羅捷 楊文瀟

開發(fā)設(shè)計(jì)

基于DTW距離的房顫信號(hào)提取方法

羅捷 楊文瀟

（廣東工業(yè)大學(xué)，廣東 廣州 510006）

在單導(dǎo)聯(lián)的房顫信號(hào)（f波）提取中，常用模板對(duì)消類方法易受QRST波群的形態(tài)變化以及點(diǎn)檢測(cè)偏差的影響。為克服上述問題，提出一種基于DTW距離的房顫信號(hào)提取方法。首先，通過高斯增強(qiáng)來提高R峰周圍心電在近鄰心拍計(jì)算中的權(quán)重，并基于DTW距離找到每個(gè)心拍對(duì)應(yīng)的近鄰心拍；然后，以各近鄰心拍為模板對(duì)目標(biāo)心拍進(jìn)行DTW重構(gòu)，并求出各近鄰心拍與對(duì)應(yīng)重構(gòu)目標(biāo)心拍的殘差；最后，通過平均殘差得到干凈的f波。對(duì)單導(dǎo)聯(lián)f波提取的仿真與實(shí)驗(yàn)結(jié)果表明：本文方法能夠更為準(zhǔn)確地提取f波，且對(duì)QRST波的形態(tài)變化以及點(diǎn)檢測(cè)偏差有較強(qiáng)的魯棒性。

房顫信號(hào)提?。籇TW距離；高斯增強(qiáng)；心拍重構(gòu)

0　引言

房顫是一種常見的心律失常疾病[1]。隨著年齡的增長(zhǎng)，房顫的發(fā)生率會(huì)不斷增加，且房顫患者中風(fēng)的概率比正常人高約五倍[2]。當(dāng)房顫發(fā)生時(shí)，心房激動(dòng)的頻率達(dá)每分鐘300~600次，正常心電的P波消失，取而代之的是連續(xù)、不規(guī)則顫動(dòng)的f波。f波的物理參數(shù)（如幅度和頻譜）常被用于房顫的臨床診斷，對(duì)房顫電生理活動(dòng)機(jī)理的研究有重要意義[3-6]。

心室QRST波與心房f波在時(shí)域和頻域都有部分混疊[7]，制約f波的檢測(cè)精度。為解決該問題，研究人員提出很多方法，可分為多導(dǎo)聯(lián)和單導(dǎo)聯(lián)2大類。多導(dǎo)聯(lián)方法[8-14]采用獨(dú)立分量分析、稀疏成分分析等技術(shù)，利用多通道心電信號(hào)之間的相關(guān)性分離QRST波和f波，雖然檢測(cè)精度較高，但需要布置多個(gè)電極，不便于移動(dòng)心電監(jiān)護(hù)設(shè)備使用。單導(dǎo)聯(lián)方法[15-19]采用平均模板消除、貝葉斯濾波等技術(shù)剔除QRST波或依據(jù)先驗(yàn)?zāi)Ｐ蛯?duì)f波進(jìn)行參數(shù)估計(jì)，從而實(shí)現(xiàn)f波的提取。單導(dǎo)聯(lián)方法雖然便于移動(dòng)心電監(jiān)護(hù)，但對(duì)異常心拍（如室性早搏導(dǎo)致的心拍形態(tài)變化）敏感，檢測(cè)精度較低。

為提高單導(dǎo)聯(lián)f波提取的精度，本文提出一種基于動(dòng)態(tài)時(shí)間規(guī)整[20]（dynamic time warping, DTW）距離的QRST波消除方法。首先，進(jìn)行R峰檢測(cè)，根據(jù)R峰的位置將心電信號(hào)切割成一個(gè)個(gè)單獨(dú)的心拍，并通過高斯增強(qiáng)來提高R峰周圍心電在近鄰心拍計(jì)算中的權(quán)重；然后，采用基于DTW距離的近鄰算法，找出與每個(gè)心拍最相似的個(gè)近鄰心拍，并計(jì)算各心拍與其近鄰心拍的DTW殘差；最后，平均這個(gè)DTW殘差求得各心拍的f波信號(hào)。

1　房顫心電模型

2　基于DTW距離的房顫信號(hào)提取

由于畸變心拍與f波信號(hào)的影響，R峰檢測(cè)算法并不能準(zhǔn)確地找到每個(gè)心拍真實(shí)的R峰。當(dāng)以R峰為基準(zhǔn)來分割心拍時(shí)，R峰采樣點(diǎn)的偏差將導(dǎo)致整個(gè)心拍時(shí)間采樣的偏差。為克服時(shí)間采樣偏差帶來的影響，需利用DTW算法對(duì)每個(gè)目標(biāo)心拍與其近鄰心拍進(jìn)行重構(gòu)。與其他模板法不同的是，本文沒有求出各心拍對(duì)應(yīng)的模板，而是先求出各心拍與其個(gè)近鄰心拍之間的DTW重構(gòu)殘差，再對(duì)這個(gè)重構(gòu)殘差進(jìn)行平均得到f波，算法流程如圖1所示。具體步驟如下：

1）采集患者心電信號(hào)1 min，采用4 Hz高通濾波器去除基線漂移及T波；

3）因?yàn)樵趩蝹€(gè)心拍內(nèi)心房f波能量的分布相對(duì)穩(wěn)定，而心室心電的波幅在R峰達(dá)到最大值，所以在R峰周圍，心室心電的信噪比較高；因此，本文通過高斯增強(qiáng)來提高R峰周圍心電在近鄰心拍計(jì)算中的權(quán)重：

6）為使各心拍f波在拼接邊緣不出現(xiàn)明顯的信號(hào)失真，本文采用crossfading方法[19]來處理信號(hào)邊緣的拼接問題，具體公式為

式中，fs為信號(hào)采樣率；L信號(hào)長(zhǎng)度；c為拼接邊緣的信號(hào)重疊長(zhǎng)度。

3　實(shí)驗(yàn)和結(jié)果分析

3.1 實(shí)驗(yàn)設(shè)置

為驗(yàn)證本文提出的基于DTW距離的房顫信號(hào)提取方法的有效性，將文中提出的基于DTW距離的模板法（GDTW-BS）及其不加高斯增強(qiáng)的方法（DTW-BS）分別在仿真房顫數(shù)據(jù)集和真實(shí)房顫數(shù)據(jù)集上進(jìn)行測(cè)試，并與平均模板法（ABS）、最優(yōu)模板對(duì)消法（WABS）和近鄰模板法（KNN-BS）進(jìn)行比較。

仿真房顫信號(hào)通過正常心電信號(hào)疊加仿真房顫信號(hào)生成，其中仿真f波采用STRIDH M等人提出的方法[12]來生成。采用歸一化均方誤差（normalized mean square error, NMSE）表示仿真所提取房顫信號(hào)的準(zhǔn)確度：

真實(shí)房顫信號(hào)取自PTB數(shù)據(jù)庫(kù)中房顫患者的心電數(shù)據(jù)。采用所提取f波信號(hào)的頻譜集中度（spectral concentration, SC）作為評(píng)價(jià)指標(biāo)：

房顫信號(hào)的中心頻率一般在3 Hz~10 Hz之間，提取房顫信號(hào)中的心室信號(hào)殘留成分一般在房顫信號(hào)的中心頻率之外。因此，越高，表明算法的提取效果越好。

3.2 實(shí)驗(yàn)結(jié)果和分析

表1和表2分別比較了ABS，WABS，KNN-BS，DTW-BS和GDTW-BS等方法在10條仿真房顫信號(hào)和真實(shí)房顫信號(hào)上的提取性能。圖2、圖3和圖4分別顯示了仿真f波提取效果、真實(shí)房顫信號(hào)、真實(shí)f波提取效果。

表1 仿真房顫信號(hào)提取結(jié)果

表2 真實(shí)房顫信號(hào)提取結(jié)果

圖2 仿真f波提取效果

圖3 真實(shí)房顫ECG

圖4 真實(shí)房顫ECG的f波提取效果

與ABS，WABS和KNN-BS相比，GDTW-BS在仿真f波提取性能指標(biāo)上分別提高了36.5%，28.6%和14.2%；在真實(shí)f波提取性能指標(biāo)上分別提高了19.8%，23.6%和18.8%。仿真與真實(shí)房顫信號(hào)提取實(shí)驗(yàn)的結(jié)果表明：本文提出的基于DTW距離的房顫信號(hào)提取方法能夠從心電信號(hào)中提取出更為干凈的f波，且魯棒性更高。

由于R峰在心電信號(hào)中能量最高，周圍的心電信號(hào)信噪比也較高。對(duì)R峰周圍的心電進(jìn)行高斯增強(qiáng)，一方面有利于找到更為準(zhǔn)確的近鄰心拍，另一方面可以提高DTW算法對(duì)R峰檢測(cè)偏差的糾正能力。相比于DTW-BS，GDTW-BS在仿真f波提取性能指標(biāo)上提高了7.5%；在真實(shí)f波提取性能指標(biāo)SC上提高了1.1%。實(shí)驗(yàn)結(jié)果表明：R峰增強(qiáng)能夠有效地提高算法的f波提取準(zhǔn)確度。

4　結(jié)語

針對(duì)f波提取算法的提取精度易受QRST波群變異性以及R峰定位誤差的影響，本文提出一種基于DTW距離的房顫信號(hào)提取方法。在仿真與真實(shí)f波的提取實(shí)驗(yàn)中，本文所提出的方法相較于ABS，WABS和KNN-BS等單導(dǎo)聯(lián)方法取得了更好的f波提取準(zhǔn)確度。

[1] CHUGH Sumeet S, ROTH Gregory A, GILLUM Richard F, et al. Global burden of atrial fibrillation in developed and developing nations[J]. Global Heart, 2014, 9(1):113-119.

[2] CHUGH Sumeet S, BLACKSHEAR Joseph L, SHEN Winkuang, et al. Epidemiology and natural history of atrial fibrillation: clinical implications[J]. Journal of the American College of Cardiology, 2001, 37(2):371-378.

[3] HUSSER D, STRIDH M, SORNMO L, et al. Time–frequency analysis of the surface electrocardiogram for monitoring antiarrhythmic drug effects in atrial fibrillation[J]. The American Journal of Cardiology, 2004, 95(4):526-528.

[4] RAYGOR VP, NG J, GOLDBERGER JJ. Surface ECG f wave analysis of dofetilide drug effect in the atrium[J]. J Cardiovasc Electrophysiol, 2015, 26(6):644-648.

[5] LANKVELD T, ZEEMERING S, SCHERR D, et al. Atrial fibrillation complexity parameters derived from surface ECGs predict procedural outcome and long-term follow-up of stepwise catheter ablation for atrial fibrillation[J]. Circulation: Arrhythmia and Electrophysiology, 2016, 9(2):e003354.

[6] ALCARAZ R, HORNERO F, RIETA J J. Electrocardiographic spectral features for long-term outcome prognosis of atrial fibrillation catheter ablation[J]. Annals of Biomedical Engineering, 2016, 44(11):3307-3318.

[7] ROONIZI E K, SASSI R. An extended bayesian framework for atrial and ventricular activity separation in atrial fibrillation[J]. IEEE Journal of Biomedical and Health Informatics, 2017, 21(6):1573-1580.

[8] RIETA J J, CASTELLS F, S′ANCHEZ C, et al. Atrial activity extraction for atrial fibrillation analysis using blind source separation[J]. IEEE Transactions on Biomedical Engineering, 2004, 51(7):1176-1186.

[9] CASTELLS F, RIETA J J, MILLET J, et al. Spatiotemporal blind source separation approach to atrial activity estimation in atrial tachyarrhythmias[J]. IEEE Transactions on Biomedical Engineering, 2005, 52(2):258-267.

[10] VAY′A C, RIETA J J, S′ANCHEZ C, et al. Convolutive blind source separation algorithms applied to the electrocardiogram of atrial fibrillation: Study of performance[J]. IEEE Transactions on Biomedical Engineering, 2007, 54(8):1530-1533.

[11] LLINARES R, IGUAL J. Application of constrained independent component analysis algorithms in electrocardiogram arrhythmias[J]. Artificial Intelligence in Medicine, 2009, 47(2):121-133.

[12] STRIDH M, SORNMO L. Spatiotemporal QRST cancellation techniques for analysis of atrial fibrillation[J]. IEEE Transactions on Biomedical Engineering, 2001, 48(1):105-111.

[13] LEMAY M, JACQUEMET V, FORCLAZ A, et al. Spatiotemporal qrst cancellation method using separate QRS and t-waves templates[C]. //Computers in Cardiology, IEEE, 2005:611-614.

[14] LEMAY M, VESIN J-M, VAN OOSTEROM A, et al. Cancellation of ventricular activity in the ECG: evaluation of novel and existing methods[J]. IEEE Transactions on Biomedical Engineering, 2007, 54(3):542-546.

[15] SLOCUM J, SAHAKIAN A, SWIRYN S. Diagnosis of atrial fibrillation from surface electrocardiograms based on computer-detected atrial activity[J]. Journal of Electrocardiology, 1992, 25(1):1-8.

[16] SHKUROVICH A, SAHAKIAN V, SWIRYN S. Detection of atrial activity from high-voltage leads of implantable ventricular defibrillators using a cancellation technique[J]. IEEE Transactions on Biomedical Engineering, 1998, 45(2):229-234.

[17] ALCARAZ R, RIETA J J. Adaptive singular value cancelation of ventricular activity in single-lead atrial fibrillation electrocardiograms[J]. Physiological Measurement, 2008, 29(12):1352.

[18] DAI H, JIANG S, LI Y. Atrial activity extraction from single lead ECG recordings: Evaluation of two novel methods[J]. Computers in Biology and Medicine, 2013, 43(3):176-183.

[19] MALIK J, REED N, WANG C-L, et al. Singlelead f-wave extraction using diffusion geometry[J]. Physiological Measurement, 2017, 38(7):1310.

[20] BERNDT D J, CLIFFORD J. Using dynamic time warping to find patterns in time series[J]. KDD Workshop, 1994, 10(16):359-370.

Extraction Method of Atrial Fibrillation Signal Based on DTW Distance

Luo Jie Yang Wenxiao

(Guangdong University of Technology, Guangzhou 510006, China)

In single lead atrial fibrillation (f-wave) extraction, the common template cancellation methods are easily affected by the morphological changes of QRST complex and R-point detection bias. In order to overcome the above problems, a method of extracting atrial fibrillation signal based on DTW distance is proposed. Firstly, Gaussian enhancement is used to improve the weight of ECG around R peak in the calculation of neighbor beats, and the nearest neighbor beats corresponding to each beat are found based on the DTW distance. Then, the DTW reconstruction of the target beat is performed with each neighbor beat as the template, and the residuals of each neighbor beat and the corresponding reconstructed target beat are calculated. Finally, clean f-wave is obtained by average residual. The simulation and experimental results of single lead f-wave extraction show that the proposed method can extract f-wave more accurately, and has strong robustness to the morphological changes of QRST waves and R-point detection bias.

atrial fibrillation signal extraction; DTW distance; beat reconstruction

TN911.7

A

1674-2605(2020)06-0006-05

10.3969/j.issn.1674-2605.2020.06.006

羅捷，男，1995年生，碩士研究生，主要研究方向：模式識(shí)別，機(jī)器學(xué)習(xí)，生物信號(hào)處理。E-mail:joyluo@mail2.gdut.edu.com