Sleep Apnea Detection from Single-Lead ECG Using Features Based on ECG-Derived Respiration (EDR) Signals

Background and objective: One of the important applications of non-invasive respiration monitoring using ECG signal is the detection of obstructive sleep apnea (OSA). ECG-derived respiratory (EDR) signals, contribute to useful information about apnea occurrence. In this paper, two EDR extraction methods are proposed, and their application in automatic OSA detection using single-lead ECG is investigated. Methods: EDR signals are extracted based on new respiration-related features in ECG beats morphology, such as ECG variance (EDRVar) and phase space reconstruction area (EDRPSR). After evaluating the EDRs by comparing them to a reference respiratory signal, they are used in an automatic OSA detection application. Fantasia and Apnea-ECG database from PhysioNet are used for EDRs assessments and OSA detection, respectively. The final performance of our OSA detection is tested on an independent test data which is also compared with results of other techniques in the literature. Results: The extracted EDRs, EDRVar and EDRPSR show correlations of 72% and 70% with reference respiration, which outperform the other state-of-the-art EDR methods. After feature extraction from EDRs and RR intervals series, the combination of RR and EDRPSR feature sets achieved 100% accuracy in subject-based apnea detection on independent test data, and also minute-based apnea detection is done with accuracy, sensitivity and specificity of 90.9%, 89.6% and 91.8%, which is better than other automatic algorithms in the literature. Conclusions: Our OSA detection system using EDRs features yields better independent test results compared with other state-of-the-art automatic apnea detection methods. The results indicate that ECG-based OSA detection system can classify OSA events with high accuracy and suggest a promising, non-invasive and efficient method for apnea detection. (C) 2018 AGBM. Published by Elsevier Masson SAS. All rights reserved.