摘要
重症监护室(Intensive Care Unit,ICU)病人的生命体征信号如心电图(Electrocardiogram,ECG)、动脉血压(Arterial Blood Pressure,ABP)等是医师对病人进行诊断、治疗的重要参考指标。然而,由于这些生命体征信号经常受到诸如噪声、伪迹和数据缺失等的干扰,造成监护信号参数估计出现错误,导致ICU监护仪的错误报警率居高不下,从而使医生对监护仪报警信号的信任缺失,忽略真正的危急情报,大大削弱了监护效果。目前监护仪大多依赖对心电信号的分析估计心率,心律失常的诊断又与心率密切相关,结果受干扰的影响很大。本文研究了同时对心电和动脉血压两源信号进行数据融合,抑制监护仪心律失常的错误报警,提高监护系统的报警准确率和灵敏度。
首先我们对心电信号和动脉血压信号分别经QRS波形识别算法和血压搏动识别算法获得逐搏心率值,其次应用信号质量评估算法获得ECG和ABP的信号质量指数(Signal Quality Index,SQI)。信号质量指数可以提供判断信号质量好坏的客观指标。然后应用卡尔曼滤波方法对逐搏心率进行心率估计,并通过信号质量指数对卡尔曼滤波增益系数进行动态调整,当信号质量低时,可控制停止卡尔曼滤波更新,使心率估计保持上一个高质量的心率值不变,从而避免严重干扰对心率估值的影响。最后用卡尔曼滤波的残差和信号质量指数作为权重系数,对心电和血压信号进行心率数据融合,计算融合心率。
我们研究了基于融合心率和信号质量指数抑制监护仪严重心动过缓和严重心动过速错误报警的算法。首先对ECG和ABP数据进行融合心率估计,并根据信号质量指数的高低判断融合心率是否可信,进行错误报警的抑制。当计算得到的融合心率未超过监护仪设定的心率报警阈值时,如果此时ECG或ABP中至少一导联信号的SQI＞0.5时,我们信任此时的心率并未超界,认为监护仪报警是误报警并加以抑制,否则接受该报警。以美国麻省理工学院健康科学与技术中心(MIT/HST)建立的MIMICⅡ重症病人多参数智能监护数据库为评价数据库,应用本算法对其中记录的监护仪产生的2432个严重心动过速和严重心动过缓报警对应的ECG和ABP信号进行重新分析,将该抑制算法的结果与专家作出的报警注释逐一对比,得出本算法对真实报警正确识别率为99.64%,对错误报警的抑制率为66.18%。
本文研究的基于ECG和ABP两通道信号数据融合算法可扩展应用于多通道数据融合,对ICU病人的生命体征信号进行融合估计。下一步我们计划在基于心率估计的监护仪错误报警抑制算法的基础上,进一步结合心电及血压波形形态学分析对室性心律失常报警信号进行研究,进一步改善监护仪的报警质量。
Physiological vital signals of intensive care unit (ICU) patients such as electrocardiogram(ECG) and arterial blood pressure (ABP) are the important indicators to physician's diagnosis and treatment for the patients. Nevertheless, these signs are often severely corrupted by noise, artifact and missing data, which lead to large errors in the estimation of signal parameters, keeping a high incidence of false alarms from ICU monitors, resulting in lack of trust of clinical staff to monitor alarming signals, neglecting the real critical information, greatly diminishing the effect of care. Currently heart rate estimation relies mostly on ECG analysis, the diagnosis in arrhythmias is closely related with heart rate, so the results are greatly influenced by the interference. This paper studies ECG and ABP at the same time and gives a comprehensive analysis and intelligent diagnosis of signals from two sources, suppressing false alarms of arrhythmia and increasing accuracy and sensitivity of the alarm monitoring system.
Firstly, we acquired heart rate values respectively from ECG by using QRS signal waveform recognition algorithm, and from ABP by using pulse and blood pressure identification algorithm, followed by applying signal quality assessment algorithm to get signal quality index (SQI)of ECG and ABP. Signal quality index can provide the objective indicators in judging the signal's quality. Then, Kalman filters were applied in heart rate estimation one by one, and we adjusted Kalman filter gain coefficient adaptively through signal quality index. When the signal quality was low, we did not update the Kalman filter, keeping the heart rate estimation as the previous value, so as to avoid the serious effects of noise on heart rate estimation. Finally, the residual error of Kalman filter and signal quality index were applied as a weighting coefficient in data fusion of ECG and ABP, to calculate the fusing heart rate.
We studied the algorithm that integrated fusing heart rate and signal quality index to inhibit serious bradycardia and tachycardia false alarms created by monitors. Firstly ECG and ABP data were used in heart rate integration estimations, and we judged the credibility of integrated HR according to the level of signal quality index in order to suppress false alarms. When the integrated heart rate we calculated did not exceed the alarming thresholds of heart rate set in monitor, and if the SQI of at least one signal (either ECG or ABP) was above 0. 5, we trusted the heart rate at this time and considered that it did not exceed the threshold and the alarm was false and to be suppressed, or else accepted the alarm. To evaluate our algorithm, we used the Multi-Parameter Intelligent Monitoring for Intensive Care (MIMIC) II database established by the Health Sciences and Technology of Massachusetts Institute of Technology Centre (MIT/HST). The ECG and ABP data recorded in the database according to 2432 severe tachycardia and bradycardia alarms produced by the monitor were reanalyzed using this suppression alarm algorithm. The result was compared with the expert group report notes one by one. The results show that the correct rate identifying true alarms using the algorithm is 99.64%, and the rate of depressing false alarms is 66.18%.
Data fusion algorithm based on two-channel ECG and ABP signals can be extended to multi-channel data fusion, and to evaluate synthetically the physiological parameters of ICU patients. Next, we plan to improve the algorithm by combining ECG and blood pressure waveform morphology to analyze alarming signals of ventricular arrhythmia, so as to improve the quality of intensive care.
引文
[1]萧正伦.重症监护病房的进展与危重症监护医学[J].新医学,2000,31(5):305-306.
[2]Michael Imhoff,Silvia Kuhls.Alarm algorithms in critical care monitoring[J].AnesthAnalg,2006,102:1525-1537
[3]Tsien C L,Fackler J C.Poor prognosis for existing monitors in the intensive care unit[J].Crit Care Med,1997,25(4):614-619.
[4]姜淑霞,赵秀娥.现代监护仪的工作原理和使用特点[J].医疗装备,200(),13(10)39-40.
[5]陈文彬,潘祥林主编.诊断学(第六版)[M].北京:人民卫生出版社,2004,2.
[6]Zong W,Moody G B,Jiang D.A robust open-source algorithm to detect onset and duration of QRS complexes[J].Comput in Cardiol,2003,30:737-740.
[7]傅娟,宁新宝,李德华.小波分析应用于对心电图信号实时处理的研究[J].南京大学学报(自然科学),2004,40(2):245-250.
[8]范宝林,彭屹.小波变换用于心电信号消噪及QRS波检测的研究进展[J].北京生物医学工程,2005,24(2):147-150.
[9]赵治栋等.小波阈值去噪技术研究及其在生物医学信号处理中的应用[J].中国医疗器械杂志,2004,28(4):78-80.
[10]Ercelebi,E.Electrocardiogram signals de-noising using lifting-based discrete wavelet transform[J].Comput Biol Med.,2004,34(6):479-93.
[11]Kalman,R.E.A New Approach to Linear Filtering and Prediction Problems[J].Journal of Basic Engineering,1960,82(Series D):35-45.
[12]李群力.基于 Kalman滤波的数据融合研究[J].计算机仿真,2007,24(8):299-302.
[13]韩清鹏,王平.基于卡尔曼滤波技术的心电信号在线检测[J].生物医学工程学杂志,2007,24(1):215-218.
[14]Sittig,D F,Factor,M.Physiologic trend detection and artifact rejection:a parallel implementation of a multi-state Kalman filtering algorithm[J].Computer Methods Programs Biomed,1990,31(1):1-10.
[15]赵浩,周卫东,钟凌惠.独立分量分析在生物医学信号处理中的应用[J].生物医学工程研究,2003,22(4):56-60.
[16]赵治栋等.ICA及其在消除心电信号中肌电伪迹的应用.浙江大学学报(工学版)[J],2003,38(1):8-10.
[17]苏永春等.基于独立分量分析的脉搏波信号的降噪处理[J].医疗设备信息,2006,21(6):4-7.
[18]He T,Clifford GD,Iarassenko L.Application of independent component analysis in removing artefacts from the electrocardiogram[J].Neural Comput & Applic,2006,15(2):105-16.
[19]杨冬等.基于BP神经网络的非接触生命参数信号检测[J].第四军医大学学报,2004,25(14):1249-1251.
[20]张泾周,梁佳明.基于自适应线性神经网络的心电信号滤波[J].陕西科技大学学报,2005,23(3):48-52.
[21]王进军,金登男.应用神经网络的理论对动态心电信号模式分类的研究[J].计算机应用与软件,2006,23,(3):83-85.
[22]Tsien,C L,Kohane,I S,Mclntosh,N.Multiple signal integration by decision tree induction to detect artifacts in the neonatal intensive care unit[J].Artificial Intelligence in Medicine,2000,19:189-202.
[23]王耀南著.智能控制系统—模糊逻辑.专家系统.神经网[M].长沙:湖南大学出版社.1996,10.
[24]Zong,W.,Moody,G.B.,Mark.R.G.Reduction of false arterial blood pressure alarms using signal quality assessment and relationships between the electrocardiogram and arterial blood pressure[J].Med Biol Eng Comput,2004,42:698-706.
[25]Moody,G.B.,Mark,R.G.QRS Morphology Representation and Noise Estimation using the Karhunen-Loeve Transform[J].Computers in Cardiology,1989,16:269-272.
[26]Cao,C.,Kohane,I.S.,McIntosh,N.Artifact detection in cardiovascular time series monitoring data from preterm infants[J].Proc AMIA Symp,1999:207-211.
[27]Makivirta,A,Koski,E,Kari,Aet al.The median filter as a preprocessor for a patient monitor limit alarm system in intensive care[J].Comput Methods Programs Biomed,1991,34(2-3):139-144.
[28]Hamilton P S,Tompkins W J.Quantitative investigation of QRS detection rules using the MIT/BIH arrhythmia database[J].IEEE Trans on BME,1986,33(12):1157-1165.
[29]Zong W.,Moody G B,Mark R G.An open-source algorithm to detect onset of arterial blood pressure pulses[J].Comput.in Cardiol.,2003,30:259-262.
[30]李桥,俞梦孙.重症监护病人心电导联信号质量评估[J].山东大学学报(医学版),2007,45(9):868-872,890.
[31]Zong W,Moody G B,Mark R G.Reduction of false arterial blood pressure alarms using signal quality assessment and relationships between the electrocardiogram and arterial blood pressure[J].Med.Biol.Eng.Comput,2004,42(5):698-706.
[32]吴艳.多传感器数据融合算法研究[D].西安电子科技大学博士学位论文,2003.
[33]李桥,Mark R G,Clifford G D等.基于信号质量评估和卡尔曼滤波的心率估计算法[J].中国医学物理学杂志,2007,24(6):454-457.
[34]Saeed M,Lieu C,Raber G,et al.MIMIC Ⅱ:a massive temporal ICU patient database to support research in intelligent patient monitoring[J]. Comput in Cardiol, 2002,29:641-644.
[35]Feldman,J.M., Ebrahim,M. Robust sensor fusion improve heart rate estimation: clinical evaluation[J]. Journal of clinical monitoring, 1997, 13(6):379-384
[36]Feldman, J. M. , Ebrahim,M. Bar-Kana,Izhak. A robust sensor fusion method for heart rate estimation[J]. Journal of clinical monitoring, 1997,13(6):386-393
[37]Elena MM, Quero JM, Borrego I. An Optimal Technique for ECG Noise Reduction in Real Time Applications [J]. Comput in Cardiol , 2006, 33:225 - 228.
[38] Friesen GM, Jannett TC, Jadallah MA, et al. A comparison of the noise sensitivity of nine QRS detection algorithms[J]. IEEE Trans on Biomed Eng, 1990, 37 (8) :85 - 98.
[39] Tarassenko L, Mason L and Townsend N. Multi-sensor fusion for robust computation of breathing rate[J]. Electronics Letters, 2002,38(22) : 1314-16.