Sleep Apnea Classification Algorithm Development Using a Machine-Learning Framework and Bag-of-Features Derived from...

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Heart rate variability (HRV) and ECG-derived respiration (EDR) are established methods for sleep apnea (SA) detection.
• Existing methods have limitations that newer approaches aim to overcome.

Purpose of the Study:

• To develop and validate a novel SA detection algorithm.
• To leverage machine learning and ECG spectrograms for improved SA diagnosis.

Main Methods:

• Utilized overnight ECG recordings from 83 subjects.
• Applied signal preprocessing, continuous wavelet transform (CWT) for time-frequency analysis, and bag-of-features (BoF) generation.
• Employed machine learning classifiers (SVM, EL, KNN) with cross-validation.

Main Results:

• Achieved cross-validation accuracies of 90.5% (10s window) and 91.4% (60s window).
• Identified specific frequency bands (0.1-50 Hz and 8-50 Hz) yielding optimal results.

Conclusions:

• Successfully developed an SA detection algorithm using BoF and machine learning.
• The algorithm demonstrates satisfactory classification accuracy and high temporal resolution for SA detection.