光电容积脉搏波的睡眠呼吸暂停综合征筛查方法

doi:10.3964/j.issn.1000-0593(2019)06-1852-06

Abstract
References
Related Citation (15)

Download: PDF (2338 KB)
Export: BibTeX | EndNote (RIS)

Abstract Sleep Apnea Syndrome (SAS) is known as the “sleep killer”. The diagnostic rate is low due to the limitations of the Polysomnography (PSG) diagnostic criteria. Studies have shown that heart rate rhythm changes when apnea occurs, so automatic screening of SAS can be achieved by measuring Electrocardiograph (ECG) signals based on Heart Rate Variability (HRV) analysis. However, the electrodes used in the ECG-SAS method are cumbersome, can easily cause skin allergy, and affect sleeping comfort. Due to Pulse Rate Variability (PRV) analysis being highly correlated with HRV analysis and photoplethysmography (PPG) signals being simpler to acquire than ECG signals, this study proposes using synchronously acquired PPG and ECG signals and applying the same modeling method to compare the recognition ability of the two methods. The benefits for acquiring PPG instead of ECG are that the electrode does not cause skin allergyand is easier to wear so that it has little interference with sleeping comfort. The Back-Propagation (BP) neural network is applied to establish the automatic screening models of PPG-SAS and ECG-SAS, respectively. The 10-fold cross validation method and the Receiver Operating Characteristic (ROC) curves are used to compare and to evaluate the models. The experimental data are from MIT-BIH Polysomnographic Data base that contains 8 248 samples, including 6 227 normal samples. First of all, we established PPG-SAS and ECG-SAS models using a three-layer BP neural network with the default parameters, and compare their classification performances through the 10-fold cross validation method and the ROC curves. And then, we successively adjusted the number of hidden layer nodes, training functions and transfer functions to establish corresponding PPG-SAS and ECG-SAS models, and compare the respective optimal models obtained by using the 10-fold cross validation method. Through the comparisons of the recognition and prediction accuracies and the area of the ROC curves, the results illustrate that the PPG-SAS model is better than the ECG-SAS model when default parameters were applied. By comparing the average classification performances, we obtained the optimal model of PPG-SAS with 50 hidden layer nodes, trained function based on one-step secant method, and transferred function based on hyperbolic tangent sigmoid. The optimal PPG-SAS model has the highest recognition accuracy of 80.30% and prediction accuracy of 80.13%. Similarly but with a different transfer function of radial basis, the optimalECG-SAS model has the highest recognition accuracy of 77.60% and prediction accuracy of 77.67%. The results showed that the optimal PPG-SAS model is better than the optimal ECG-SAS model. The above experimental results demonstrated that the SAS classification ability by using PPG signals is superior to that by using ECG signals, which proved the feasibility and reliability of the PPG-SAS screening method. Therefore, the PPG-SAS screening method will lay a theoretical foundation on early detection of SAS and improvement of its diagnostic rate.

Key words：Sleep apnea syndrome; PPG; ECG; Neural networks; 10-fold cross-validation

Received: 2018-05-16 Accepted: 2018-10-28

ZTFLH:

TP391

Corresponding Authors: NI Wei-guang E-mail: niwg@jlu.edu.cn

Cite this article:

URL:

https://www.gpxygpfx.com/EN/10.3964/j.issn.1000-0593(2019)06-1852-06 OR https://www.gpxygpfx.com/EN/Y2019/V39/I06/1852

[1] Chang C C, Chuang H C, Lin C L, et al. Sleep Medicine, 2014, 15(4): 410.
[2] Yaggi H Klar, John Concato, Walter N Kernan, et al. The New England journal of medicine, 2005, 353(19): 2034.
[3] Selim B J, Junna M R, Morgenthaler T I. Sleep Medicine Clinics, 2014, 9(1): 37.
[4] Hang L W, Lin H H, Chiang Y W, et al. Applied Mathematics & Information Sciences, 2013, 7(1): 227.
[5] Chazal P D, Heneghan C, Sheridan E, et al. Computers in Cardiology, 2000, 27: 745.
[6] Mietus J E, Peng C K, ChIvanov P, et al. Computers in Cardiology, 2000, 27: 753.
[7] Zywietz C W, Von E V, Widiger B, et al. Methods of Information in Medicine, 2004, 43(1): 56.
[8] Chazal P D, Heneghan C, Sheridan E, et al. IEEE Transactions on Biomedical Engineering, 2003, 50(6): 686.
[9] TONG Guang-ming, GUO Ji-hong, HAN Fang, et al(佟光明, 郭继鸿, 韩芳, 等). National Medical Journal of China(中华医学杂志), 2006, 86(122): 1545.
[10] Acharya U R, Chua E C, Faust O, et al. Physiological Measurement, 2011, 32(3): 287.
[11] Khaldon Lweesy, Luay Fraiwa, Natheer Khasawneh, et al. Journal of Medical Systems, 2011, 35: 723.
[12] Kesper K, Canisius S, Penzel T, et al. Medical & Biological Engineering & Computing, 2012, 50(2): 135.
[13] Poupard Laurent, Philippe Carole, Goldman Michael David, et al. Sleep and Breathing, 2012, 16(2): 419.
[14] Ahmet Resit Kavsaoglu, Kemal Polat, Mehmet Recep Bozkurt. Turkish Journal of Electrical Engineering & Computer Sciences, 2016, 24: 1782.
[15] Schäfer A, Vagedes J. International Journal of Cardiology, 2013, 166(1): 15.
[16] Bulte C S E, Keet S W M, Boer C, et al. European Journal of Anaesthesiology, 2013, 68(7): 775.
[17] Poh M Z, Swenson N C, Picard R W. IEEE Transactions Information Technology in Biomedicine, 2010, 14(3): 786.
[18] Li Y, Gao H, Ma Y. Medicine, 2017, 96(18): e6755.
[19] LI Jun-feng, WANG Yue-le, HU Sheng, et al(李俊峰, 汪月乐, 胡升, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2016, 36(10): 3261.