A Novel Clinical Method for Detecting Obstructive Sleep Apnea using of Nonlinear Mapping

Due the long-time admission of patients in the ICU, it is very expensive. Therefore, solutions, which can increase the quality of care and decrease costs, can be helpful. Separation of the patients based on the acute conditions can be useful in providing appropriate therapy. In this study, we present a classifier to predict the OSA based on heart rate variability of patients. In this analytical study, we used the recorded ECG signals from PhysioNet Database. At first, in the preprocessing stage, the noise from the ECG signal was removed, and R spikes were detected to generate the HRV. The next stage was related to linear and non–linear features extraction. We used the paired sample t-test that is a statistical technique to compare two periods (apnea and non-apnea). These features were applied as the inputs of two different classifiers, including MLP and SVM to find the best method and distinguish patients with higher death risk. The results showed that the SVM classifier is more capable to separate the four periods seperated from each other. The sensitivity for detecting the OSA event was 95.46% and the specificity was 97.57% for the non-OSA period. Accurate and timely diagnosis of the disease can ensure the health of the individual, family, and community. Based on the proposed algorithm, the HRV signal and novel feature, presented in this study, had the highest specificity and sensitivity for the detection of the OSA event of the non-OSA, respectively.