Edward J. Berbari

Critical analysis of the signal-averaged electrocardiogram. Improved identification of late potentials.

BackgroundThis study performed a critical analysis of signal-averaging methods. The objective was to optimize detection of late potentials. Methods and ResultsWe studied two patient populations: a low-arrhythmia-risk group with no evidence of heart disease and a group with clinically documented ventricular tachycardia (VT). Filtered QRS duration (QRSD) and terminal QRS amplitude (RMS40) were measured from the vector magnitude. A QRS duration based on the latest detectable ventricular activity in any of the three individual XYZ leads was also measured. Because of improved signal-to-noise ratio, both individual lead analysis and extended (600- versus 200-beat) averaging yielded significant changes in signal-averaged ECG parameters. Both approaches gave an increased sensitivity for VT identification. Sensitivity, specificity, and accuracy were evaluated as functions of critical values of QRSD and RMS40. RMS measurements in the terminal QRS, ranging from 20 to 100 msec and including RMS40, did not contribute to maximizing sensitivity and were highly correlated with QRSD. Our results from the low-arrhythmia-risk group suggest that age and sex should be considered in the definition of late potentials. ConclusionsWe propose a VT risk stratification scheme using signal-averaged ECG parameters obtained from both individual lead and vector magnitude analysis. This allows definition of four categories of VT risk derived statistically from the study data. This definition is based on combined measures of sensitivity, specificity, and negative and positive predictive value.

Principles and signal processing techniques of the high-resolution electrocardiogram

T HIS ARTICLE presents an overview of the signal processing methods used to estimate and subsequently analyze late potentials in the high-resolution electrocardiogram (ECG). The high-resolution ECG is universally acquired using signal averaging techniques. The goal in acquisition is to improve the signal-tonoise ratio of the ECG: Resolution of signals on the order of 0.5 uV is required to detect late potentials accurately. The essential elements of signal averaging include computerized data acquisition, QRS detection and alignment, averaging, and noise measurement, and each is discussed in detail. Analysis of the signal average in the time domain is presented with discussions of filtering, vector magnitude transformation, and measurement techniques. Frequency domain analysis is introduced, with a discussion of spectral analysis techniques. Application to the high-resolution ECG is presented. Arguments for time-varying spectral representations and advanced time-frequency distributions are then considered. A summary of applications of other signal processing techniques to the high-resolution ECG follows. Several classes of Wiener filtering methods that have the potential to improve the signal-to-noise ratio significantly, compared with conventional signal averaging, are introduced. The review concludes with a description of techniques of high-resolution ECG analysis from Holter recordings and possible time-varying late potential activity.

Signal-Averaged Electrocardiography: A New Noninvasive Test To Identify Patients at Risk for Ventricular Arrhythmias

Signal-averaged electrocardiography (ECG) is a new noninvasive test for identifying patients at risk for ventricular arrhythmias. This computerized method of analyzing standard ECGs identifies particular microvolt-level signals called late potentials. Late potentials have been correlated with clinical ventricular tachycardia, are predictive of ventricular tachycardia inducibility at the time of electrophysiologic testing, and are predictive of arrhythmic events after myocardial infarction. In this review, we describe late potentials, the method of obtaining and processing the signal-averaged ECG, and clinical studies in various patient groups that have assessed the predictive value of the signal-averaged ECG for identification of patients at risk for subsequent ventricular arrhythmias.

Evaluation of Esophageal Electrodes for Recording His-Purkinje Activity Based Upon Signal Variance

Signal averaging the ECG to observe low level signals generated by His¿Purkinje system (HPS) has many disadvantages and has not been widely used. This is partly due to the inability to distinguish between atrial and HPS potentials within the PR segment. To more clearly observe the onset of the HPS activity, several leads with an esophageal electrode were studied and compared to a bipolar surface lead. Signal variance was calculated to estimate the noise levels present in all lead systems. The bipolar surface lead consistently provided reproducible HPS waveforms with accurate estimates of noise levels. The leads with an esophageal electrode (bipolar or esophago¿thoracic) failed to show reproducible HPS waveforms and had higher noise levels than the bipolar surface lead as measured by signal variance. The cause of these problems is the motion of the electrode within the esophagus, as visualized fluoroscopically. While the novel lead systems were not adequate for recording HPS waveforms, the analytic methods for evaluating signal averaged recordings from various lead systems provided a basis for optimizing this approach for quantifying low level cardiac signals.

Selective beat signal averaging and spectral analysis of beat intervals to determine the mechanisms of premature ventricular contractions

Selective beat signal averaging and spectral analysis were implemented to identify whether reentry, triggered activity or parasystole is the mechanism of the isolated premature ventricular contraction (PVC). 2-hour, high resolution ECG recordings were digitally obtained from 9 patients with frequent PVCs (>250/hr). For reentry, in 7/9 patients with a positive average for late potentials (LP) there were no changes of LP in sinus beats before and after PVCs. However, changes in the T/U waves in the difference waveform between the pre-PVC beat and its precedent beat (in 2/9 patients), and their T/U wave amplitudes (in 9/9 patients, mean amplitude difference of 0.0269 mV) implied a triggered activity, mechanism. Also, significant periodicities in PVC interval series and low correlation between sinus and PVC intervals (in 9/9 patients), suggested a parasystolic mechanism. Based on the limited data set, triggered activity and parasystole may be significant PVC mechanisms while reentry may not.<<ETX>>

Time-frequency structure of the high-resolution ECG

This study considers the problem of representing high-resolution ECG (HRECG) signals in the time-frequency plane using spectrotemporal mapping (STM). High-resolution ECG signal components overlap in both time and frequency. The central issue with STM techniques is whether sufficient time-frequency resolution exists to discriminate normal and abnormal QRS signals. The trade-off between signal resolution in time and in frequency must be made without a priori knowledge of the HRECGs time-frequency structure. A simulation experiment was performed to examine the time-frequency distribution of normal, abnormal low-level (late potentials), and bundle branch block components of the QRS. Results suggest that discrimination of these signals with STM is problematic. Signals and noise within the HRECG ensemble can, however, be easily distinguished. This observation forms the basis of a new optimally filtered ensemble averaging technique for signal-to-noise ratio enhancement.

Changes in Late Potential Measurements as a Function of Decreasing Bandwidth

Bandwidth of Late Potentials Introduction: Limited bandwidth systems such as Holter recorders are being used to record cardiac late potentials. Standard late potential systems have a low‐pass frequency of 300 Hz. This suggests that Holter tape systems may significantly distort the late potential measurements, when compared to standard late potential systems. Methods and Results: Signal‐averaged recordings were obtained with an analog bandwidth of 0.05‐300 Hz. Digital filters were then used to do high‐pass filtering and low‐pass filtering. All XYZ signal‐averaged recordings were low‐pass filtered at frequencies of 250, 150, 100, 75, 50, and 25 Hz using a second order Butterworth filter. A fourth order 40 Hz bidirectional high‐pass filter was then applied to each set of recordings. The QRS duration (QRSd) was measured in the filtered vector magnitude. Thirty‐eight syncope patients who had negative electrophysiologic studies for ventricular tachycardia, no prior infarction, normal ejection fractions (> 50%), and a QRSd <120 msec served as the negative late potential group. Thirty‐seven patients with clinical ventricular tachycardia, a positive electrophysiologic study for ventricular tachycardia, and a QRSd>120 msec served as the positive late potential group. Conclusions: A statistically significant lengthening of the QRSd (P <0.05) was observed in the patients with late potentials between the 250 Hz and 25 Hz low‐pass filter. There was a statistically significant shortening of the QRSd in the negative late potential patients between 250 Hz and 25 Hz, as well as between 250 Hz and 50 Hz, with P<0.05 in both cases. Thus, significant changes in late potentials are seen in limited bandwidth recordings that may limit the clinical usefulness of such systems.

Spectral Estimation of the Electrocardiogram

Langner introduced the idea of a high fidelity electrocardiogram (ECG) using an oscilloscope display and a film recorder.’ Subsequently, he examined this high resolution ECG in the context of myocardial infarction2q3 and coronary artery disease4 and concluded that slurs and notches within the QRS complex were strongly correlated with the presence of disease. This association was strengthened by Flowers et al. with correlation between QRS notching and anatomic identification of infarct scam5-’ Initial approaches for quantifying the spectrum of these notches and slurs in the Q R S complex were done with narrow band analog filters in an attempt to characterize the frequency bands of normal and notched QRS complexes.89 Subsequently, the use of digital computers gave rise to the discrete Fourier transform (DFT) method of spectral estimation.” These methods, for the most part, were applied to the QRS complex for analysis. Two trends in cardiac electrophysiology gave rise to a reexamination of spectral estimation of the ECG. The first was the maturing of invasive cardiac electrophysiology and the study of arrhythmogenesis. Directly recorded cardiac electrograms identified arrhythmia substrates where slow conduction and nonuniform propagation occur. The second trend was the development of high resolution ECGs (HRECG) using computer implemented signal averaging to improve the signal-to-noise ratio of the ECG and allow the identification of very low level signals. The most common application of this methodology is the identification of signals after the QRS complex.”-” These “late potentials” have been extensively studied in various clinical populations and are assumed to originate from damaged or diseased regions of the ventricles such as the surviving border zones of myocardial infarction scars. Thus the notches and slurs identified in the earliest studies may likely originate in these regions and late potentials are those that occur latest in the cardiac cycle. Goldberger et al. applied Fourier analysis to signal averaged ECGs but did not consider late po ten t ia l~ . ’~*’~ High pass filtering has been considered a necessary tool for

Methods and results of noninvasive detection of ventricular late potentials

In 1973, Boineau and Cox [1] showed that electrograms from ischemic regions of the canine heart were delayed and fractionated. Waldo and Kaiser [2] described bridging of the diastolic interval with low level electrical activity. Then Scherlag et al. [3], Hope et al. [4], Williams et al. [5] and El-Sherif et al. [6, 7], all working in the same laboratory, evolved an animal model for monitoring and relating this delayed or continuous electrical activity to the generation of ventricular arrhythmias. A wide variety of observations verified the relationship between late potentials and ventricular arrhythmias. These included patterns such as 2:1 block in the ischemic/infarct zone as well as progressive beat-by-beat prolongation of late potentials evolving to continuous electrical activity anteceding ectopic ventricular discharges and ventricular tachycardia. Other studies [8–10] also described prolongation and conduction block of late potentials as a function of heart rate and drugs. Multi-electrode epicardial mapping by El-Sherif et al. [11] and Wit et al. [12] has further verified these late activated areas as the substrate of reentry. An editorial comment by Josephson and Wit [13] clearly emphasizes the role of continuous electrical activity in defining reentry as a mechanism of ventricular tachycardia. Initial studies in man on late potentials were reported by Josephson et al. [14] and Fontaine et al. [15] which were recorded from the endocardial and epicardial surface, respectively, of patients with ventricular tachycardia.

Optimal Filtering Of The Signal-Averaged ECG Using Time-frequency Representations

This paper proposes the use of time-frequency representations of the signal-averaged ECG for optimal filtering. The goal of optimal filtering is to improve the signal-to-noise ratio of the ECG. This is achieved by reducing the variance of the signal estimate compared to that obtained by signal averaging at the expense of introducing bias. The optimal filter can be expressed as a discrete function of time and frequency and found using averaging and smoothing techniques. The short-time Fourier spectrum was used to estimate the filter. An example is shown using simulated signals and noise.