Obstructive Sleep Apnea Detection using Frequency Analysis of Electrocardiographic RR Interval and Machine Learning...

Area of Science:

• Biomedical Engineering
• Cardiology
• Sleep Medicine

Background:

• Obstructive Sleep Apnea (OSA) is a common sleep disorder characterized by airway obstruction during sleep.
• Polysomnography (PSG) is the standard diagnostic tool but involves uncomfortable sensor application.
• There is a need for non-invasive and comfortable OSA detection methods.

Purpose of the Study:

• To propose and evaluate an Obstructive Sleep Apnea (OSA) detection method.
• Utilize Fast Fourier Transform (FFT) statistics of electrocardiographic RR Intervals (RRIs).
• Employ machine learning algorithms for OSA classification.

Main Methods:

• A case-control study using the Apnea ECG database from MIT-BIH.
• Extracted five FFT-based features from RRIs: mean, Shannon entropy, standard deviation, median, and geometric mean.
• Compared performance of Linear Discriminant Analysis (LDA), Artificial Neural Network (ANN), K-Nearest Neighbors (K-NN), and Support Vector Machine (SVM).

Main Results:

• Artificial Neural Network (ANN) demonstrated the highest performance in OSA detection.
• ANN with gradient descent backpropagation achieved 84.64% accuracy, 94.21% sensitivity, and 64.03% specificity.
• Linear Discriminant Analysis (LDA) showed the lowest performance among the tested algorithms.

Conclusions:

• ANN, particularly with gradient descent backpropagation, is a promising approach for OSA detection.
• This method offers a potentially more comfortable alternative to traditional PSG.
• Further research can optimize ANN parameters for improved OSA diagnosis.