifying sleep apnea events from ECG data.
• The proposed method shows significant potential for widespread clinical application in OSA diagnosis.