An Experimental Review on Obstructive Sleep Apnea Detection Based on Heart Rate Variability and Machine Learning Techniques

Obstructive sleep apnea (OSA) is a respiratory syndrome of high incidence in the general population and correlated with some cardiovascular diseases. Several techniques have been proposed in the last decades to find a surrogate method to polysomnography (PSG), the gold standard for the diagnosis of OSA. The present study comprises an experimental review on the state-of-the-art methods for OSA detection through the public Apnea-ECG database, which is available at PhysioNet. Precisely, traditional time-frequency domain features were extracted from the heart rate variability (HRV) signal, together with some common complexity measures. Given their ability to deal with real-world time series, two additional entropy-based measures were also tested, i.e., Renyi and Tsallis entropies. Moreover, univariate and multivariate classifiers were applied, including diagnostic test, support vectors machine, and k-nearest neighbors. Ultimately, two sequential feature selection (SFS) algorithms were employed to reduce the computational cost of the resulting discriminant models. The major findings reported that multivariate classifiers reached similar results to those found in the literature. Moreover, univariate classification results suggested that the frequency domain features provided the best OSA detection, although a well-known entropy index also obtained a good performance.