0331 Apnea Recognition in Wearable Device using the Intensive Evaluation of the ECG Power Spectral Density. (27th April 2018)

• Record Type: Journal Article
• Authors: Lin, Y; Wang, P; Lin, C; Sadrawi, M; Lin, C; Hsieh, Y; Kuo, C; Chien, J; Haraikawa, K; Abbod, M F; Shieh, J
• Is Part Of: Sleep. Volume 41(2018)Supplement 1
• Page Start: A127
• Page End: A127
• Publication Date: 2018-04-27
• Subjects: Sleep -- Physiological aspects -- Periodicals; Sleep disorders -- Periodicals; Sommeil -- Aspect physiologique -- Périodiques; Sommeil, Troubles du -- Périodiques; Sleep disorders; Sleep -- Physiological aspects; Sleep -- physiological aspects; Sleep Wake Disorders; Psychophysiology; Electronic journals; Periodicals; 616.8498
• DOI: 10.1093/sleep/zsy061.330
• Languages: English
• ISSNs: 0161-8105

Abstract

Abstract: Introduction: Apnea is one of critical parts for cardiovascular diseases and stroke. The evaluations of the pulse rate and heart rate variabilities are the main features in obstructive sleep apnea (OSA). The evaluations of power spectral density (PSD) and electrocardiography (ECG) have shown good correlation. The purpose of this study is to recognize normal and apnea conditions using several features, especially for the intensive evaluation of PSD, only from ECG signal with the utilizing of wearable device. In further, the apnea-hypopnea index (AHI) evaluation is also investigated. Methods: Initially, artificial neural network (ANN) is trained using Apnea-ECG database from PhysioNet to form a model. For testing the ANN model, dataset collected from 24 patients during whole sleep time was collected a single lead BC1 ECG device (Bio Clothing One, Kinpo Inc., Taipei, Taiwan). The apnea evaluation is calculated based on 30s with 1-minute sliding window. Different algorithms were used, such as fast Fourier transform (FFT)-based algorithm, sample entropy and detrended fluctuation analysis (DFA) for extracting features that are fed to ANN as the training data. The PSD features are intensively evaluated segmenting the bandwidth into 0.3 Hz segments to form totally 64 inputs. Results: The used evaluations are defined as: sensitivity (SE), positive predictive value (PPV), specificity (SP), accuracy (ACU) and AHI. Results of this study show that training of the PhysioNet Apnea-ECG data for SE, PPV, SP and ACU has 0.86, 0.83, 0.85 and 0.85, respectively. Meanwhile for testing, the PhysioNet Apnea-ECG test data has 0.90, 0.66, 0.73, and 0.79 respectively for SE, PPV, SP and ACU. Furthermore, the BC1 ECG data, simultaneously collected with the PSG, is used for the AHI evaluation. This study shows that the difference of the AHI between ANN and PSG less than 10 is 71.43%. Conclusion: The intensive evaluation of PSD algorithm with utilizing cross validation ANN has been applied to the wearable device that shows accurate evaluation results. Consequently, deep evaluation of PSD has contributed contributed positively to the apnea investigation. Support (If Any): Kinpo Electronics, Inc.