Pages: 1-12

Abstract

Abstract: Sleep apnea is very common in patients with heart failure and is considered a major cause of death. The main causes of this apnea are patients’ unawareness, lack of diagnosis, and ignoring the disease without considering any treatment. Currently, sleep apnea is being diagnosed primarily based on night Polysomnography. A complete recording, which is based on the apnea occurrence, is costly, cumbersome, and difficult to conduct. This study aims to provide an algorithm for the diagnosis of sleep apnea from electrocardiogram signals under the treatment of chemical warfare veterans. For this purpose, a study has been conducted with a combination of extracted features from changes in heart rate and signals of the electrocardiogram. Reducing the computational effort and the number of features as well as maintaining the high performance of the classifier are the subjects that are considered in this report. In other words, using ECG signal processing, especially HRV and EDR signal, the apnea is examined and to diagnose, the designed neural network in this study achieved a specificity of 98.94%, sensitivity of 77.21%, and accuracy of 98.38%. This test was conducted in Baqiyatallah Hospital and the mean absolute error in detecting AHI for 96 patients was 2.6. To evaluate the performance, the comprehensive Physionet database with the specificity of 99.73%, sensitivity of 87.43%, and accuracy of 92.95% has been used in the ANFIS model, and for further investigation of the AHI, patients were also studied in the formerly designed neural networks.