Journal Article

Automatic detection of obstructive sleep apnea through nonlinear dynamics of single-lead ECG signals

Chen, Liangjie, Wang, Qinghui, Zeng, Wei, Yuan, Chengzhi, Wang, Ying, Liu, Fenglin

2025

Overview

Obstructive Sleep Apnea (OSA) is a sleep disorder where the brain and body receive insufficient oxygen during sleep. Traditional diagnosis involves Polysomnography (PSG), which is time-consuming, tedious, subjective, and costly in clinical settings. To address these drawbacks, computer-assisted diagnosis techniques have emerged, utilizing a single physiological signal. This study aims to introduce an innovative method for automatically detecting OSA based on the dynamics of the ECG system. The approach combines tunable quality factor (Q-factor) wavelet transform (TQWT), variational mode decomposition (VMD), and three-dimensional (3D) phase space for feature extraction, capturing clinically relevant information from OSA ECG recordings. Neural networks are employed to model and identify ECG system dynamics via deterministic learning theory, classifying normal and OSA ECG signals based on differences in dynamics using a bank of dynamical estimators. An assessment is conducted utilizing a 10-fold cross-validation methodology on a PhysioNet apnea-ECG dataset, which comprises 70 nocturnal recordings derived from an equal number of subjects. The empirical outcomes demonstrate that the introduced approach, which amalgamates a classifier based on neural network principles and the recommended attributes, attains superior accuracy (98.27 % ), sensitivity (97.68 % ), and specificity (98.63 % ) in contrast to conventional PSG. The results corroborate the suggested technique as a viable substitute for automatic OSA detection in a clinical setting.

• Publisher: Springer US
• Subject: Artificial Intelligence / Computer Science / Machines / Manufacturing / Mechanical Engineering / Processes