Electronic patient monitors play a fundamental role in anesthetic management during surgery (1,2). They acquire, process, and display wave form signals, such as the electrocardiogram (ECG), arterial blood pressure, oxygen saturation, and respiration. In addition, these wave form signals are analyzed to derive clinical variables, such as heart rate, systolic, diastolic, and mean blood pressures, O2 saturation values, and end-tidal CO2 (EtCO2) values. The wave form signals, along with the derived parametric values, provide critical information on the clinical status of the patient, which is used directly to provide anesthesia care. Thus, it is imperative that patient monitors be able to process and display clinical information accurately and without distortion. Moreover, the displayed variables should represent the current status of the patient with minimal processing delays and signal averaging effects. However, achieving these desirable features is difficult because of artifacts that affect the wave form signals and subsequently the derived parameters. To minimize the effects of artifacts, currently marketed patient monitors use several filtering techniques. However, the success of these techniques has been minimal and artifacts still remain a significant problem.
Over the last decade, there have been considerable advances in signal processing techniques, particularly for biomedical signals. Many of these techniques, particularly those using nonlinear filters and artificial intelligence, have the potential to significantly improve artifact detection and elimination. In fact, the adoption of these techniques could revolutionize the way patient monitors process wave form signals, leading to more accurate displays of clinical information. In this article we review the clinical problem of artifacts and the artifact filtering methods adopted by currently marketed patient monitors. We also review artifact correction methods that have been recently developed by the research community. Possible approaches that the new generations of patient monitors could adopt to detect and eliminate artifacts are also proposed. Only artifacts that are related to technical and environmental factors are considered in this review. Artifacts that have a physiological origin are beyond the scope of the review, and are not considered.
ORIGIN OF COMMON ARTIFACTS
The main sources of artifacts that typically affect patient monitor data are well known (2–8). These sources can be either physiological or nonphysiological in origin. Most artifacts affect the clinical signals directly, resulting in erroneous derived parameter values. However, limitations in the patient monitor algorithms that process the source signals sometimes cause inaccuracies in the derived parameters. The common artifacts that affect ECG, invasive arterial blood pressures (ART), pulmonary artery blood pressure (PA) and central venous pressure (CVP), noninvasive arterial blood pressure (NIBP), pulse oximetry, capnography and temperature signals are tabulated in Table 1.
METHODS USED IN PATIENT MONITORS
We surveyed products currently sold by four of the major patient monitor manufacturers (GE Medical Systems, Milwaukee, WI; Phillips Medical Systems, Andover, MA; Datex Ohmeda, Helsinki, Finland; and Ivy Biomedical Systems, Bradford, CT) to assess the steps taken by industry to deal with the problem of artifacts. Patient monitors, in general, use simple linear filters to remove artifacts and noise. Such filters are either hardware or software filters, or a combination of both, and are implemented as part of the patient monitor and any interfacing signal acquisition modules. These filters remove artifact interference in frequency ranges outside those of the signals of interest. Though such filters help reduce the effects of artifacts, they are not entirely successful in artifact elimination. This is because there is a considerable overlap of frequency spectrum between the acquired clinical signals and artifact interference. Additionally, the parametric data derived from the wave form signals are also filtered by the patient monitors. Examples of filtering of parametric data include averaging over the last several valid beats (V24 monitor Phillips Medical Systems), averaging over the last several seconds (Vital Guard 450 monitor, Ivy biomedical systems), and infinite impulse response filtering (Solar 9500 monitor, GE Medical Systems). Though such techniques can minimize or “smooth” the effect of artifacts in the parametric data, they are not always successful in eliminating artifactual parameter values. Additionally, the larger the number of samples the monitor uses for averaging, the less “current” the displayed parameters will be.
Only a few manufacturers have developed some specialized schemes to remove specific types of artifacts. They are sometimes implemented as enhanced features of the monitor that the user can select. However, in many cases, such features are never enabled because of either user ignorance or because the user is not confident of the abilities of such algorithms.
The steps taken by industry to eliminate specific artifacts in the main signals of the patient monitor are described below.
Most monitors use some method to minimize the effect of electrosurgical unit (ESU) interference. These methods comply with American Association of Medical Instrumentation (AAMI) standards and generally use band-pass filters to minimize the effects of ESU noise. Most monitors keep the pass-band of such filters constant, but there has also been an attempt to adaptively vary the pass-band (Ivy Biomedical Systems) if ESU interference is detected. ESU interference is detected by monitoring the “slew rate” (the first differential of the signal). With this scheme, rapid fluctuation in the ECG signal caused by ESU interference causes the measured slew rate to be above a threshold. When ESU interference is detected, the pass-band of the filters is reduced, so that ESU interference is minimized without excessively distorting the primary ECG characteristics. Such a step will ensure that the heart rate parameter is computed with more reliability.
Interference from power sources could also affect ECG signals. To eliminate artifacts from power-line frequency, the monitors use notch filters to remove the frequency component specific to power-line interference.
None of the monitors use any artifact correction scheme specifically intended to eliminate baseline wander or motion artifacts. The effect of baseline wander is minimized by using a filter to remove low frequency components (generally below 0.5 Hz).
Invasive Arterial Blood Pressure Signals
Most models of patient monitors do not use any schemes to correct specific artifacts in arterial blood pressure signals. In general, simple band-pass filters are used to limit the effect of artifacts that could affect the computation of systolic, diastolic, and mean blood pressures. One manufacturer (GE Medical Systems) has used a “Smart BP” feature to automatically detect artifacts caused by zeroing and flushing of the pressure line or by drawing blood samples. Detection of these artifacts is performed through identifying patterns in the blood pressure signals that are created by each of the three sources of artifacts. The patterns represent the general trends in systolic, diastolic, and mean blood pressure values when a zeroing, flushing, or blood draw is performed. These patterns are detected by simple signal processing steps and heuristic rules. The “Smart BP” feature tested on a limited data set (18) seems to detect aforementioned artifacts accurately and consistently. However, further testing of this feature is necessary to fully assess its true accuracy. Moreover, the basic assumption of the algorithm that the patterns in blood pressure wave forms due to the aforementioned artifacts remain constant might be flawed. This could lead to artifacts being detected with low sensitivity and specificity.
Damping (overdamping or underdamping) of pressure signal is a common source of artifacts that could distort the pressure wave form. This could lead to over- or underestimation of pressure parameters. The current patient monitors do not use any technique to detect or compensate for this problem.
To the best of our knowledge, the currently marketed patient monitors do not use any specific algorithms to detect and eliminate specific artifacts that affect PA and CVP.
Noninvasive Arterial Blood Pressure
Most patient monitors use the oscillometry method to measure noninvasive arterial blood pressure (NIBP) (1,19,20). Though this method is immune to ambient noise, motion artifact is a major potential source of error (1,14,20–22). Irregularities in the oscillometric pulses are detected by most monitors, and if a pulse search fails, arterial blood pressure measurement is reattempted. To minimize the effect of motion artifacts, most monitors use techniques to sense oscillometric pulses in synchrony with the patient’s pulse (from Spo2 signal) or ECG. An example is the Dynamap technology (14,23) used by GE monitors that uses stepped deflation to identify pulses matched in frequency and amplitude. This algorithm has been improved (Dynamap SuperStat) to use a modified Gaussian curve to fit the oscillometric data, thus minimizing measurement time and the effects of artifacts (24). Similarly, the “Smart Cuff” technique (Protocol Systems, Beaverton, OR) directly synchronizes ECG and NIBP data to eliminate noncardiac pulses related to artifacts. An evaluation of the Smart Cuff algorithm has shown promising results (21,22). To improve the accuracy of NIBP measurements, additional steps have been adopted. An example is the use of a two-tube system (Dynamap) that uses separate tubes to inflate and sense pulsations, thus minimizing pressure damping effects, particularly when measuring low arterial blood pressures. Another manufacturer (Nihon-Kohden, Tokyo, Japan) uses a parameter indicative of changes in arterial blood pressure to trigger NIBP measurements, thus automatically initiating more frequent NIBP measurements when the arterial blood pressure is changing rapidly. Though all manufacturers comply with American Association of Medical Instrumentation (AAMI)/American National Standards Institute (ANSI) or European standards for NIBP measurements, there is still wide variability in measured values when using different monitors, or even when using different algorithms used by the same manufacturer (25,26).
Radial tonometry is an alternative to the oscillometry method and can measure arterial blood pressure more frequently. Additionally, this noninvasive method can provide an averaged tracing of the arterial wave form. However, this technique is sensitive to motion artifact and position of pressure sensor (27). To the best of our knowledge, currently marketed patient monitors do not use this technique, though stand-alone pressure measurement devices using this technique have been on the market for some time (27).
As described previously, motion artifact is the main type of artifact that affects pulse oximetry. Almost all pulse oximetry manufacturers have developed their own methods to minimize the effect of motion artifacts. The attempts include using adaptive noise cancellation (Masimo, Irvine, CA), frequency-based fourier artifact suppression (Phillips Medical Systems), combinations of time and frequency domain analysis and fuzzy logic (Nellcor, Pleasonton, CA). These methods have been shown to have varying success on the basis of the measurement type and situations (28). The patient monitors in general do not directly perform the pulse oximetry measurements but, rather, interface with measurement modules supplied by the pulse-oximeter manufacturer. Hence the patient monitors act as passive display devices as far as the Spo2 signal is concerned. They simply display the O2 saturation wave form and the Spo2 parameter value that are detected and computed by the pulse oximetry hardware module. For this reason, specific methods to correct artifacts in pulse oximetry data are generally not implemented in patient monitors.
Most capnography modules that interface to patient monitors use an infrared spectrography technique of gas analysis (15,16,29,30). These modules generally use a side-stream method to analyze gas samples aspirated from the breathing circuit (15,31,32). Moisture and secretions entering and clogging the breathing circuit and/or sampling line are common problems, although most monitors can detect this scenario and generate alarms (15,31). However, interference due to the presence of inhaled anesthetics, nitrous oxide, and high concentrations of O2 could potentially produce erroneous EtCO2 values (15,17). Monitors, in general, do not detect or correct for such interference. However, with recent monitors that use Raman scattering technology, this is less of a problem. Though Raman spectrography can measure CO2 faster and with greater accuracy than infrared technology, it is comparatively more expensive (15,16).
Most monitors do not detect or alarm when a temperature probe is dislodged from its intended measurement position. Similarly, artifacts in the PA catheter temperature measurements are not directly detected. However, if this leads to failure in cardiac output measurement, monitors generally generate error messages.
METHODS PROPOSED BY THE RESEARCH COMMUNITY
Research literature describes several methods to eliminate artifacts from the ECG, arterial blood pressure, and pulse oximetry signals. The methods described in the literature can generally be classified into two broad categories: 1) methods that extract the hemodynamic signal of interest from the artifactual signal, and 2) methods that explicitly detect artifact signals. The methods from the first category rely on approximate characterizations of the signals, while those of the second are based upon constructed models of artifacts.
A survey of the research literature revealed active research in developing methods that eliminate artifacts in the ECG signals. However, most of the efforts are geared towards improving ECG quality in nonoperating room (OR) environments, such as with ambulatory and Holter ECG recordings. Methods based on signal extraction (3,33,34) as well as artifact elimination (4,35–37) have been tested.
An example of the first method was the use of P, QRS, and T-wave frequency components (3) to differentiate the various features of the ECG from noise. A sample template of the ECG wave form was constructed using averaging techniques applied over several beats. Using the sample template, the frequency components of the ECG were derived and used to specify the pass frequency band of filters. However, the earlier application of such a technique did not adapt to the variation in the ECG signal among different subjects, and sometimes even in the same subject (5). Also, the technique failed to consider that the frequency spectrum of the ECG signal often overlaps with the spectrum of noise. As an improvement, in later studies (38), linear band-pass filters were modified to compensate for the nonstationary nature of the ECG signals. The newer filters had the capability to adapt to the changes in the ECG signal so that their operational parameters were based upon some feedback information. The feedback information is, in general, the difference between an estimate of the signal and the measured signal. The filter parameters were adjusted such that the difference between the estimated and the measured signals was forced to a minimum. This approach is used in techniques such as least mean square, recursive least square, and matching pursuit (39) to improve the accuracy of the adaptive filters. However, even these improved methods did not change the filters’ linear characteristics, and hence were still unreliable in dealing with the nonlinearity of the ECG noise characteristics. To accommodate nonlinearity, more recent ECG processing techniques have adopted methods of artificial intelligence (3). Artificial neural network is one such artificial intelligence method that is inspired by the way biological nervous systems, such as the brain, process information, and comprises a set of interconnected processing nodes that simulate neurons. Artificial neural network has the ability to successfully learn the embedded characteristics of the ECG signal and its inherent nonlinearity (3). However, the significant computational cost and processing time required by the algorithm has prevented its real-time application. Considerable research has also been done to extract ECG features by using wavelet filters (40–44). Wavelet transform decomposes the signal into multiple scales while preserving the temporal features of the signal. The scale functions are further analyzed to extract ECG features successfully (41,42) from noisy and nonstationary signals. Recently, the accuracy of ECG feature extraction has been improved by using a wavenet technique that combines wavelets with artificial neural networks (45,46).
Methods based on artifact elimination have used different types of adaptive filters to extract artifact signals (4,39). As an example, Aase et al. (4) used a multichannel Wiener adaptive filter to extract the cardiopulmonary resuscitation (CPR) artifact. A Wiener filter operates in the frequency domain to extract the noise signal by statistical means. However, this method suffered from computational complexity and was impractical for real-time application. In later studies (39), a matching pursuit algorithm was used to provide an efficient mechanism for detecting the CPR artifact. Using a moving average filter is another example of an artifact subtraction method to detect motion artifacts (47). In this approach, a least mean square method was used to adapt the noise filter parameters to accommodate the varying artifact situations. The artifact extraction and subtraction method generally requires several filters to extract the different types of artifact. Because ECG artifacts have various origins, it is not realistic to extract every type of artifact and then subtract it from the main ECG signal.
Invasive Blood Pressure and NIBP
Most of the current research work (35,48) focuses on artifacts, particularly motion artifacts that affect NIBP measurement. Initial attempts (48) to deal with motion artifacts were based on averaging techniques to minimize the fluctuations due to artifacts. Artifact elimination was improved by applying adaptive techniques, such as Kalman filtering (35), that was superior in dealing with time varying oscillation amplitudes and stochastic noise. The Kalman filter operates in time domain and recursively estimates a signal based on the previous estimate and the current measurement. The feasibility of artificial intelligence methods such as fuzzy logic has been studied successfully (6) to remove artificial interference and reconstruct the oscillation amplitude pattern. Though active research is conducted to minimize NIBP artifacts, our literature search revealed that very few attempts have been made to eliminate artifacts affecting the ART signal. A particular attempt by Nagai and Nagata (49) is worth mentioning. They applied different digital filters to detect systolic and diastolic arterial blood pressure values. Artifact values were eliminated by applying Smirnov’s rejection method, which is based on finding of outlier blood pressure values when the standard deviation of blood pressure values was more than 5%. The presence of outlier values was used as a trigger to eliminate segments of blood pressure values. Though this technique was successful in eliminating artifacts with reasonable accuracy, it adopted a very conservative approach and eliminated several nonartifact data points along with actual artifactual points. An interesting approach by Zong et al. (50) used fuzzy logic to analyze the relationship between ECG and arterial blood pressure wave forms to assess blood pressure signal quality and artifacts. Initial evaluation of this method in intensive care units (ICU) demonstrated that arterial blood pressure artifacts can be detected with high accuracy and specificity.
The pulse oximetry device evaluates blood photoplethysmographic (PPG) characteristics and detects hypoxia when the blood is not well perfused. The PPG can be corrupted by motion artifacts, ambient light interference, and low blood perfusion. Several algorithms are described in the literature that aims to eliminate artifacts in the Spo2 signal (7). Generally, these algorithms identify the artifacts by analytical means, through either spectral analysis of the signal or statistical correlation with other transducer signals. As an example, Hall (51) correlated the occurrence of a true blood pulse to the onset of cardiovascular activity, which was determined by observing the R wave in the QRS complex of an ECG signal. A microprocessor calculated the oxygen saturation using a preset formula and read the result from an experimentally determined reference table. This approach predominantly characterized the incoming PPG signal and rejected any part of the signal that did not fit the prior assumed signal characteristics, and the preset formula implicitly discriminated artifact components. Other approaches (7,52) used an artifact reduction method, where an additional measurement channel is used for deterministic removal of the modeled form of the motion artifact. The techniques described in the literature show promise in minimizing artifacts, but extensive studies evaluating their accuracy in a clinical environment are yet to be performed.
Quite recently, an entirely different way of improving the quality of the data extracted from physiological signals has been suggested. The method uses multisensor data fusion. For instance, Ebrahim et al. (53) and Feldman et al. (54) investigated the notion of combining sensors with similar information to improve the quality and reliability of the extracted data. Thus, they sought to develop a method for combining heart rate measurements from multiple sensors to obtain: 1) an estimate of heart rate that is free of artifact; 2) a confidence value associated with every heart rate estimate which indicates the likelihood that an estimate is correct; and 3) a more accurate estimate of heart rate than is available from any individual sensor. The raw signals used to estimate heart rate were the ECG, the arterial blood pressure signal, and the pulse oximetry signal. Their method sought to discriminate between high-quality and low-quality sensor signals and combine only the high-quality signals to derive an improved heart rate estimate. Past estimates of heart rate were used to derive a predicted value for the current heart rate that was mathematically “fused” with the sensor measurements. The level of consensus between the sensor measurements, the predicted value, and the physiologic credibility of the readings was used to distinguish between high-quality and low-quality readings. The authors stated that the method performed well using clinical data, noting that the fused heart rate estimate “was consistently as good or better than the estimate available from any individual sensor” and that the fused estimates “consistently reduced the incidence of false alarms compared with individual sensors without an unacceptable incidence of missed alarms.”
Few attempts are being made to improve detection and elimination of artifacts by better processing of capnography signal. However, improvements in capnography measuring technique and design of airway adapters have minimized the occurrences of some common artifacts. Molecular correlation spectroscopy, an improvement over conventional infrared spectroscopy, uses an infrared emission to precisely match the absorption spectrum of CO2 [Oridion, Needham, MA (16,55)]. This method not only improved the measurement accuracy, but also minimized the interference because of the presence of inhalations agents and nitrous oxide. Filterline airway adapters (Oridion) minimize the chance of sample line blockage due to fluids by using three channels that faces different directions. With this adaptor, aspiration of sample can take place even if only one channel is open. The use of Naflon to make the sampling tube also minimizes condensation of water vapor in the breathing circuit (16). Naflon selectively allows water vapor to pass from the interior of the tube to the exterior.
Artifacts are a common problem in the OR and ICU(8) as they distort the hemodynamic signals displayed by the patient monitor. This in turn leads to the display of wrong or invalid parametric values that are derived from the hemodynamic signals. Artifactual data can cause wrong interpretation of data and false annoying alarms (56–59). These data can also become inadvertently propagated into a data recording systems, such as a computerized anesthesia record keeping system, that automatically acquire data from the patient monitor (60–62).
To minimize the problem of artifacts, several measures can be adopted. These measures can be considered from clinical and engineering perspectives. The clinical measure should be toward adopting monitoring practices that minimize the chance of artifacts affecting measurement of biomedical signals. The engineering measures would be two-fold: 1) improving the design of biomedical sensors so that they are less tolerant toward artifacts; and 2) improving signal processing methods so that artifacts can be accurately detected and filtered. From the patient monitor design perspective, it is the second engineering measure that is the most relevant. Artifact detection and correction is a two-step process. The initial step is artifact detection, in which it is determined whether the acquired biomedical signal is corrupted by artifact. This is followed by the step of signal extraction, in which the signal of interest is extracted from the acquired signal with the artifact portion removed. The success of the second step will be determined by the quality of the signal extraction technique as well as the quality of the acquired signal. If the signal quality or signal-to-noise ratio is so low that signal extraction is neither possible nor reliable, the monitor should take the necessary steps not to process the signal, and thus not to provide erroneous parametric data. Additionally, the patient monitor should take the necessary steps to alert the clinician to the presence of artifacts so that necessary clinical steps can be taken to eliminate or minimize them. Monitors now on the market do perform the above tasks, although in a simplistic manner. Detection of artifacts is generally performed by measuring rapid and large fluctuations in the signal. Signal extraction or artifact removal is typically performed by applying linear filters. Though the techniques adopted by most patient monitors can detect obvious artifacts, they have limited success in detecting a wide variety of artifacts encountered in the intraoperative environment. Additionally, the simpler filtering techniques used in current monitors have been in existence for decades and are not ideally suited to filter time-varying and complex biomedical signals, and therefore their ability to extract the actual signal from one contaminated by artifacts is severely limited.
There is much room for improvement in the artifact detection methods adopted by current patient monitors. This is especially true considering that in recent years, there has been considerable progress in biomedical signal processing techniques and the computational power of microprocessors. These technological advancements should be used to provide more reliable and accurate artifact detection and signal extraction schemes. The new generation of patient monitors could possibly detect artifacts using a combination of artificial intelligence and signal processing techniques (63). The techniques of artificial intelligence, such as fuzzy logic, artificial neural networks, and genetic algorithms are designed to duplicate the human thought and decision-making processes. Such techniques could duplicate the decision-making process of a clinician when deciding whether a parameter value contains artifact. In the OR setting, the clinician decides whether to discriminate a parameter value as an artifact on the basis of several factors. The primary factor is the quality of the signal (or wave form) from which the parameters are derived. Inferior quality wave forms with irregular fluctuations or wave forms that show abnormal or nonphysiological patterns are obvious indicators that the parameters derived from such wave forms are artifacts. In such cases, the clinicians can select the parameters from an alternate source, if one is available. As an example, when the ECG signal is distorted by artifacts, the clinicians could choose to use the heart rate value derived from the blood pressure or pulse oximetry signals. Clinicians sometimes use additional pieces of information that are provided by the signal acquisition module, to make decisions about artifactual data. For example, some pulse oximetry manufacturers provide a signal quality index along with the oxygenation signal. During periods when the signal quality index is low, the clinician may ignore the Spo2 parameter value as artifactual. Occasionally, parameters containing artifacts are not obvious by simply qualitatively analyzing the source wave form. In such cases, clinicians might choose to refer to alternate wave forms or sources of information to judge whether a parameter is an artifact. For example, double counting of ECG complexes can be deduced by referencing the blood pressure and pulse oximetry wave forms, and the heart rate derived from these wave forms. Another clinical example occasionally encountered is the erroneous display of Spo2 values when the peripheral perfusion is poor. This situation can be deduced by gauging the pressure wave form and the cardiac output values. Artificial intelligence techniques can mimic the aforementioned human thought process. These techniques can take advantage of the fact that the patient monitor serves as a central source, where multiple signal information derived from a variety of biomedical signals is available. Additionally, such techniques can also use the underlying factor that the ECG, arterial blood pressure, and pulse oximetry signals are all generated by the beating of the heart. Hence there is a fair amount of information overlap among these signals, with the obvious common parameter being the heart rate. However, these three signals are acquired by different transducers: ECG-electrical, blood pressure-mechanical, and pulse oximetry-optical. Different signal transduction methodologies mean that there is little chance of common sources of artifact that affect these signals simultaneously.
In addition to using artificial intelligence methods, better signal processing methods can also be adopted by patient monitors to deal with artifacts. The use of better signal processing methods can help in two ways. First, it can provide more accurate information to the artificial intelligence module so that it can make better decisions on classifying signals as artifacts. Second, it will be able to efficiently and accurately extract the signal of interest from an artifactual signal. Most signal processing filters used in the existing patient monitors are linear and based on frequency domain designs. However, in reality, not only are the biomedical signals contaminated by nonlinear sources of noise, but also their frequency characteristics change over time. Several nonlinear and adaptive filtering techniques, such as the Kalman and H∞ filters, have been developed in recent decades and have been shown to have superior and optimal performance when removing nonlinear noise. Wavelet filtering is another option that patient monitors can use. Such filters, based on both time and frequency domain designs, are especially suited to filter signals whose frequency characteristics vary over time.
While dealing with artifacts, the primary task of the patient monitor is the detection and possible elimination of artifacts. However, many times, the signal is severely corrupted by artifacts such that extraction of reliable physiological information becomes impossible. Under these conditions, the patient monitor should be able to switch to the secondary task of alerting the clinician about the presence of the artifact. A message describing the possible source of artifact should be displayed. Additionally, the parametric data derived from the source signals should be displayed in a different format (different background color) to notify the clinician that the displayed values are old and that computation of current values cannot be performed because of poor signal quality. Most current patient monitors perform the above tasks to some extent; however, their ability to alert clinicians to the presence of artifacts is rather simplistic, probably because of the limited types of artifacts that are detected by current patient monitors.
In summary, our review found that artifacts remain a significant problem for processing and displaying correct clinical information by OR patient monitors. Most of the current patient monitors still use simple linear filters which are often ineffective to detect and eliminate artifacts. The new generation of patient monitors should adopt improved methods of artifact detection, elimination, and alerts. These methods could be based on artificial intelligence and advanced signal processing methods that take advantage of the fact that the patient monitors serve as a central source for multiple signals that have a fair amount of information overlap.
1. Lawrence JP. Advances and new insights in monitoring. Thorac Surg Clin 2005;15:55–70.
2. Pace NL. Technology assessment of anesthesia monitors. J Clin Monit 1992;8:142–6.
3. Xue Q, Hu YH, Tompkins WJ. Neural network based adaptive matched filtering for QRS detection. IEEE Trans Biomed Eng 1992;39:317–29.
4. Aase SO, Eftestol T, Husoy JH, et al. CPR removal from human ECG using multi-channel filtering. IEEE Trans Biomed Eng 2000;47:1440–9.
5. Chase C, Brady WJ. Artifactual electrocardiogram change mimicking clinical abnormality on the ECG. Am J Emerg Med 2000;18:312–16.
6. Lin CT, Liu SH, Wang JJ, Wen ZC. Reduction of interference in oscillometric arterial blood pressure measurement using fuzzy logic. IEEE Trans Biomed Eng 2003;50:432–41.
7. Hayes MJ, Smith PR. A new method for pulse oximetry possessing inherent insensitivity to artifact. IEEE Trans Biomed Eng 2001;48:452–61.
8. Cunningham S, Symon A, McIntosh N. The practical management of artifact in computerized physiological data. Int J Clin Monit Comput 1994;11:211–16.
9. Birkholz T, Schmid M, Nimsky C, et al. ECG artifacts during intraoperative high-field MRI scanning. J Neurosurg Anesthesiol 2004;16:271–6.
10. Gardner RM. Direct blood pressure measurement: dynamic response requirements. Anesthesiology 1981;54:227–36.
11. Kleinman B, Powell S, Kumar P, Gardner RM. The fast flush test measures the dynamic response for the entire pressure monitoring system. Anesthesiology 1992;77:1215–20.
12. Kleinman B, Powell S, Gardner RM. Equivalence of fast flush and square wave testing of blood pressure monitoring systems. J Clin Monit 1996;12:149–54.
13. Kleinman C, Frey K. Artifact mistaken for electrical interface recorded from a pulmonary artery catheter. J Clin Monit Comput 1998;14:361–3.
14. Stebor AD. Basic principles of non-invasive blood pressure measurement in infants. Adv Neonatal Care 2005;5:252–61.
15. Bhavani-Shankar K, Moseley H, Kumar AY, Delph Y. Capnography and anaesthesia. Can J Anaesth 1992;39:617–32.
17. AARC Clinical Practice Guideline. Capnography/capnometry during mechanical ventilation—2003 revision and update. Respir Care 2003;48:534–9.
18. Gardner RM, Monis S, Oehler P. Monitoring direct blood pressure: algorithm enhancements. IEEE Comput Cardiol 1986;13:607.
19. Pickering TG, Hall JE, Appel LJ, et al. Recommendations for blood pressure measurement in humans and experimental animals, Part 1: Blood pressure measurement in humans: a statement for professionals from the Subcommittee of Professional and Public Education of the American Heart Association Council on High Blood Pressure Research. Hypertension 2005;45:142–61.
20. Jilek J, Fukushima T. Oscillometric blood pressure measurement: the methodology, some observations and suggestions. Biomed Instrum Technol 2005;39:237–41.
22. Van Horn RN, Kahlke RJ, Taylor LA, Dorsett TJ. Noninvasive blood pressure performance: a reproducible method for quantifying motion artifact tolerance in oscillometry. Biomed Instrum Technol 2001;35:395–414.
23. Amoore JN. A comparative evaluation of the DINAMAP 8100 and DINAMAP Copact TS using a non-invasive blood pressure simulator. Blood Press Monit 1998;3:309–14.
24. Nelson RM, Stebor AD, Groh CM, et al. Determination of accuracy in neonates for non-invasive blood pressure device using an improved algorithm. Blood Press Monit 2002;7:123–9.
25. Murray IC, Amoore JN, Scott DHT. Differences in oscillometric non-invasive blood pressure measurements recorded by different revisions of Phillips Component Monitoring System. Blood Press Monit 2005;10:215–22.
26. Sims AJ, Reay CA, Bousfield DR, et al. Low-cost oscillometric non-invasive blood pressure monitors: device repeatability and device differences. Physiol Meas 2005;26:441–5.
27. Belani K, Ozaki M, Hynson J, et al. A new noninvasive method to measure blood pressure. Anesthesiology 1999;91:686–92.
28. Rheineck-Leyssius A, Kalkman C. Advanced pulse oximeter signal processing technology compared to simple averaging. Effect on frequency of alarms in the operating room. J Clin Anesth 1999;11:192–5.
29. Anderson CT, Breen PH. Carbon dioxide kinetics and capnography during critical care. Crit Care 2000;4:207–15.
30. Ahrens T, Sona C. Capnography application in acute and critical care. AACN Clin Issues 2003;14:123–32.
31. Block FE, McDonald JS. Sidestream versus mainstream carbon dioxide analyzers. J Clin Monit 1992;8:139–41.
32. Thomas PE. What’s the latest on carbon dioxide monitoring? Neonatal Netw 2004;23:70–3.
33. Xue Q, Tompkins W. A weight pattern study of neural networks with ECG signal pattern recognition. In: Proceedings of the annual international conference of the IEEE Engineering in Medicine and Biology Society, 1989:2023, 2024.
34. Lin K, Chang K. Classification of QRS pattern by an associative memory model. In: Proceedings of the annual international conference of the IEEE Engineering in Medicine and Biology Society, 1989: 2017, 2018.
35. Cao C, Kohane IS, McIntosh N. Artifact detection in cardiovascular time series monitoring data from preterm infants. Proc AMIA Symp 1999:207–11.
36. Cheung J, Hull S. Detection of abnormal electrocardiograms using a neural network approach. In: Proceedings of the annual international conference of the IEEE Engineering in Medicine and Biology Society, 1989:2015, 2016.
37. Langhelle A, Eftestøl T, Myklebust H, et al. Reducing CPR artifacts in ventricular fibrillation in vitro. Resuscitation 2001;48:279–91.
38. Tompkins W. Adaptive matched filtering for ECG detection. In: Proceedings of the annual international conference of the IEEE Engineering in Medicine and Biology Society, 1988;145, 146.
39. Husoy JH, Eilevstjonn J, Eftestol T, et al. Removal of cardiopulmonary resuscitation artifacts from human ECG using an efficient matching-pursuit like algorithm. IEEE Trans Biomed Eng 2002;49:1287–98.
40. Mallat S. Multiresolution channel decomposition of images and wavelet models. IEEE Trans Acoust Speech Signal Process 1989;37:2091–110.
41. Li C, Zheng C, Tai C. Detection of ECG characteristic points using wavelet transforms. IEEE Trans Biomed Eng 1995;42:21–8.
42. Sahambi JS, Tandon SN, Bhatt RKP. Using wavelet transforms for ECG characterization. IEEE Trans Eng Med Biol 1997;16:77–83.
43. Martinez U, Almeida R, Olmos S, et al. A wavelet based ECG delineator: evaluation on standard databases. IEEE Trans Biomed Eng 2004;51:570–81.
44. Lee JW, Lee GK. Design of adaptive filter with dynamic structure for ECG signal processing. Int J Control Autom Syst 2005;1:137–42.
45. Li C, Wang S. ECG detection method based on adaptive wavelet neural network. J Biomed Eng 2002;19:452–4.
46. Khajeh-Zadeh A, Rashidi-Nejah M, Farahmand H, Shojaee M. Adaptive prediction of non-stationary signals using wavelet networks. In: ACSE Conference, Cairo, Egypt, 2005.
47. Hamilton P, Curly M, Aimi R. Effect of adaptive motion-artifact reduction on QRS detection. Biomed Instrum Technol 2000;34:197–202.
48. Chon KH, Hoyer D, Armoundas AA, et al. Robust nonlinear autoregressive moving average model parameter estimation using stochastic recurrent artificial neural network. Ann Biomed Eng 1999;27:538–47.
49. Nagai R, Nagata S. New algorithmic-based digital filter processing system for real-time continuous blood pressure measurement and analysis in conscious rats. Comput Biol Med 1995;25:483–94.
50. Zong W, Moody GB, Mark RG. Reduction of false arterial blood pressure alarms using signal quality assessment and relationships between electrocardiogram and arterial blood pressure. Med Biol Eng Comput 2004;42:698–706.
51. Hall P. Apparatus for the detection of motion transients. US Patent 5,226,417, 1990.
52. Hayes MJ, Smith PR. Artifact reduction in photoplethysmography. Appl Opt 1998;37:7437–46.
53. Ebrahim MH, Feldman JM, Bar-Kana I. A robust sensor fusion method for heart rate estimation. J Clin Monit 1997;13:385–93.
54. Feldman JM, Ebrahim MH, Bar-Kana I. Robust sensor fusion improves heart rate estimation: clinical evaluation. J Clin Monit 1997;13:379–84.
55. Colman Y, Krauss B. Microstream capnography technology: a new approach to an old problem. J Clin Monit 1999;15:403–9.
56. Edworthy J, Meredith CS. Cognitive psychology and the design of alarm sounds. Med Eng Phys 1994;16:445–9.
57. Kestin IG, Miller BR, Lockhart CH. Auditory alarms during anesthesia monitoring. Anesthesiology 1988;69:106–9.
58. Lawless ST. Crying wolf: false alarms in pediatric intensive care unit. Crit Care Med 1994;22:981–5.
59. McIntyre JW. Ergonomics: anaesthetist’s use of auditory alarms in the operating room. Int J Clin Monit Comput 1985;2:47–55.
60. Hoare SW, Beatty PC. Automatic artifact identification in anaesthesia patient record keeping: a comparison of techniques. Med Eng Phys 2000;22:547–53.
61. Sanborn KV, Castro J, Kuroda M, Thys DM. Detection of intraoperative incidents by electronic scanning of computerized anesthesia records. Comparison with voluntary reporting. Anesthesiology 1996;85:977–87.
62. Gostt RK, Rathbone GD, Tucker AP. Real-time pulse oximetry artifact annotation on computerized anaesthetic records. J Clin Monit Comput 2002;17:249–57.
63. Mylrea KC, Orr JA, Westenskow DR. Integration of monitoring for intelligent alarms in anesthesia: neural networks—can they help? J Clin Monit 1993;9:31–7.