## Abstract

Traditional analyses have assumed that cardiac electrical activity is reflected on the surface ECG without distortion as the signal passes through the body tissues. This study aims to explore the frequency dependence of thoracic attenuation of surface-recorded intracardiac electrical activity. Twenty patients (14 men, 55 ± 15 yr of age) undergoing electrophysiological study were enrolled. Rectangular unipolar stimuli were applied from a catheter positioned in the right ventricular apical area and another in the posteroseptal area without contact with the myocardium. An orthogonal Frank-lead surface ECG and a unipolar intracardiac electrogram near the pacing site were recorded. Frequency domain characteristics of the signal-averaged pacing impulses were analyzed. Linear regression analysis showed significant frequency-dependent attenuation in the magnitude transfer functions (*R*^{2} = 0.84–0.89, *P* < 0.0001) and good linear fit for the phase transfer characteristics (*R*^{2} = 0.98–1.0, *P* < 0.0001). Age, physical dimension, and respiratory characteristics had significant effects on the magnitude and phase characteristics of the transfer functions. Application of models of the low- and high-slope transfer functions to signal-averaged ECGs from 33 subjects showed differences in the attenuation of P and T waves relative to the QRS.

- electrocardiography
- pacing
- Fourier analysis

body surface recordings of electrical activity arising from internal organs are routinely used in clinical practice, i.e., ECG, electroencephalography, and electromyography. For these techniques, it is assumed that although the local internal electrical activity is attenuated as it is transmitted to the body surface, the attenuation is uniform (not frequency dependent) and does not distort the signal. For ECG, one of the most widely applied clinical tests, this is a particularly important assumption, inasmuch as the frequency content of the different components of the ECG (P wave, QRS complex, and T wave) varies. Inasmuch as there are clinical implications for the measured amplitudes and relative timing of these signals, it is important that the ratio of capacitive to resistive components be small or, at the very least, similar from patient to patient. Attenuation in a purely resistive medium is the same for all frequencies, whereas a capacitive component to thoracic impedance would be associated with frequency-dependent attenuation and propagation delay. Previous in situ studies have shown that impedances for various tissues are higher at low frequency than at high frequency (2, 12–14). However, it was concluded that these capacitive influences can be assumed to be negligible for the purpose of ECG (14). This assumption has not been validated in humans.

Several obstacles need to be overcome to study the effects of thoracic tissue on the attenuation properties of cardiac signals as recorded on the body surface. The main difficulty is that cardiac potentials can vary greatly from person to person with heart size, heart orientation, heart disease, and arrhythmias. Population-based studies to examine the effects of body characteristics on the surface ECG are thus often limited to a specific subset of subjects (e.g., healthy subjects), and large sample sizes are required to find trends. The subject-to-subject variation of cardiac potentials also presents a challenge when an attempt is made to characterize the frequency dependence of transthoracic impedance.

In the present study, we employed a novel technique that allows the in vivo measurement of the frequency dependence of thoracic attenuation of surface-recorded potentials from a known source. Thoracic electrical transfer functions specific to each patient were obtained. This enabled us, for the first time, to assess how attenuation and frequency response differ with lead direction, origin of activation, and body characteristics. We hypothesized that the human thorax demonstrates significant frequency-dependent attenuation, which varies from patient to patient, and that these transfer characteristics are significant enough to alter surface ECG recordings.

## MATERIALS AND METHODS

The basic approach used in the present study was to apply a known signal from a point source from within the heart during an isoelectric, or electrically inactive, time period and record this signal on the body surface. Because the signal arises from a point source within the heart, it produces only a small signal on the body surface. To enhance the signal-to-noise ratio of this body surface signal, signal averaging was performed. The averaged signals from inside the heart and on the body surface were analyzed using standard frequency domain analysis techniques. This allowed for characterization of the frequency-dependent attenuation and evaluation of the effect of body characteristics on attenuation.

#### Patients.

Twenty patients undergoing electrophysiology studies [14 men and 6 women, 55.4 ± 15.4 (18–79) yr of age] were enrolled in the protocol. Physical dimensions (height, weight, chest circumference, chest height, and chest width) and respiratory measures obtained from spirometry [forced expiratory volume in 1 s (FEV_{1}) and forced vital capacity (FVC)] are shown in Table 1. All patients provided written, informed consent for the study, which was approved by the Institutional Review Board of Northwestern University.

#### Study protocol.

On their arrival at the electrophysiology laboratory, the patients' skin was thoroughly cleansed and mild abraded, and Ag-AgCl adhesive electrodes were applied as follows: *1*) *x* leads in the right and left axilla; *2*) *y* leads at the manubrium sternum and the left upper abdomen; and *3*) *z* leads at the fifth intercostal space in the midclavicular line and the opposite point in the posterior chest. In male patients, the chest hair was shaved as needed for proper electrode attachment. Multiple-electrode catheters were positioned in the heart, as clinically indicated, through sheaths placed in the femoral vein. After completion of the diagnostic portion of the test, a quadripolar catheter with 5-mm interelectrode spacing was positioned in the right ventricular apex. A second steerable, quadripolar catheter with 5-mm interelectrode spacing was positioned in the right ventricle in the posteroseptal area for recording of a small atrial depolarization and a large ventricular depolarization. The locations of these two catheters are diagrammed in Fig. 1. Catheter positioning was confirmed in the left and right anterior oblique views. Unipolar recordings (band-pass filters between 0.05 and 400 Hz) were made from the distal electrode of each catheter, with a skin electrode on the thigh as the indifferent electrode. Unipolar stimulation was performed with a programmable stimulator (Bloom Associates, Reading, PA) using a constant-voltage stimulus isolation unit from the second pole of each catheter (5 mm from the recording site).

The stimulation protocol was designed so that a broadband signal, which contains a wide range of frequencies, would be applied during a portion of the cardiac cycle with no spontaneous electrical activity. Additionally, the applied signal did not result in ventricular capture in any case, because it was not in direct contact with the myocardium. Thus simultaneous endocardial recordings from near (5 mm away from) the site of stimulation and surface ECG recordings were required for measurement of the body's attenuation of this signal. These measurements were made from the right ventricular apex, an anterior and apical site, and the posteroseptal area, a posterior and basal site.

A 0.1-ms 1-V rectangular stimulus was first applied from the catheter in the right ventricular apex. The stimulus, which was timed to coincide with the midportion of the ECG TP interval, was applied for 5 min in each TP interval. This test signal was chosen to approximate an impulse, which is a broadband signal. This procedure was then repeated using the catheter positioned in the right posteroseptal area. If the 1-V stimulus caused saturation of the recorded signal, a 0.1-V stimulus was used instead.

During each 5-min pacing period, ECG data from the *x*, *y*, and *z* leads and the unipolar recordings from the distal pole of the catheter from which the stimuli were applied (5 mm away) were recorded on a commercially available system (Predictor I, Arrhythmia Research Technology, Austin, TX) at a sampling frequency of 2,000 Hz and stored on an optical disk for subsequent analysis. In addition, a square-wave synchronous to the pacing stimulus was recorded from the Bloom stimulator. Sampling of 0.1-ms impulses with a 2,000-Hz sampling frequency is possible because of the 0.05- to 400-Hz band-pass filter. The filtered signal has a frequency band well below the Nyquist bound of half the sampling rate needed to sample without loss of information (Fig. 2). The same band-pass filtering was applied to the intracardiac impulse and the surface ECG impulses and, thus, would not affect the transfer function calculation.

#### Signal processing.

Digital signal processing was performed using analysis software developed with MATLAB (Mathworks, Natick, MA). With the positive deflection of the square wave used as the fiducial point, time-based signal averaging of the applied stimulus was performed to reduce noise levels. A precise fiducial point avoids the excessive smoothing of the waveform, which is a limitation of signal averaging of QRS complexes. A 50-ms segment centered around the fiducial point was selected for analysis. Baseline correction was performed using a third-order polynomial trend removal algorithm. With the assumption of a causal system, the first 50 points (25 ms) of the signal were replaced with zeroes. A Hanning window was then applied, and the data were zero padded to 400 points. The frequency domain representation of these signals was calculated using the fast Fourier transform with a frequency resolution of 5 Hz. The *x*, *y*, and *z* transfer functions were defined as the Fourier transform of the respective ECG recording divided by the Fourier transform of the intracardiac recording Thus the magnitude transfer functions were calculated by dividing the magnitude of the Fourier transform of the ECG recording by the magnitude of the Fourier transform of the intracardiac recording The phase transfer functions were calculated by subtracting the unwrapped phase angle of the Fourier transform of the intracardiac recording from the unwrapped phase angle of the ECG recording

#### Data analysis.

For each subject, two sets of three magnitude and three phase transfer functions were generated from the right ventricular stimulation/recordings and the posteroseptal stimulation/recordings. If the signal averaging and other data processing did not adequately reduce noise levels or introduced obvious artifacts, the data for that particular lead were excluded from analysis. As a result, only two patients were excluded from the subsequent analyses. The magnitude transfer function were expressed in decibels and plotted on a logarithmic frequency scale. The phase transfer functions were expressed in radians and plotted on a linear scale.

#### Technique validation.

A simulated setup consisting of a saline (0.9% NaCl) bath was used to test the transfer function of a homogenous system. Standard Ag-AgCl electrodes were also used to record the electrical impulses applied to the saline bath from a quadripolar catheter. The transfer function with the saline bath setup had a magnitude slope of 0.1 dB/decade frequency (*P* = not significant for the linear regression) and only a 2-dB range, demonstrating minimal frequency dependence of attenuation over the 25- to 250-Hz frequency band. The phase slope of only −0.001 rad/Hz (*P* < 0.0001 for the linear regression) indicates negligible conduction delay. The transfer function characteristics thus show that the saline and system (electrodes, Predictor hardware) have primarily resistive effects.

#### Effect of the transfer function on the P wave, QRS complex, and T wave.

Five minutes of continuous, orthogonal-lead ECGs were obtained from 33 healthy volunteers (19 men, 53 ± 7 yr of age) and then signal averaged. With the assumption that the ECGs were recorded from a thorax with transfer functions with mean magnitude slopes, least-square finite-impulse response approximations of the transfer functions that would be required to obtain signals attenuated by the upper and lower bound of the 95% confidence intervals of magnitude transfer function slopes were applied to the signal-averaged ECGs. The transfer function for <25-Hz frequencies was extrapolated from the 25- to 250-Hz band. The transfer function for >250-Hz frequencies was considered band pass. After normalization of the adjusted ECGs to the QRS amplitudes, the P and T wave amplitudes were compared for the low- and high-slope conditions.

#### Statistics.

The data were analyzed with multilevel mixed-effect regression models using HLM 6.0 software (Scientific Software, Lincolnwood, IL). Magnitude and phase functions were evaluated separately. Maximum likelihood estimation was used to obtain slope-and-intercept models for individual patients. These individual slopes and intercepts, in turn, were used to investigate the difference between leads, stimulus location, and the explanatory power of patient-level characteristics listed in Table 1. To explore patient variables, the two sets of recordings from the right ventricular apex and posteroseptal area were combined for each lead. Patient variables were then entered into the regression model in a stepwise fashion. Strongest predictors of TF magnitude, slope, and their difference between the posteroseptal area and right ventricular apex were entered first, followed by additional predictors that had significant partial correlations. If the significance for a variable already in the model dropped below 0.10, it was removed from the final model. Therefore, patient-level variables in the final model represent an optimal subset. *P* < 0.05 was considered significant.

## RESULTS

#### Signal-averaged impulses.

The peak-to-peak amplitudes of the signal-averaged impulses from the right ventricular apex recorded in the *x*, *y*, *z*, and intracardiac leads for the 20 patients ranged from 0.003 to 0.238 mV (0.060 ± 0.062 mV), 0.044 to 0.653 mV (0.250 ± 0.184 mV), 0.002 to 0.347 mV (0.079 ± 0.882 mV), and 0.299 to 1.921 mV (0.800 ± 0.453 mV), respectively (*P* < 0.0003 for *x* vs. *y* and *y* vs. *z*). For signal-averaged impulses from the right ventricular posteroseptal area, *x*, *y*, *z*, and intracardiac peak-to-peak amplitudes were 0.001–0.274 mV (0.062 ± 0.076 mV), 0.062–0.691 mV (0.278 ± 0.187 mV), 0.002–0.197 mV (0.052 ± 0.050 mV), and 0.300–1.890 mV (0.808 ± 0.456 mV), respectively (*P* < 0.0001 for *x* vs. *y* and *y* vs. *z*). Comparison of the amplitude of right ventricular apex impulses with that of the corresponding posteroseptal impulses showed a significant difference only in the *y* lead (*P* < 0.0003 for the paired comparison).

#### Frequency characteristics of the intracardiac signal.

A representative signal-averaged intracardiac impulse and the corresponding signal-averaged surface ECG impulses are shown in Fig. 3. The fast Fourier transform magnitude of this intracardiac signal was 1.25–1.42 mV in the 25- to 250-Hz band, which shows that the test signal is indeed a broadband impulse in this frequency range.

#### Transfer function modeling.

The *x*, *y*, and *z* transfer functions could be calculated for 14, 18, and 14 patients, respectively, for recordings from the right ventricular apex and 14, 18, and 16 patients, respectively, for recordings from the right posteroseptum. The mean transfer functions are displayed in Fig. 4, with the saline transfer function shown for comparison. Linear regression analysis demonstrated an excellent fit, with *R*^{2} = 0.84–0.89 for the mean magnitude transfer functions. The slopes and intercepts for the *x*, *y*, and *z* magnitude functions are shown in Table 2. Thus, over the 25- to 250-Hz frequency range, there was ∼8-dB (∼2.5 times) change in magnitude attenuation.

The *R*^{2} values from linear regression of the phase transfer functions showed excellent fit (*R*^{2} = 0.98–1.0), indicating little to no overall phase distortion. The slopes and intercepts for the *x*, *y*, and *z* phase transfer functions are shown in Table 2. Similar findings were observed in the apical and the posteroseptal recordings.

Table 3 shows the mean, standard deviations, and 95% confidence intervals of the modeled transfer function characteristics. The modeled magnitude transfer function for each lead was summarized with two variables: the intercept at 250 Hz (in dB) and the slope (expressed as dB/decade of frequency in Hz). The modeled phase transfer function was summarized by the slope (in rad/Hz).

#### Comparison of intraindividual x, y, and z transfer functions.

The intraindividual modeled *x*, *y*, and *z* transfer functions were compared by lead and impulse location. The magnitude slopes of the transfer functions in the right ventricular apex recordings were larger in the *z* lead than in the *y* lead (*P* < 0.04 for the paired comparison). There were no significant differences in magnitude slopes among the leads with the posteroseptal recordings. There were no differences in magnitude slopes between the right ventricular apex and posteroseptal recordings.

The magnitude intercepts at 250 Hz showed significant differences between all leads (*P* < 0.003 for all), except between *x* and *z* leads from the posteroseptal recordings (*P* < 0.06). The *y* lead showed an increase in magnitude intercept from the right ventricular apex to the PS (*P* < 0.006), whereas the *z* lead showed a decrease (*P* < 0.001).

#### Effect of clinical characteristics on transfer functions.

Table 4 lists the optimal sets of patient characteristics that affected the magnitude and phase transfer functions. The *R*^{2} values quantifying the percentage of transfer function characteristic that is explained by the patient characteristics that were found to be significant are also shown in Table 4.

Magnitude intercepts were affected by chest circumference, chest height, age, and respiratory measures. The negative coefficients show that increases in each of these variables were associated with smaller transfer function intercepts and, in turn, greater attenuation of the electrical signals. Magnitude slopes were affected by height, age, and respiratory measures. Increases in FEV_{1} were associated with more positive slopes in the magnitude transfer functions (or more attenuation of low frequencies), whereas greater heights and age were associated with a smaller slope.

#### Effect of the transfer function on the P wave, QRS complex, and T wave.

To examine the effects of the transfer function on the surface ECG, the differences in magnitude slopes as displayed in Table 3 were modeled and applied to the signal-averaged ECGs of the 33 subjects. The upper and lower bounds of the 95% confidence intervals are shown in Table 3. Figure 5 shows an ECG before and after application of the low-magnitude slope and high-amplitude slope correction and normalization to QRS amplitude. The high-magnitude slope correction showed a significant relative attenuation of the T wave. The low-magnitude slope correction resulted in an amplified T wave. P waves in this example did not show significant change. The overall effect of the high and low slope corrections on the 33 subjects is shown in Table 5.

## DISCUSSION

The amplitude of cardiac potentials as recorded on the surface ECG is considerably less than that of the potentials generated by the heart. This attenuation is caused in part by the high impedance of the thorax and in part by cancellation of activation wavefronts propagating in opposite directions. If the thorax is purely resistive, the attenuation due to impedance would simply result in differences in scaling. Even with the assumption of a heterogeneously conductive thorax, the waveform morphologies would remain unaltered, with only amplitude being affected. However, if significant capacitive elements are present in the system, frequency dependence of attenuation will be present as well. Our study found significant frequency dependence in the transfer function, which confirms that the thorax does indeed have resistive and capacitive effects in humans. The findings that the transfer functions have a wide range of characteristics and are affected by body characteristics also have significant implications, since they imply that body type can affect the morphology of the surface ECG waveform.

It is worthwhile to consider from first principles why the thoracic impedance should have a significant capacitive effect. The voltage generated on the surface of the heart, which is ultimately recorded on the body surface as the ECG, creates current flow in the thoracic tissues. The thoracic tissues have an extracellular space and the cellular space through which current may flow. The extracellular space allows for free flow of ions and is essentially purely resistive. On the other hand, cell membranes have a capacitance of ∼1 μF/cm^{2} (3). Current flow through the cellular space will therefore have resistive and capacitive attenuation. The relative impedance to current flow in the extracellular and cellular spaces will determine whether the thorax has predominantly resistive attenuation or resistive and capacitive attenuation.

The most comprehensive study of the resistivity and capacitivity of living tissue was performed by Schawn and Kay more than 50 years ago (13, 14). In their studies, impedances, dielectric constants, and the ratios of capacitive to resistive currents were measured in canine lung, muscle, liver, heart muscle, and fatty tissue. Impedance was shown to be higher at low frequencies than at higher frequencies. Higher dielectric constants were found at low frequencies. The ratio of the capacitive to resistive effects was ∼0.15 at 10 Hz and 0.035 at 100 Hz, except in fatty tissue, in which it was significantly smaller. They concluded that “it is justified to a good approximation… to consider body tissues as a resistive medium.”

Rudy et al. (11) showed altered QRS amplitudes and morphologies in patients undergoing bronchopulmonary lavage after their lungs were filled with saline. In this study, air, which does not conduct currents, was replaced with saline, providing a large volume of low-resistance medium. Rudy et al. noted a change in QRS morphology that was attributed to a change in the vectorcardiogram. Careful evaluation of the figures demonstrates changes in the ST-T wave morphology and timing. Specifically, changes in the timing of the peak of the T wave can be appreciated in some of the examples. Although this may also be due to differences in heart rate, the data could support the notion that altering the thoracic impedance by the addition of saline changed the ratio of resistive to capacitive effects, thereby altering the frequency-dependent attenuation. Thus visualization of frequency-dependent attenuation in this study supports significant capacitive effects on the human thorax.

Comparison of the transfer characteristics of the three surface lead recordings showed that the *y* lead had magnitude transfer function slopes with significantly higher intercepts at 250 Hz. A similar difference was seen in the peak-to-peak time domain amplitudes. The extent to which the distances of the ECG from the source are responsible for the magnitude differences in the *x*, *y*, and *z* leads is not clear. However, the change in stimulus location from the right ventricular apex to the posteroseptal area did produce expected changes in the *z* lead, inasmuch as the magnitude intercept decreased as the catheter was moved away from the chest wall.

Although the magnitude intercept was shown to be very dependent on the leads and location of the impulse, the same level of differences was not seen for the magnitude slopes. This finding suggests that the thoracic tissue most responsible for the electrical attenuation is primarily uniform in composition in all directions, thus resulting in similar frequency responses, despite differences in overall attenuation.

The linearity of the phase transfer functions showed little phase distortion for the frequency range studied. The positive regression slope that was computed was not expected, inasmuch as it signifies a negative delay of the surface recording relative to the intracardiac recording. Because of the proximity of the intracardiac recording electrode to the stimulus electrode, a smaller time delay from this signal than from the surface-recorded signal should have been expected. This discrepancy may have resulted from the comparison of the unipolar intracardiac recording with bipolar surface lead recordings, which may introduce some nonlinear components. In Fig. 3, the change in morphology from the unipolar recording, which has a greater second deflection, to the bipolar surface recordings, which have larger initial deflections, can be seen. In any case, the magnitude characteristics of the surface ECG signals in the frequency domain are maintained.

Analysis of the patients' clinical characteristics showed that physical dimensions, respiratory measures, and age affected the transfer function characteristics. Other studies using signal-averaged ECG have shown the effect of patient characteristics on QRS amplitude and duration (4–7, 9, 15). However, these population-based studies have the limitation of cardiac activation being widely variable between patients, and thus the transfer functions specific to the patients could not be obtained.

The results of the present study have several implications for traditional ECG analysis. The key features in the ECG, which include the P wave, QRS complex, ST segment, and T wave, each have specific frequency ranges that contain the most information among the respective signals. The utility of analyzing the timing, amplitude, and morphology of these features has been well demonstrated. In particular, the prognostic value of the QT interval has been well documented. As shown in this study, the P wave, QRS complex, and T wave are subject to different attenuation, and these differences may depend on the body characteristics of the patient. Greater attenuation of the T wave could affect the detection of the T wave offset and, thus, affect QT interval measurement. Thus body characteristics, and not just cardiac electrophysiology, may affect QT interval measurements.

The results also have implications for the research in “forward” and “inverse” problems of electrocardiology. The forward problem of electrocardiology attempts to estimate the surface body potentials on the basis of the known potentials detected from the epicardium of the heart (1). The inverse problem attempts to estimate the cardiac source potentials from body surface mapping (8, 10). The solution for the forward problem is unique, whereas the solution for the inverse problem is not. The study of both problems has in common the assumption that the tissue is mainly resistive with negligible capacitance.

#### Limitations.

The present study was designed to introduce a new method to obtain patient-specific transfer functions of the thorax in humans. Further study is needed to test the clinical applicability of the technique, as well as to further understand the effects of stimulus location and thorax characteristics on the transfer function. Finally, the total variance in the ECG due to body characteristics will need to be determined to assess whether “correction” for thorax characteristics can improve the diagnostic utility of the ECG.

The recorded intracardiac signal was near, but not at, the site of stimulation. This would likely affect the quantitative assessment of attenuation; however, the qualitative assessment should remain valid if it is assumed the blood impedance is purely resistive. The frequency bands outside the 25- to 250-Hz range were not analyzed. Although frequencies <25 Hz are of importance in ECG analysis, the signal-to-noise ratio was not adequate for the computation of the transfer functions in this band. The frequencies >250 Hz are not important in most applications. Nonlinear effects of the thoracic attenuation were not studied. The mechanisms for the observed capacitive effects and the actual ratio of capacitance to resistive currents could not be determined from the present study.

#### Conclusions.

A new method to obtain patient-specific thoracic transfer functions was described. The frequency dependence of signal attenuation implies that the thorax has significant capacitive, as well as resistive, effects. Inasmuch as analyses of surface ECG recordings are constantly being developed to provide increasing information regarding cardiac electrical activity, the effects of thoracic attenuation/distortion of the cardiac signals must be considered. Accounting for the effects of thoracic attenuation/distortion of cardiac signals recorded on the body surface and the influence of body characteristics may further enhance the diagnostic utility of such recordings.

## Footnotes

The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “

*advertisement*” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

- Copyright © 2007 by the American Physiological Society