摘要
现代加强医疗病房应用了大量的监护设备对危重病人的生理和病理状态进行监测,然而由于监护信号经常受到噪声、伪迹和数据缺失等的干扰,造成监护参数的估值错误,导致监护仪错误报警。有报道表明监护仪报警假阳性率有时竟高达86%。居高不下的错误报警率不但剥夺了病人的正常休息,增加了病人和医护人员的压力,造成时间、资源上的浪费,更为严重的是造成医护人员对报警信号的不信任和麻痹大意,这严重影响了临床医生对真实危重报警的及时响应和处理,大大降低了重症监护效果。
本文从危重病人生命体征信号的质量评估入手,导出了反映信号质量高低的信号质量指数(Signal Quality Index,SQI),研究了基于SQI的在严重干扰存在时生命体征信号的估计,实现了对严重心律失常错误报警的抑制。
本文首先研究了心电(Electrocardiogram,ECG)与动脉血压(ArterialBlood Pressure,ABP)信号的信号质量评估方法,提出了通过比较对干扰敏感性不同的QRS识别算法结果来确定心电信号质量的思想,并综合多导联同步分析与比较、信号峰度分析和信号功率谱分析等方法导出了心电信号质量指数;应用模糊逻辑算法和启发式约束算法对血压信号进行波形特征分析,导出了血压信号质量指数。信号质量指数是一个介于0和1之间的数值,数值越大表明信号质量越好,数值越小表明信号质量越差。
导出SQI之后,研究了在严重干扰存在时生命体征信号的估计方法,主要研究了心率和动脉血压值的估计。直接从ECG或ABP信号的搏动识别得到的心率值和直接从ABP波形特征分析得到的血压值很容易受到干扰的影响,造成估值的严重误差。卡尔曼滤波器可以避免干扰对信号估计的影响,提供对随机信号的最佳估计,适用于对心率、血压等心血管系统信号的估计。我们研究了应用卡尔曼滤波器实现心率和血压估计的算法,提出了通过信号质量指数调节卡尔曼滤波器增益系数的思想,当信号质量高时,使卡尔曼滤波器残差的增益系数增加,此时算法较多地依赖高质量的心率(或血压)测量值来调整心率(或血压)估计;反之,当信号质量低时,使滤波器残差的增益系数减小,算法较少地依赖低质量的测量值来调整参数估计。同时设定了一个信号质量指数阈值,当信号质量低于阈值时,停止更新卡尔曼滤波器,以避免信号质量过低时干扰对参数估计的严重影响,从而对卡尔曼滤波估计进行了优化。
心率信息可以通过对ECG、ABP及脉搏血氧饱和度等信号进行搏动检测获得,由于各通道间数据冗余且近似独立,同时,各通道间的干扰往往不具有相关性,通过传感器融合算法对多源信号进行数据融合可获得更精确的心率估计。我们首先研究了从多个包含干扰的信号估计中获取最佳估计的数据融合算法,导出了多导数据融合公式。然后对基于多导心电和动脉血压信号的卡尔曼滤波算法获得的心率估计进行数据融合,得出了鲁棒性的融合心率估计;提出了用信号质量指数和卡尔曼滤波器残差作为权重系数控制数据融合的思想,当某导联由于干扰造成心率估计误差时,一方面由于干扰的存在使该导联的SQI减小,另一方面心率估计误差的存在使卡尔曼滤波器的残差增大,从而在数据融合时控制减小该导联的权重。数据融合方法在仅有一个通道具有高质量信号的情况下,仍能获得精确的参数估计。
通过数据融合算法获得心率估计后,研究了应用融合心率和信号质量指数抑制监护仪严重心律失常错误报警的算法,提出了通过信号质量指数判断阈值的选取进行真实报警识别率和错误报警抑制率调节的思想。当监护仪发出严重心律失常报警时,用此时算法估计的融合心率与监护仪设定的报警阈值进行比较,并根据信号质量是否可信判断接受或拒绝报警。当融合心率未达到报警阈值且信号质量指数指示信号质量好时,则信任此时的心率估计,认为此时监护仪的报警是错误报警并加以抑制。
为对本文提出的算法进行验证和评价,应用美国麻省理工学院多参数智能重症监护数据库Ⅱ(Multi-Parameter Intelligent Monitoring forIntensive CareⅡ,MIMICⅡ)中的高质量连续监护数据作为标准信号集,共选取437例、3762段长度为1小时以上(1.62±0.69 h),总计6084小时的连续高质量的同步多导联ECG和ABP波形数据,对标准信号集添加不同类型和强度的心电和动脉血压干扰,建立了心电和血压评价数据集。心电干扰采用了麻省理工学院标准心电干扰数据集的干扰数据,包括三类干扰和6种信噪比的18类干扰数据,动脉血压干扰应用我们自行设计的六类动脉血压干扰生成算法生成的干扰数据,包括六类干扰和5种不同干扰段占据比例(占空比)的30类干扰数据。同时基于MIMICⅡ数据库中的监护仪报警数据建立了错误报警抑制评价数据集,包括各类严重心律失常报警信号5344次,经专家确定,其中的错误报警率为42.74%。
为进行心率估计算法性能评价,研究了在心率估计抗干扰研究中采用的确定心率金标准的算法,以标准信号集为基准计算得出了心率的金标准值,用于比较不同心率估计算法的抗干扰效果。我们对八种心率估计算法和三种动脉血压估计算法进行了抗干扰效果评价。
研究结果表明,评价数据集的ECG和ABP信号的信号质量指数均随添加干扰的增加而减小,信号质量指数较好地反映了生理信号受干扰的水平;信号质量指数在应用卡尔曼滤波算法进行心率和血压估计中发挥了重要作用,基于卡尔曼滤波和信号质量指数的估计算法可获得较好的心率和血压估计,如不采用信号质量指数对卡尔曼滤波器参数进行控制,当信号质量较低时估计误差显著增加;数据融合算法可有效克服干扰造成的心率估计误差,获得较好的融合心率估计,在严重ECG干扰存在时心率估计误差小于1次/分,在严重ABP干扰存在时心率估计误差小于2.1次/分;基于SQI的动脉血压估计算法可以不同程度地减小血压估计误差,当SQI对干扰敏感时,可显著降低干扰对血压估计的影响,当SQI不敏感时,干扰影响仍不同程度的存在;错误报警抑制算法对严重心律失常报警的真实报警正确识别率为100.00%,对错误报警的抑制率为53.68%,错误报警率从抑制前的42.74%降低为19.80%。
正确的生命体征信号估计和信号质量评估是智能监护的前提,只有获取了正确的监护参数和可信的信号,才能进行下一步对病人生理和病理状态的正确解释、预测和评估。本文在这方面进行了有益的尝试,但仍有许多不足之处:如获取的血压信号SQI对某些类型的干扰仍不很敏感,只对生命体征信号中的ECG和ABP进行了研究,只研究了心律失常错误报警的抑制,对其他生命体征信号和其他类型的报警尚需进行进一步研究。
Modern intensive care units (ICU) employ an impressive array of technologically sophisticated instruments to provide detailed measurements of the physiological and pathophysiologic state of each patient. But the physiological signals in the ICU are often severely corrupted by noise, artifact and missing data, which lead to large errors in the estimation of the signals values. This can result in a high incidence of false alarms from ICU monitors, which can sometimes be as high as 86% for some alarm types. Frequent false alarms due to data corruption will lead not only to sleep deprivation for patients and stress for patients and staff, but also to wasted time, resources, and to a desensitization of clinical staff to real alarms and a consequent drop in the overall level of care.
This paper started with the signal quality assessment of vital signs in intensive care patients, derived the signal quality index (SQI) to reveal the degree of signal quality. Based on the SQI, the vital signs were estimated in the presence of high levels of noise and artifact. And then the arrhythmia false alarm reduction algorithm in ICU monitors was accomplished.
The methods of signal quality assessment for electrocardiogram (ECG) and arterial blood pressure (ABP) were studied. Since different ECG beat (QRS) detection algorithms are sensitive to different types of noise, a novel idea was presented that the signal quality can be reflected by comparing the accuracy of different QRS detectors. The SQI of ECG signals was obtained by combining four analysis methods: the comparison of multiple beat detection algorithms, the comparison of different synchronous ECG leads, evaluation of the kurtosis (randomness) of an ECG segment and calculating the proportion of the spectral distribution of a given ECG segment. The SQI of ABP signals was obtained by a combination of two algorithms: a beat-by-beat fuzzy logic-based assessment of features in the ABP waveform and heuristic constraints of each ABP pulse to determine normality. The SQI ranges between 0 and 1 inclusively. High value of SQI means good quality and low value means bad quality.
After obtaining the SQI, we studied the estimation of the vital signs, such as heart rate (HR) and blood pressure, in the presence of high levels of noise and artifact. The HR directly obtained from the beat detection of ECG or ABP and the blood pressure directed obtained from ABP feature extraction are easily corrupted by noise. The Kalman filter is an optimal state estimation method for stochastic signals and can minimize the estimation error caused by noise. We studied the algorithms of HR and blood pressure estimation based on Kalman filter and proposed the idea to optimize the Kalman filter by adjusting the gain of Kalman filter according to the SQI. When the SQI is high, we elevate the Kalman gain and force the Kalman filter to trust the current measurement and use the current measurement to adjust the system state. When the SQI is low, we depress the Kalman gain and force the Kalman filter to trust the current measurement less. Furthermore, an upper limit that defines the cusp between good and bad data is defined. When SQI is lower than the upper limit, the KF is not updated. So it will escape the severe influence of noise when SQI is too low.
HR information can be obtained easily by beat detection from the ECG, ABP and pulse oximetry waveforms. These sources provide approximately redundant and independent measures of HR. Furthermore, the sources of the noise and artifact are often weakly or uncorrelated with both the signals that are cardiovascular in origin, and with each other. Reliable estimation of HR can therefore be obtained by sensor fusion. We studied the data fusion algorithm to obtain the optimal estimation from multi noisy signal sources. Then we realized the HR data fusing estimation based on the SQI and the innovation (residual error) of Kalman filter. We presented the novel idea that the data fusion can be weighted by SQI and the innovation of Kalman filter. When one channel is corrupted by artifact and the HR from this channel is miscalculated, the SQI will be low and the sudden change of HR will make the residual error large. So the weight for this channel will be set to a small value. Data fusion method can provide robust cardiovascular parameter estimates even when only one channel of data is relatively noise free.
After obtained the fusing HR, the arrhythmia false alarm reduction algorithm was performed based on the fusing HR and SQI. When the monitor alarm was set, we compared the fusing HR with the threshold of the monitor and made the decision of accepting or rejecting the alarm based on SQI. If the fusing HR did not exceed the threshold of the monitor and the SQI was high enough, we believed in the fusing HR estimation and suppressed the false alarm.
To evaluate the algorithms in this paper, we established a noise free dataset, named clean dataset, by selecting the noise free data from the Multi-Parameter Intelligent Monitoring for Intensive Care II (MIMIC II) database. The clean dataset included 437 subjects, comprising 3762 1 h or longer (1.62±0.69 h) data segments or 6084 h in total, with the ABP and at least one channel of ECG simultaneously present. The ECG and ABP evaluation dataset were established by adding ECG and ABP noise with different types and intensities to the clean dataset. The ECG noise sourced from MIT standard ECG noise database (NSTDB), including 3 types and 6 different signal-noise-ratio noises. We created models of ABP noise, including 6 types and 5 different noise percentages. The false alarm suppression evaluation dataset included a subset of over 5344 life-threatening arrhythmia alarms taken from the MIMIC II database and was annotated by experts. The false alarm rate of the dataset is 42.74%.
To evaluate the HR estimation algorithms, we obtained the golden standard HR which can be applied to evaluate the methods under noisy conditions. We evaluated eight different methods of HR estimation and three methods of blood pressure estimation.
The results show that the ECG and ABP SQIs of evaluation dataset decrease along with the increase of noise. The SQI reflects the degrees of noise in physiological signals effectively. The SQI plays an important role in eliminating the effect of noise and artifact from HR estimation. It is evident that when there is no SQI control, the error is much higher when the SQI is low (and the noise level is high). The Kalman filter and SQI based estimation algorithms give good HR and blood pressure estimations. The data fusion algorithm performs better. The root mean squared error (RMSE) of the fusing HR estimation is less than 1 beat/min when high levels of ECG noise exist. And the RMSE is less than 2.1 beats/min in the presence of high levels of ABP noise. The blood pressure estimation algorithm can suppress the estimation error differently accord to the sensitivity of SQI to noise types. The results of false alarm reduction algorithm show that the false alarm reduction rate was 53.68%, and the corresponding true alarm acceptance rate was 100.00% for extreme arrhythmia alarms. The false alarm rate decreases from 42.74% to 19.80%.
The sound vital signs estimation and the signal quality assessment are the base of intelligent monitoring. Only when the correct and reliable data were obtained, the physiological and pathophysiologic state of patient can be interpreted, predicted and assessed correctly. This paper still has some shortcomings. The SQI of ABP is not sensitive to some types of noise. This paper only studied ECG and ABP signals. The false alarm reduction algorithm only dealt with arrhythmia alarms. We would do more detail researches on other vital signs and other types of false alarm reduction later.
引文
Aboukhalil A, Mark RG, Clifford GD. A Method to Significantly Reduce the Incidence of False ECG Alarms in the ICU, Using Correlated Pressure Signals. Submitted to Biomedical Informatics, Special Issue, September 2007.
Allen J and Murray A 1996 Assessing ECG signal quality on a coronary care unit Physiol. Meas. 17 249-58
ANSI/AAMI EC57 1998 Testing and Reporting Performance Results of Cardiac Rhythm and ST Segment Measurement Algorithms American National Standards Institute
Chambrin M C 2001 Alarms in the intensive care unit: how can the number of false alarms be reduced? Critical Care 5 184-8
Brown, R G 1983 Introduction to Random Signal Analysis and Kalman Filtering, John Wiley and Sons, New York, NY
Clifford G D, McSharry P E, and Tarassenko L 2002 Characterizing abnormal beats in the normal human 24-Hour RR tachogram to aid identification and artificial replication of circadian variations in human beat to beat heart rate, Comput. in Cardiol. 29 129-32
Clifford G D, Aboukhalil A, and Mark R G, 2006 Using the Blood Pressure Waveform to Reduce Critical False ECG Alarms, Comput. in Cardiol. 33 829-832
Clifford G D 2006 Linear Filtering Methods, Chapter 5 in Advanced methods and tools for ECG data analysis Clifford G D, Azuaje F and McSharry P E (Eds.) (Norwood, MA: Artech house)
Chen L, McKenna T, Reisner A and Reifman J 2006 Algorithms to qualify respiratory data collected during the transport of trauma patients Physiol. Meas. 27 797-816
Ebrahim M H, Feldman J M and Bar-Kana I 1997 A robust sensor fusion method for heart rate estimation J Clin Monit 13 385-93
Feldman J M, Ebrahim M H, Bar-Kana I 1997 Robust sensor fusion improves heart rate estimation: clinical evaluation. J. Clin. Monit. 13 379-84
Friesen G M, Jannett T C, Jadallah M A, Yates S L, Quint S R and Nagle H T 1990 A comparison of the noise sensitivity of nine QRS detection algorithms IEEE Trans. Biomed. Eng 37 85-98.
Hamilton P S and Tompkins W J 1986 Quantitative investigation of QRS detection rules using the MIT/BIH arrhythmia database IEEE Trans. on BME 33 1157-65
Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PC, Mark RG, Mietus JE, Moody GB, Peng CK, Stanley HE. Physiobank, physiotoolkit, and physionet: Components of a new research resource for complex physiologic signals. Circulations 2000;101(23): e215-e220.
He T, Clifford G D, Tarassenko L 2006 Application of independent component analysis in removing artefacts from the electrocardiogram, Neural Comput. & Applic. 15 105-16
Jakob S, Korhonen I, Ruokonen E, Virtanen T, Kogan A and Takala J 2000 Detection of artifacts in monitored trends in intensive care Comput. Methods Programs Biomed. 63 203-9
Kaiser W and Findeis M 2000 Novel signal processing methods for exercise ECG Int. J. Bioelectro 2 61-65
Kohler B U, Hennig C and Orglmeister R 2002 The principles of software QRS detection IEEE Eng. Med. Biol. Mag. 21 42-57
Lawless S T 1994 Crying wolf: false alarms in a pediatric ICU Crit. Care Med. 22 981-5
Li Q, Clifford G D 2008, Realistic artificial arterial blood pressure artifact algorithms and an evaluation of a robust blood pressure estimator. In submission to Biomedical Engineering Online.
Mendelson Y 1992 Pulse Oximetry: theory and applications for noninvasive monitoring Clin. Chem. 38 1601-7
Moody G B, Muldrow W E and Mark R G 1984 A noise stress test for arrhythmia detectors Comput in Cardiol. 11 381-4
Moody G B and Mark R G 1997 Integration of real-time and off-line clinical data in the MIMIC Database. Comput. in Cardiol, 24 585-8
Murthy V K, Grove T M, Harvey G A and Haywood L J 1978 Clinical usefulness of ECG frequency spectrum analysis Computer application in medical care 1978 Proceedings 610-612
Saeed M, Lieu C, Raber G and Mark R G 2002 MIMIC II: a massive temporal ICU patient database to support research in intelligent patient monitoring Comput, in Cardiol 29 641-4
Sittig D F and Factor M 1990 Physiologic trend detection and artifact rejection: a parallel implementation of a multi-state Kalman filtering algorithm Comput. Methods Programs Biomed. 31 1-10
Sun J X, Reisner A T and Mark R G 2006 A signal abnormality index for arterial blood pressure waveforms Comput. in Cardiol. 33 13-16
Tarassenko L, Townsend N, Clifford G, Mason L, Burton J and Price J 2001 Medical signal processing using the software monitor Intelligent sensor processing (Kef. No. 2001/050) 3/1-3/4
Tarassenko L, Mason L and Townsend N 2002 Multi-sensor fusion for robust computation of breathing rate Electronics Letters 38 1314-16
Tarassenko L, Townsend N and Price J D 2003 Combining measurements from different sensors United States Patent Application number: 311250 (http://www.freepatentsonline.com/20030187337.html)
Townsend N 2001 Merging multiple estimates to form a single estimate Unpublished Technical Report (Oxford: Department of Engineering Science, University Oxford)
Tsien C L and Fackler J C 1997 Poor prognosis for existing monitors in the intensive care unit Crit. Care Med. 25 614-9
Tsien C L,Kohans I S and Mclntosh N 2001 Building ICU artifact detection models with more data in less time Proc.AMIA Symp.706-10
Welch G and Bishop G 2004 An introduction to the Kalman filter Dept.Comp.Sci.Univ.North Carolina Chapel Hill TR95-041
Zong W,Heldt T,Moody G B and Mark R G 2003a An open-source algorithm to detect onset of arterial blood pressure pulses Comput.in Cardiol.30 259-62
Zong W,Moody G B and Jiang D 2003b A robust open-source algorithm to detect onset and duration of QRS complexes Comput.in Cardiol.30 737-40
Zong W,Moody G B and Mark R G 2004 Reduction of false arterial blood pressure alarms using signal quality assessment and relationships between the electrocardiogram and arterial blood pressure Med.Biol.Eng.Comput 42 698-706
[1] C. Tsien, J. Fackler, "Poor prognosis for existing monitors in the intensive care unit," Crit. Care Med. 1997; 25(4):614-619
[2] A. Makivirta, E. Koski, A. Kari and T. Sukuvaara, "The median filter as a preprocessor for a patient monitor limit alarm system in intensive care," Comput. Methods Programs Biomed. 1991; 34:139-144
[3] D. F. Sittig, M. Factor, "Physiologic trend detection and artifact rejection: a parallel implementation of a multi-state Kalman filtering algorithm," Comput. Methods Programs Biomed. 1990;31(1):1-10
[4] W. Zong, G. B. Moody and R. G. Mark, "Reduction of false arterial blood pressure alarms using signal quality assessment and relationships between the electrocardiogram and arterial blood pressure," Med. Biol. Eng. Comput. 2004,42(5):698-706
[5] G. D. Clifford, A. Aboukhalil, J. X. Sun, W. Zong, B. A. Janz, et al., "Using the Blood Pressure Waveform to Reduce Critical False ECG Alarms," Comput. in Cardiol. 2006; 33:829-832
[6] M. Saeed, C. Lieu, G. Raber, R. G. Mark, "MIMIC Ⅱ: A massive temporal ICU patient database to support research in intelligent patient monitoring," Comput. in Cardiol. 2002; 29:641-644
[7] J. X. Sun, A. T. Reisner, R. G. Mark, "A signal abnormality index for arterial blood pressure waveforms," Comput. in Cardiol. 2006; 33:13-16
[8] W. Zong, T. Heldt, G. B. Moody, R. G. Mark, "An open-source algorithm to detect onset of arterial blood pressure pulses," Comput. in Cardiol. 2003; 30:259-262
[9] R. G. Brown and P. Y. C. Hwang, "Introduction to random signals and applied Kalman filtering", John Wiley & Sons, 1997
[10] Q. Li, R. G. Mark, G. D. Clifford, "Robust heart rate estimation from multiple asynchronous noisy sources using signal quality indices and a Kalman filter," Physiol. Meas., 2008; 29 (1): 15-32