Mayo Clin Proc. 2004;79:339-350 © 2004 Mayo Foundation for Medical Education and Research

Review

Real-Time 12-Lead High-Frequency QRS Electrocardiography for Enhanced Detection of Myocardial Ischemia and Coronary Artery Disease

From the Neuroautonomic Laboratory (T.T.S.), Wyle Laboratories (W.B.K.), and Human Adaptation and Countermeasures (Biostatistics) (A.H.F.), NASA Johnson Space Center, Houston, Tex; Department of Engineering, Colorado State University-Pueblo (J.L.D.); Department of Engineering, Baylor University, Waco, Tex (J.S.W.); Division of Cardiology, University of Texas Medical Branch, Galveston (M.A.R.); and University of Texas, Houston Medical School, Houston (M.W.B.).

This work was supported in part by NASA grant 199161157, a Presidential Early Career Award, and Johnson Space Center Director’s Discretionary and Technology Development funds. The NASA Johnson Space Center Summer Faculty Fellowship and student cooperative programs supported Dr DePalma and Mr Wilson, respectively.

None of the authors presently has a financial interest in the technology. NASA has filed a patent application for the technology that specifies Drs Schlegel and DePalma as co-inventors.

Address reprint requests and correspondence to Todd T. Schlegel, MD, NASA Johnson Space Center, Mail Code SK3, Houston, TX 77058 (e-mail: todd.t.schlegel@nasa.gov).

Abstract

Several studies have shown that diminution of the high-frequency (HF; 150-250 Hz) components present within the central portion of the QRS complex of an electrocardiogram (ECG) is a more sensitive indicator for the presence of myocardial ischemia than are changes in the ST segments of the conventional low-frequency ECG. However, until now, no device has been capable of displaying, in real time on a beat-to-beat basis, changes in these HF QRS ECG components in a continuously monitored patient. Although several software programs have been designed to acquire the HF components over the entire QRS interval, such programs have involved laborious off-line calculations and postprocessing, limiting their clinical utility. We describe a personal computer–based ECG software program developed recently at the National Aeronautics and Space Administration (NASA) that acquires, analyzes, and displays HF QRS components in each of the 12 conventional ECG leads in real time. The system also updates these signals and their related derived parameters in real time on a beat-to-beat basis for any chosen monitoring period and simultaneously displays the diagnostic information from the conventional (low-frequency) 12-lead ECG. The real-time NASA HF QRS ECG software is being evaluated currently in multiple clinical settings in North America. We describe its potential usefulness in the diagnosis of myocardial ischemia and coronary artery disease.

Mayo Clin Proc. 2004;79:339-350

In the setting of a known or suspected acute coronary syndrome (ACS), the 12-lead conventional electrocardiogram (ECG) remains the single most important initial test for revascularization triage. However, an initial 12-lead ECG may be diagnostic for acute injury in only 24% to 60% of patients with a final diagnosis of acute myocardial infarction (MI).¹ Although the performance of serial conventional ECGs or continuous monitoring of conventional 12-lead ST segments improves the sensitivity of the ECG for detection of ACS,² diminution of the high-frequency (HF) components present within the central QRS complex of the ECG is known to be a more sensitive indicator for the presence of ischemia than changes in conventional ST segments.^3-5 Moreover, alterations within the HF QRS complexes of resting baseline HF QRS ECG recordings have been proposed as sensitive indicators for underlying coronary artery disease (CAD) in individuals with normal conventional 12-lead ECGs.^6-8

In this article, we describe a personal computer (PC)–based ECG software program recently developed at the National Aeronautics and Space Administration (NASA) that, for the first time to our knowledge, analyzes and displays, in real time during the actual patient encounter, changes in the HF QRS components in each of the 12 ECG leads during any chosen period of monitoring. Although several other programs have been designed to acquire the central HF QRS components, such programs have involved laborious off-line calculations and postprocessing, limiting their clinical utility. The NASA software currently is being evaluated in multiple clinical settings including emergency medicine departments (for enhanced evaluation of ACS), operating rooms (for enhanced monitoring of known or suspected myocardial ischemia in high-risk patients), cardiac catheterization laboratories (for determination of the success vs failure of percutaneous coronary interventions), and preventive medicine and cardiology clinics (for early and inexpensive diagnosis of CAD and for monitoring of heart failure).

The current version of the NASA software is focused especially on the real-time isolation, analysis, and display of HF components within the entire QRS interval and between the important frequencies of 150 to 250 Hz.^4,5,7,9 These HF components should be distinguished from the more familiar “late potentials,” which are signals of higher amplitude that are isolated in a broader frequency band (generally 40-250 Hz) in only the very latest portion of the QRS interval and not in the central QRS itself.¹⁰ The term HF QRS therefore refers to those signals between 150 and 250 Hz located throughout the signal-averaged QRS interval that are isolated by software via digital filtering.

To visualize the HF QRS components in real time, the following steps are taken: First, using PC-based ECG hardware and software (eg, CardioSoft, Houston, Tex, or Cardiax, IMED Co Ltd, Budapest, Hungary) plus any modern Windows-compatible PC, amplified ECG signals are continuously acquired at a rate of 1000 samples/s or faster. Next, the amplified, R-wave–detected incoming conventional QRS complexes are signal-averaged by the software to improve the signal-to-noise ratio. Premature complexes and noisy beats are automatically eliminated in real time (ie, not added to a growing template beat in each ECG channel) by a software-based cross-correlation function that rejects any incoming beat that is not at least 97% cross-correlated⁵ to the existing templates. The averaged complexes of the accepted beats are then bandpass-filtered by the software in real time using a digital filter that allows only those frequencies between 150 and 250 Hz to pass. The resulting microvolt-level HF QRS signals for each lead are then plotted in real time on the computer monitor and updated on a beat-to-beat basis adjacent to the higher-amplitude conventional ECG complexes. An example of a normal HF QRS complex from a healthy person derived from a lead I signal average is shown in Figure 1, A.

The 2 most commonly measured parameters of HF QRS signals are the root mean squared (RMS) voltage^{4,5,7-9,11,12} and the presence vs the absence of a splitting phenomenon known as a reduced-amplitude zone (RAZ).^6-8,12-15 Whereas RMS voltage provides an estimate of the total energy or amplitude of an HF QRS signal, the presence of a RAZ provides a morphologic indicator of potential pathology. As originally defined by Abboud,⁷ a RAZ occurs in an HF QRS signal when at least 2 local maxima of the signal’s upper envelope or 2 local minima of the signal’s lower envelope are present (Figure 1, B-D). A local maximum or minimum (darkened circles in the HF QRS signals in Figure 1, A-D) is in turn defined as an HF envelope sample point (peak or trough) within the QRS interval wherein the absolute value of its voltage exceeds that of the 3 envelope sample points (peaks or troughs) immediately preceding and following it. The RAZ (arrows, Figure 1, B-D) is thus the region lying between the 2 neighboring maxima or minima. Reduced-amplitude zones are important because they often are formed in the HF QRS complexes of individuals with CAD and during ischemia, even when there is no simultaneous change in the conventional ST segments.^{3,7,13,14,16-18} Abboud et al¹⁹ have successfully modeled both HF QRS complexes and the generation of RAZ phenomena using a computer simulation that includes the torso as a volume conductor, the cells of the myocardium, and a ventricular conduction system that branches 7 or more times in a fractal (self-similar) fashion. When an ischemic activation process is simulated in the model by reducing the conduction velocity in a small ventricular region that includes the late-branching Purkinje fibers, the local myocardial cells, or both, a RAZ is formed.¹⁹

The NASA software automatically searches in real time for local maxima and minima of the HF QRS envelope, not only according to the original criteria of Abboud but also separately according to stricter criteria that were developed to improve the usefulness of RAZ detection for clinical diagnoses. These empirical criteria identify, in addition to the generic RAZ described previously (ie, what we have termed the Abboud RAZ [RAZ A], after its original descriptor [Figure 1, B]), stricter RAZs that we have termed the Abboud Percent RAZ (RAZ AP) and the NASA RAZ (RAZ N). A RAZ AP (Figure 1, C) is simply a RAZ A that meets one additional criterion: its secondary local maximum (or minimum) has an absolute voltage that is at least X% of the absolute voltage of the primary local maximum (or minimum) located on the same side of the HF QRS envelope. The variable “X%” is user-adjustable in the software but currently defaults to a number between 25% and 33% (commonly 30%), depending on the exact clinical situation and based on prior analyses of several hundred HF QRS ECGs. In turn, the strictest type of RAZ (the RAZ N [Figure 1, D]) has both a secondary local maximum and a secondary local minimum, both having an absolute voltage of at least X% of their respective primary local maximum and minimum. The RAZ nomenclature is such that when a RAZ N is present in any given HF QRS complex, a RAZ AP and a RAZ A also must be present in the same complex by definition. Similarly, whenever a RAZ AP is present, a RAZ A must be present by definition.

As noted previously, the RMS voltage is a measure of HF QRS amplitude rather than morphology. The RMS voltage is calculated by squaring the amplitude of each sample point within the QRS interval, determining the mean of these squares, and then taking the square root of this mean.⁵ The RMS voltage and related measures of amplitude are important because they often are decreased during acute myocardial ischemia, even when conventional ST segments are unchanged.^4,5,7 Nevertheless, RMS voltage levels in single-baseline 12-lead HF QRS ECG recordings are subject to wide interindividual (and interlead) variability.^6,11 Thus, although serial measurements of RMS voltage in a given individual are potentially useful for monitoring ischemia over time,^4,5,7 RMS voltage levels alone in single resting recordings are probably insufficient to distinguish between individuals with and without CAD⁶ and between individuals with and without prior MI.²⁰ However, Seegobin et al^8,12 have claimed the contrary.

In contrast, Abboud et al⁶ have shown that a measure of morphology (ie, presence vs absence of RAZ) is potentially more useful than a measure of voltage (RMS) for identifying CAD from single-baseline multichannel HF QRS ECGs. This may not be surprising given that, in an isolated conventional 12-lead ECG, it is also generally the morphology of a QRS complex, rather than its total voltage, that is diagnostically most useful for identifying MI and myocardial ischemia. Although the concept of RAZ has limitations (eg, the appearance of a RAZ sometimes may represent an improvement rather than a deterioration in physiology if the basis for the RAZ appearance is the addition of energy, and therefore new envelope sample points, to an area where the HF voltage had previously been “zero”²¹), the importance of determining the presence or absence of RAZ needs to be reiterated.

Several examples of the diagnostic utility of the 12-lead HF QRS ECG are discussed subsequently. Some of the device requirements (hardware and software) necessary for real-time 12-lead HF QRS monitoring are detailed in the Appendix.

DIAGNOSTIC UTILITY AND FUNCTIONALITY

Figure 2, A shows the main HF QRS ECG screen of a healthy 25-year-old man with no history of cardiovascular disease. The HF QRS ECG complexes are shown below their respective signal-averaged ECG conventional QRS complexes. This snapshot was obtained with the NASA software’s autoscaling function in an “on” position, a setting that allows HF QRS morphology to be most readily evaluated by ignoring absolute voltages. Figure 2, B shows the same tracing as Figure 2, A but with the software’s autoscaling function switched to the alternative “off” position, allowing the absolute voltages rather than the morphologies of the HF QRS complexes to be compared. In this view, the RAZs are less visible because the 2 leads that contain them have relatively low RMS voltages. Overall, the 12-lead HF QRS ECG shown in Figure 2 is normal and is typical for a healthy young individual.

Figure 3 shows a morphology-optimized snapshot from the main HF QRS ECG screen of a 65-year-old man who had a normal conventional 12-lead ECG but advanced 2-vessel CAD, which was verified later the same day by cardiac catheterization performed as part of a work-up for chest pain. This individual had widespread RAZ formation, with a RAZ N present in HF leads II, aVR, V₁, and V₄ through V₆ and a RAZ A present in HF leads III, aVF, and V₃. His case illustrates the fact that resting 12-lead HF QRS ECG often can distinguish between individuals with (Figure 3) vs without (Figure 2) CAD, whereas conventional 12-lead ECG performed at the same time provides no such distinction. An ECG from a 57-year-old man who had 2-vessel CAD and a prior inferior MI is shown in Figure 4. This patient had even more extensive RAZ formation, with RAZ Ns present in all leads except for I (which has a RAZ AP) and V₁.

In our experience, we have found that referencing the HF precordial leads to the right arm electrode rather than to the Wilson central terminal (via a user’s switch in software), and thus using the CR precordial leads rather than the V precordial leads for HF QRS analysis (which maximizes QRS voltages^22,23), often allows an even better distinction between individuals with and without CAD. Some of the promising results of an observational study at the University of Texas using the standard limb HF QRS leads plus the CR precordial HF QRS leads for diagnosis of CAD are discussed subsequently.

Accuracy

Pettersson et al⁵ found that during percutaneous transluminal coronary angioplasty, changes in RMS voltages in 12-lead surface HF QRS ECGs are 100% sensitive for detecting acute left anterior descending artery occlusion, 87% sensitive for detecting acute right coronary artery occlusion, and 73% sensitive for detecting acute left circumflex artery occlusion, for an overall sensitivity of 88%. This sensitivity was substantially higher than that obtained from an analysis of the changes in the conventional ST segments during the same study (71%-79%), even considering the fact that the conventional ST segment change criteria were clinically liberal. Similarly, Aversano et al⁴ reported that changes in HF QRS RMS voltages in the Frank XYZ leads are 100% sensitive and 92% specific for revealing failed reperfusion therapy for ST-segment elevation ACS in an emergency department setting, thereby providing more useful information than resolution of either chest pain or ST-segment elevation. Beker et al⁹ presented evidence suggesting that changes in the HF QRS RMS voltage are more sensitive than changes in conventional ST segments for identifying CAD during exercise stress testing.

Abboud et al⁶ analyzed the sensitivity and specificity of resting-baseline HF QRS ECGs for identifying CAD in a group of patients presenting with chest pain but with normal conventional 12-lead ECGs. They specifically measured HF QRS complexes only in leads V₃ through V₅ and prospectively defined an HF QRS test as positive for CAD when a RAZ (RAZ A) was present in at least 2 of these 3 precordial leads. The sensitivity and specificity of this 3-precordial-lead morphologic HF QRS ECG test for identifying advanced CAD subsequently detected at catheterization were 75% and 80%, respectively. Using full 12-lead HF QRS ECGs and a new set of RAZ-based criteria that uses RAZ N and RAZ AP (rather than RAZ A) data from all channels, our preliminary retrospective data suggest that the sensitivity of resting HF QRS ECG for detection of CAD can be increased to 90% or more, with little compromise in specificity (Table 1).

Reproducibility

By using a Frank XYZ lead configuration and by studying serial HF QRS recordings every 15 minutes in 32 individuals, Aversano et al⁴ determined that the mean SD of RMS voltage during a 3-hour period in a single lead was approximately 0.3 μV and that the coefficient of variation of RMS was approximately 10%. They suggested that a change of RMS voltage of either 0.6 μV or 20% of the baseline voltage (ie, twice the nominal intraday variations) could be considered “significant” when evaluating HF QRS RMS voltage in a monitoring situation. However, Pettersson et al¹¹ subsequently reported lower ranges (approximately 4%) for intrasession HF QRS RMS variation when using all 12 conventional leads during a 5-minute period of monitoring in both healthy subjects and those who are ill.

Although RAZ reproducibility has not been studied extensively, Goldberger et al²⁴ found that waveform morphologies of central HF QRS signals in the range of 80 to 300 Hz were strikingly stable during a period of 100 days in a population of 10 subjects. Our more recent data, derived from 16 healthy subjects, suggest that the month-to-month reproducibility of RAZ Ns (ie, their presence vs absence) exceeds 87% in the average lead, with reproducibility of RAZ APs and RAZ As being slightly lower (>78% and >72% in the average lead, respectively). The reproducibility of RAZ Ns and RAZ APs compares favorably to that of more established signal-averaged ECG parameters such as late potential-related vector magnitude measurements.^25,26

RECENT HF QRS STUDIES AT NASA AND THE UNIVERSITY OF TEXAS

12-Lead HF QRS ECG as a Screening Device

We recently collected resting 12-lead HF QRS ECGs (100- to 300-beat signal averages) from more than 150 asymptomatic volunteers at NASA’s Human Test Subject Facility at the Johnson Space Center in Houston, Tex. These individuals ranged in age from 20 to 80 years and had 0 to 3 or more verified risk factors for CAD as defined in Table 1 (footnote).²⁷ In this table, the 103 individuals in the NASA database who had no known history of CAD and a normal conventional 12-lead ECG are grouped according to the number of CAD risk factors present, with elevated high-sensitivity C-reactive protein being included as a risk factor.²⁷ The numbers of RAZs that occur in the 12-lead (limb leads plus CR₁ through CR₆ precordial leads) HF QRS ECGs of these 103 individuals are also shown, with a given RAZ being considered present in a given lead only when it occurred during two thirds or more of the HF QRS recording (ie, to avoid overreliance on any single HF QRS snapshot beat). The probability of observing a lead containing a RAZ was significantly positively correlated with the number of risk factors present (1-sided P values were <.001, <.001, and .03 for the RAZ N, RAZ AP, and RAZ A, respectively). Statistical inference was performed using the method of generalized estimating equations on a binary response (RAZ presence vs absence as defined previously) with logit link function and an equicorrelated covariance structure on observations (ie, leads) pertaining to the same subject.²⁸

The lower half of Table 1 shows the number of RAZs occurring in the 12-lead (limb leads plus CR₁ through CR₆ precordial leads) HF QRS ECGs of 34 individuals who underwent elective cardiac catheterization for evaluation of chest pain at the University of Texas Medical Branch. None of the individuals in Table 1 had a QRS interval greater than 120 ms, cardiomyopathy, or left ventricular hypertrophy (ie, conditions that can cause notches and slurs in the conventional QRS and therefore possible RAZ formation even in the absence of CAD^14,29) or arterial collaterals of grade 2 or higher to a singularly diseased culprit vessel at catheterization, a pacemaker, right axis deviation, persistent arrhythmia, or preexcitation syndrome. Using the same statistical analysis noted previously and considering only patients with no prior MI, the incidence of all 3 types of RAZs was significantly increased in patients with (vs without) critical stenoses at catheterization (1-sided P values for the test of no mean increase were .04, .004, and .004 for the RAZ Ns, RAZ APs, and RAZ As, respectively). Furthermore, individuals with proven CAD and a previous MI had significantly more RAZs than those with proven CAD and no previous MI (1-sided P values were .005, <.001, and .02 for the RAZ Ns, RAZ APs, and RAZ As, respectively). We currently are using these data (and additional studies) to further develop resting 12-lead HF QRS ECG criteria that can optimally distinguish between individuals with and without critical CAD for use in prospective trials that will begin soon. Regarding the last column in Table 1, a simple retrospective analysis that incorporates as its diagnostic criterion the presence of either RAZ Ns in 3 or more spatially contiguous HF QRS leads or RAZ APs in 4 or more spatially contiguous HF QRS leads is by itself 92% sensitive and 77% specific for identifying uncomplicated CAD.

12-Lead HF QRS ECG Before and After Balloon Angioplasty

As noted previously, Pettersson et al⁵ recently described the substantial changes that occur in HF QRS RMS voltages of the 12 surface leads during balloon angioplasty. Their work complements the earlier work of Abboud et al,³ who used fewer leads but reported simultaneous changes in HF QRS morphology. We have preliminarily investigated these angioplasty-related changes in 12-lead HF QRS complexes (RMS voltage and morphology) that occur with vs without concomitant elevations in cardiac enzymes. Figure 5 shows a representative case of changes in cardiac enzymes that are more sensitively mirrored by changes in HF QRS complexes than by changes in conventional ST segments.

LIMITATIONS

Although the HF QRS ECG is clearly more sensitive than the conventional ECG for identifying myocardial ischemia and MI in monitoring situations,^4,5,7 its use in the diagnosis of CAD requires further study and verification before it can enter the clinical mainstream. For example, because loss of energy in the higher-frequency (150-250 Hz) QRS range often corresponds to an addition of energy (and therefore notching and slurring) in the conventional QRS complex,¹⁴ it is not surprising that notch-producing pathologic conditions such as ventricular hypertrophy,^30,31 collagen diseases affecting the heart,³² and noncoronary cardiomyopathies³² are, according to our preliminary data, also commonly associated with extensive RAZ formation, even in the absence of advanced CAD at catheterization. Although no published studies have specifically investigated the 150- to 250-Hz HF QRS in the context of these other cardiac conditions, idiopathic dilated cardiomyopathy is clearly associated with the formation of multiple local maxima and minima within the higher frequencies of the QRS.²⁹ In short, any other condition associated with a derangement in cardiac conduction might theoretically produce RAZs. Nevertheless, this limitation might be put to actual use in population screening environments in which the early detection of any type of cardiac disease—not only CAD—may be the goal.

Figure 6 shows a 12-lead HF QRS ECG from a patient with a complete left bundle branch block and illustrates the fact that individuals with severely prolonged QRS intervals (>120 ms) invariably have extensive RAZ formation in resting recordings regardless of coronary artery status. Therefore, a single baseline 12-lead HF QRS ECG is probably of little value in individuals with bundle branch blocks unless the ECG is obtained in the context of continuous monitoring or for the purpose of analyzing changes between serial tracings over time. Moreover, we have noted that healthy children, young adults, and athletic individuals who have a normal QRS interval but a relatively vertical QRS axis in the frontal plane (>80°) often have higher-than-expected RAZ counts as well. This finding is consistent with the larger number of notches and slurs found in the conventional QRS complexes of young and/or athletic populations, which has been attributed in the past to physiological preponderance of the right ventricle.^33,34 Thus, RAZ formation in children or in young, asymptomatic, and/or athletic individuals with rightward axes should be interpreted with great caution.

Finally, how HF QRS is affected by obesity, diabetes, microvascular disease, endothelial dysfunction, supraventricular arrhythmias such as atrial fibrillation and flutter, and commonly used antihypertensive and antiarrhythmic medications such as β-blockers, calcium channel blockers, and digitalis is largely unknown and requires further investigation.

CONCLUSIONS

We recently developed an advanced ECG software program at NASA that in conjunction with commercially available PC-based ECG hardware, acquires, analyzes, and displays HF QRS ECG components (150-250 Hz) within each of the 12 conventional ECG leads in real time and on a beat-to-beat basis while serving as a conventional 12-lead ECG monitor. The real-time aspect of the NASA HF QRS ECG software represents a step toward bringing what was previously regarded as a highly promising research technique into routine clinical use. Real-time 12-lead HF QRS ECG shows considerable promise, not only for enhancing the detection of myocardial ischemia and MI during continuous ECG monitoring but also, in carefully selected populations, for inexpensive screening for CAD and other cardiac conditions using suitable resting 12-lead ECG equipment and accompanying software.

Physicians interested in evaluating the HF QRS ECG software can request an educational (playback) version by e-mailing  todd.t.schlegel@nasa.gov.

APPENDIX

DEVICE REQUIREMENTS FOR REAL-TIME HF QRS ECG MONITORING

Hardware Aspects

As discussed, the intra-QRS HFs of greatest interest lie specifically in the range of 150 to 250 Hz.^4,5,7,9 Thus, the raw analog incoming ECG signals handled by the PC ECG hardware must first be sampled (digitized) at no less than 500 samples/s to adequately satisfy the Nyquist sampling criterion of at least twice the rate of the highest frequency of interest for HF signal retention. The current version of the NASA software specifically assumes an analog-to-digital sampling rate in the PC ECG hardware of at least 1000 samples/s but can be adjusted easily to conform to whatever greater sampling rates might be attainable by the PC ECG hardware.

Of note, most commercial “stand-alone” 12-lead ECG devices in clinical use today, including even most PC-based ECG devices, have low-pass anti-aliasing filters in hardware that intentionally eliminate all ECG frequencies higher than approximately 100 to 150 Hz from the acquired signals. Thus, almost all commercial ECG devices available today do not have a frequency-response range high enough to allow for HF QRS ECG analyses. Given that the major manufacturers of conventional 12-lead ECG devices have been interested principally in the conventional signal, to allow their devices to routinely perform 12-lead HF QRS analyses they would need to make a concerted commitment to change³⁵ their hardware amplifiers and to generally increase their sampling rates, which currently tend to be in the range of only 250 to 500 samples/s. Although some ECG manufacturers market specialized ECG machines with wider frequency responses and faster sampling rates that are used to analyze, eg, ventricular late potentials, one limitation concerning these more specialized machines is that they most often exclusively use the Frank XYZ lead configuration and do not easily allow for the standard 12-lead configuration.³⁵

Software Aspects

Several steps must be taken in software to prepare incoming multichannel ECG signals for real-time signal averaging and HF QRS analyses. For example, the first incoming conventional QRS complex (its R wave detected at a fiducial point) constitutes an obvious initial template beat for signal averaging in each channel. However, because the first beat may actually be an inappropriate beat for inclusion in the real-time signal average (eg, it may be a premature ventricular contraction), the templates must be built up initially such that ultimately, only appropriate beats are included and inappropriate beats are excluded. Although this can be accomplished in several different ways, we briefly describe one simple method for building the initial templates.

If the conventional QRS complex of the second incoming beat adequately cross-correlates with that of the first in each ECG channel, then the second beat is “accepted,” meaning it is added to, aligned with, and made a constituent of the growing template in each channel. If the conventional QRS complex of the second incoming beat does not adequately cross-correlate with that of the first in any channel, then the first (earliest) beat is discarded from the templates, and the second beat then becomes the provisional first beat in the reset templates. This process continues (ie, dropping of the earliest beat in the existing templates when any new incoming beat does not adequately cross-correlate) until “N” well cross-correlated accepted beats constitute each of the templates. After the templates consist of “N” accepted beats (the present default for “N” being 10 accepted beats, although “N” is adjustable), any new incoming beat that does not adequately cross-correlate with the existing template of “N” or more beats in any channel is itself rejected. Reasons for poor cross-correlation (and therefore beat rejection) include (especially) premature ventricular and atrial contractions, mechanical interference with one or more of the patient leads due to physical movement or other artifacts, and/or poor skin-electrode preparation. A percentage cross-correlation of any given new beat of 97% or greater with its lead’s existing template is generally considered adequate for beat acceptance,⁵ although the rejection threshold percentage is also user-adjustable.

In an emergency department, operating room, cardiac catheterization laboratory, or intensive care unit setting, continuous monitoring of changes in HF QRS-related parameters is particularly desirable. As alluded to previously, HF QRS ECG is more sensitive than conventional ECG for determining the success vs failure of thrombolysis, even when only 1 to 3 leads are used for HF QRS recordings.^4,7 Therefore, for continuous monitoring, in addition to the snapshot screens shown in the previous figures, which themselves update on a beat-to-beat basis, NASA’s HF QRS ECG software includes both long-term and short-term trend screens that plot not only the presence vs the absence of RAZs against time but also the values of kurtosis, RMS voltage, and other related parameters³⁶ against time.

Clinician-Device Interaction

The individual clinician who uses the software must decide the number of accepted beats that will constitute his/her signal average for HF QRS analyses. If the clinician is interested in real-time monitoring and the number of accepted beats desired and set for signal averaging is 100, for example, then the moment the 101st accepted beat enters the software, the first beat that had been accepted into the template will be dropped from the template, such that a sliding template consisting of exactly 100 accepted beats is maintained going forward in time. Immediately thereafter, when the 102nd accepted beat arrives, it will replace the second beat that had been earlier accepted into the template, and so forth. The ideal number of accepted beats chosen for the sliding template depends on the clinical situation. For a primary care physician who is principally interested not in real-time monitoring but rather in using HF QRS to screen for CAD in an ambulatory population,^6,8,12 having a good signal-to-noise ratio will be paramount. Because the signal-to-noise ratio increases by a factor of the square root of the number of beats in the signal average, this user would likely choose a sliding template consisting of a rather large number of accepted beats—perhaps 200 or more—whatever his/her busy clinical schedule might allow. Under these circumstances, the number of beats chosen for the sliding template would ideally exceed the number of beats that the physician intends to collect to ensure that all collected beats are definitively included in the final signal average. In contrast, for an emergency department physician, anesthesiologist, surgeon, or cardiologist who is interested in monitoring real-time changes in the HF QRS signals,^3-5 a sliding template of 200 or more beats may be too large, given that the relative influence of any suddenly ischemic new beats might be “washed out” by an existing template that already consists of many nonischemic beats. For these latter clinicians, responsiveness may be just as important as the signal-to-noise ratio, and a sliding template of 50 to 100 accepted beats might be chosen instead.

In addition to an adequate signal-to-noise ratio, other important factors that must be realized in software to ensure a robust HF QRS ECG recording include both the correction of R-wave detector jitter and the accurate delineation of the QRS interval. All R-wave detectors have jitter, meaning the slight temporal inconsistency that often occurs with respect to locating the R-wave fiducial point as QRS morphology varies from beat to beat, eg, with respiration. If such jitter is not corrected during signal averaging, a “fuzzy” QRS template results, with a consequent false attenuation of the HFs.^3,20 In the NASA software, jitter is eliminated by independently maximizing, in real time, the cross-correlation of incoming beats with their respective templates in each individual ECG channel. Although accurate delineation of the QRS interval is an even greater challenge, especially in a real-time application, the influence on the RMS voltage of a slight inaccuracy of QRS interval delineation can be dampened greatly by “padding” the QRS interval, eg, by 10 ms on either side.³⁶ The NASA software therefore also incorporates an optional and adjustable pad to the QRS interval for HF QRS analyses.

REFERENCES

1. Fesmire FM, Smith EE..  Continuous 12-lead electrocardiograph monitoring in the emergency department. Am J Emerg Med. 1993;11:54-60.
2. Drew BJ, Krucoff MW.  ST-Segment Monitoring Practice Guideline International Working Group. Multilead ST-segment monitoring in patients with acute coronary syndromes: a consensus statement for healthcare professionals. Am J Crit Care. 1999;8:372-386.
3. Abboud S, Cohen RJ, Selwyn A, Ganz P, Sadeh D, Friedman PL.  Detection of transient myocardial ischemia by computer analysis of standard and signal-averaged high-frequency electrocardiograms in patients undergoing percutaneous transluminal coronary angioplasty. Circulation. 1987;76:585-596.
4. Aversano T, Rudikoff B, Washington A, Traill S, Coombs V, Raqueno J.  High frequency QRS electrocardiography in the detection of reperfusion following thrombolytic therapy. Clin Cardiol. 1994;17:175-182.
5. Pettersson J, Pahlm O, Carro E, et al. Changes in high-frequency QRS components are more sensitive than ST-segment deviation for detecting acute coronary artery occlusion. J Am Coll Cardiol. 2000;36:1827-1834.
6. Abboud S, Belhassen B, Miller HI, Sadeh D, Laniado S.  High frequency electrocardiography using an advanced method of signal averaging for non-invasive detection of coronary artery disease in patients with normal conventional electrocardiogram. J Electrocardiol. 1986;19:371-380.
7. Abboud S.  High-frequency electrocardiogram analysis of the entire QRS in the diagnosis and assessment of coronary artery disease. Prog Cardiovasc Dis. 1993;35:311-328.
8. Seegobin RD, Mohamed SA, Ropchan G, Pym J.  Frequency content and sex difference of the Frank lead signal-averaged ECG in a population with significant coronary artery disease: comparison with concurrent 12-lead ECG morphology. J Electrocardiol. 1995;28(suppl):228-233.
9. Beker A, Pinchas A, Erel J, Abboud S.  Analysis of high frequency QRS potential during exercise testing in patients with coronary artery disease and in healthy subjects. Pacing Clin Electrophysiol. 1996;19(12, pt 1):2040-2050.
10.  Signal-averaged electrocardiography. J Am Coll Cardiol. 1996;27:238-249.
11. Pettersson J, Carro E, Edenbrandt L, et al. Spatial, individual, and temporal variation of the high-frequency QRS amplitudes in the 12 standard electrocardiographic leads. Am Heart J. 2000;139(2, pt 1):352-358.
12. Seegobin RD, Mohamed SA.  Sex differences and frequency content of the Frank lead signal-averaged ECG in a normal population. J Electrocardiol. 1996;29:105-109.
13. Mor-Avi V, Akselrod S.  Spectral analysis of canine epicardial electrogram: short-term variations in the frequency content induced by myocardial ischemia. Circ Res. 1990;66:1681-1691.
14. Mor-Avi V, Akselrod S.  Some aspects of the wideband recording of the electrocardiogram. Clin Cardiol. 1990;13:120-126.
15. Akselrod S, Mor-Avi V, Oz O, David D.  Effect of localized surface cooling of the heart muscle on the high-frequency content of ECG waveforms in dogs. Clin Cardiol. 1988;11:112-118.
16. Abboud S.  Subtle alterations in the high-frequency QRS potentials during myocardial ischemia in dogs. Comput Biomed Res. 1987;20:384-395.
17. Abboud S, Smith JM, Shargorodsky B, Laniado S, Sadeh D, Cohen RJ.  High frequency electrocardiography of three orthogonal leads in dogs during a coronary artery occlusion. Pacing Clin Electrophysiol. 1989;12(4, pt 1):574-581.
18. Abboud S, Cohen RJ, Sadeh D.  A spectral analysis of the high frequency QRS potentials observed during acute myocardial ischemia in dogs. Int J Cardiol. 1990;26:285-290.
19. Abboud S, Berenfeld O, Sadeh D.  Simulation of high-resolution QRS complex using a ventricular model with a fractal conduction system: effects of ischemia on high-frequency QRS potentials. Circ Res. 1991;68:1751-1760.
20. Ringborn M, Pahlm O, Wagner GS, Warren SG, Pettersson J.  The absence of high-frequency QRS changes in the presence of standard electrocardiographic QRS changes of old myocardial infarction. Am Heart J. 2001;141:573-579.
21. Berkalp B, Baykal E, Caglar N, Erol C, Akgun G, Gurel T.  Analysis of high frequency QRS potentials observed during acute myocardial infarction. Int J Cardiol. 1993;42:147-153.
22. De Gasperi RN, McCulloh DH.  CR leads in cardiac emergencies: a preliminary study. Chest. 1991;99:904-910.
23. De Gasperi RN, Ezzudin SH, Bauerlein EJ, Sequeira R, Lemberg L, Duncan RC.  Digitized electrocardiograms recorded with bipolar right-infraclavicular leads compared to electrocardiograms recorded with unipolar chest V leads and bipolar lead II. J Electrocardiol. 2002;35:125-134.
24. Goldberger AL, Bhargava V, Froelicher V, Covell J.  Effect of myocardial infarction on high-frequency QRS potentials. Circulation. 1981;64:34-42.
25. Aganauskiene J, Sornmo L, Atarius R, Blomstrom-Lundqvist C.  Reproducibility of the signal-averaged electrocardiogram using individual lead analysis. Eur Heart J. 1995;16:1244-1254.
26. Savelieva I, Aytemir K, Hnatkova K, Camm AJ, Malik M.  Short-, mid-, and long-term reproducibility of the atrial signal-averaged electrocardiogram in healthy subjects: comparison with the conventional ventricular signal-averaged electrocardiogram. Pacing Clin Electrophysiol. 2000;23:122-127.
27. Patel VB, Robbins MA, Topol EJ.  C-reactive protein: a ‘golden marker’ for inflammation and coronary artery disease. Cleve Clin J Med. 2001;68:521-524,527-534.
28. Zeger SL, Liang KY, Albert PS.  Models for longitudinal data: a generalized estimating equation approach [published correction appears in Biometrics. 1989;45:347]. Biometrics. 1988;44:1049-1060.
29. Maehara K, Kokubun T, Awano N, et al. Detection of abnormal high-frequency components in the QRS complex by the wavelet transform in patients with idiopathic dilated cardiomyopathy. Jpn Circ J. 1999;63:25-32.
30. Flowers NC, Horan LG, Thomas JR, Tolleson WJ.  The anatomic basis for high-frequency components in the electrocardiogram. Circulation. 1969;39:531-539.
31. Flowers NC, Horan LG.  Diagnostic import of QRS notching in high-frequency electrocardiograms of living subjects with heart disease. Circulation. 1971;44:605-611.
32. Reynolds EWJr, Muller BF, Anderson GJ, Muller BT.  High-frequency components in the electrocardiogram: a comparative study of normals and patients with myocardial disease. Circulation. 1967;35:195-206.
33. Raunio H, Anttonen VM, Savola P, Pyykonen L.  Notching in the QRS complex in the electrocardiograms of sportsmen. Ann Clin Res. 1969;1:5-12.
34. Raunio H, Anttonen VM.  Initial and terminal notching of the QRS complex in the conventional electrocardiogram. Am Heart J. 1972;83:717-719.
35. Selvester RH.  A call to concerted action for change: the recording, display, and reporting of the routine electrocardiogram. J Electrocardiol. 1998;31:367-370.
36. Xue Q, Reddy BR, Aversano T.  Analysis of high-frequency signal-averaged ECG measurements. J Electrocardiol. 1995;28(suppl):239-245.