Noninvasive ECG Imaging of Electrophysiologically Abnormal Substrates in Infarcted Hearts
A Model Study
Background—Myocardial infarction and subsequent remodeling create substrates with altered electrophysiological (EP) properties that are highly arrhythmogenic. Existing ECG methods cannot always detect the existence of such substrates nor provide any detailed information about their EP characteristics. A noninvasive method with such capabilities is greatly needed for identifying patients at risk of arrhythmias and for guidance and evaluation of therapy. Recently, we developed a noninvasive ECG imaging modality that can reconstruct epicardial EP information from body surface potentials. We extended its application to hearts with structural disease and examined its ability to detect and characterize abnormal EP substrates.
Methods and Results—Epicardial potentials were recorded with a 490-electrode sock from an open-chest dog. Recordings were obtained from a normal heart and from the same heart 2 hours after left anterior descending coronary artery occlusion and ethanol injection to create an infarct. Body surface potentials were generated from these epicardial potentials in a human torso model. Realistic geometry errors and measurement noise were added to the torso data, which were then used to noninvasively reconstruct epicardial potentials and electrograms (EGMs), with excellent accuracy. EP characteristics associated with the infarct substrate were reconstructed, including (1) a negative region over the infarct, (2) EGMs with large predominant negative deflections (eg, Q-wave EGMs), (3) Q-wave EGMs with superimposed RS deflections reflecting local activation of surviving myocardium within the infarct border zone, (4) reduced magnitudes of EGM negative derivatives, and (5) negative QRS integrals of EGMs over the infarct.
Conclusions—ECG imaging can noninvasively detect and map abnormal EP substrates associated with infarction and structural heart disease.
Myocardial infarction and subsequent remodeling creates altered electrophysiological (EP) substrates that are highly arrhythmogenic.1 Experimental and clinical data from infarcted hearts have helped to define measures for identifying abnormal EP substrates2 3 4 5 and for stratifying their arrhythmogenic potential.6 7 Identification of arrhythmogenic substrates before an arrhythmia occurs could reduce the risk of sudden death by indicating the need for device, drug, or ablation therapy. Current noninvasive methods for assessing abnormal EP substrates rely on ECG measurements from the body surface.8 These methods cannot provide detailed EP information about the cardiac substrate. ECG imaging (ECGI) is a noninvasive method for reconstructing EP information on the heart surface from body surface measurements. Studies from our laboratories9 10 11 demonstrated that ECGI can accurately reconstruct epicardial potentials, electrograms (EGMs), and isochrones in normal hearts. The purpose of this study was to evaluate the ability of ECGI to noninvasively locate and characterize abnormal cardiac EP substrates associated with myocardial infarction.
A combination of experimental and modeling methods were used to evaluate ECGI with the use of measured epicardial potentials in an accurate heart-torso model. Figure 1A⇓ shows the strategy used. Epicardial potential maps were recorded from a single anesthetized dog. The heart was exposed through a sternotomy, and a sock with 490 unipolar silver electrodes was pulled over the heart. Unipolar signals were recorded with respect to the left leg. The system simultaneously sampled 490 electrodes at 1 kHz with 12-bit resolution and a frequency response of 0.03 to 500 Hz. Potentials were recorded during anterior left ventricular (LV) epicardial pacing and right atrial (RA) pacing (simulating sinus rhythm). Stimulation was by 2-ms current pulses at twice diastolic threshold. A region of infarcted/necrotic tissue was then created by ligating the left anterior descending coronary artery (LAD) distal to the first diagonal branch (Figure 1B⇓). A needle was inserted into the LAD and 2 mL of a 50:50 ethanol/saline solution was injected. The estimated area perfused by the LAD is outlined in Figure 1B⇓, showing the expected region of infarction created by the combined arrest of blood flow and necrotizing effect of ethanol. This protocol yields a border-zone infarct similar in dimensions to the 5-day-old LAD occlusion model.12 A 10- to 20-mV ST elevation was observed over this region after occlusion-injection, which fell to 3 to 4 mV after 2 hours. The same pacing protocols as in the preinfarction heart were repeated 2 hours after occlusion-injection, when the infarct was well developed. As in our previous studies,9 10 11 ECGI reconstructed potentials were evaluated by direct comparison to measured epicardial potentials in the same heart.
The mathematics of ECGI and modeling methods used here have been described previously.9 13 Importantly, computation of epicardial potentials from body surface potentials using the ECGI method only requires information that can be obtained noninvasively. In this study, epicardial potentials at all 490 epicardial points were noninvasively computed at 1-ms increments throughout the cardiac cycle. Potentials were displayed in space for a single instant of time to create epicardial potential maps and in time for a single point in space to create epicardial EGMs. The general quality of reconstructed potential maps and EGMs was evaluated with respect to measured data using absolute error (AE), relative error (RE), and correlation coefficient (CC) (RE and CC were defined previously).13 The AE is an average of absolute differences between measured and reconstructed potentials at each point in space for potential maps and each point in time for EGMs. The average AE, RE, and CC presented for potential maps were averaged over all reconstructed time frames. The average AE, RE, and CC presented for EGMs were averaged over all 490 epicardial points. Epicardial isochrones were created with the use of epicardial activation times determined as the time of maximum negative derivative (−dV/dtmax) in the EGMs. QRS integrals were computed for each EGM by summing the potentials over the QRS window determined using lead II.
Potentials recorded from the open-chest heart were placed on a digitized 3-D stylized canine heart model and aligned on the basis of positions of coronary arteries (eg, LAD). The heart model was mathematically placed in its anatomic position within a homogenous computer model of the human torso, constructed using 458 nodes. Potentials on the torso surface were computed from the measured epicardial potentials (Figure 1A⇑).13 The reference of the computed torso surface potentials was converted from the left leg to a Wilson central terminal. A 12-lead ECG was derived with the use of torso potentials from the standard surface positions shown in Figure 1C⇑. Fifty-microvolt peak-to-peak gaussian noise (typically 0.5% of the signal) was added to the computed torso potentials and 1 mm-gaussian geometric error was added to the torso surface points. We based the levels of signal noise and geometric error on previous, detailed studies13 and our experience with clinical data acquisition. The clinical ECGI procedure currently being evaluated at Case Western Reserve University obtains body surface electrode positions with the use of a 3-D digitizer with better than 1-mm accuracy. Simulations in which added signal and geometric noise are both doubled result in only modest increases (on the order of 10%) in RE and AE and minimal reductions in CC, reflecting the effectiveness of the regularization procedure used in the ECGI reconstructions.13 The computed torso potentials, contaminated by measurement noise and position errors, were then used as inputs to noninvasively compute epicardial potentials with the use of the ECGI methodology.
The combination of the measured epicardial potentials and the heart-torso model used in this study provides a well-controlled tool for evaluating the ability of ECGI to noninvasively reconstruct epicardial potentials, EGMs, and isochrones in an infarcted heart. The directly measured preinfarction and postinfarction epicardial potential maps provide physiologically realistic data for evaluation of the noninvasively reconstructed epicardial potentials. The use of ethanol results in fast infarct formation, allowing for the infarcted heart to be its own control. These features facilitate a rigorous evaluation of ECGI in the context of electrophysiologically abnormal substrates in the presence of structural heart disease.
Standard 12-Lead ECG
Figure 2⇓ shows a standard 12-lead ECG derived from the computed torso potentials. The 12-lead ECG from the preinfarction heart (top panel) serves as control and illustrates that the model generates clinically realistic data on the torso surface. The ECG is consistent with that of a normal subject, as characterized by several properties such as the progression from negative in V1 and V2 to isoelectric in V3 to positive in V4 to V6. Precordial lead V3 exhibits a notch that reflects a similar notch in the epicardial EGM near the site of right ventricular breakthrough (RVBT, arrival of the activation front at the epicardium and beginning of epicardial activation). The 12-lead ECG generated by the infarcted heart (bottom panel) shows the presence of myocardial infarction when compared with the control ECG in panel A. ECG features indicating infarction include the appearance of a Q wave in lead II and deep pathologic Q waves in V3 and V4. Close examination of the computed ECGs reveals minor differences compared with ECGs observed in humans (eg, the positivity in aVR). These differences can be explained by differences in anatomy between the human and canine hearts and their different orientation within the torso.
Epicardial and Torso Potential Maps During LV Pacing
Figure 3⇓ (top row) shows torso potentials, measured epicardial potentials, and noninvasively reconstructed epicardial potentials for the preinfarction heart at 30 ms and 200 ms after pacing. The bottom row shows corresponding potential maps from the infarcted heart. Paced activation in the control heart (30 ms) produces the expected elliptical region of large-magnitude negative potentials (white) around the pacing site. This elliptical pattern is aligned with the local epicardial fiber direction.14 A similar pattern (light gray) is observed in the measured potentials from the infarcted heart. However, although the potential pattern is similar to that observed in the control heart, the magnitudes of the potentials are reduced by >50% in the presence of the infarction in the underlying myocardium. The noninvasively reconstructed epicardial potentials (obtained with the use of the ECGI methodology) closely approximate the directly measured potentials in both the control and infarcted hearts. The ECGI-reconstructed potentials capture the elliptical negative region around the pacing site and its orientation along the direction of epicardial fibers. The 50% reduction of potential magnitudes over the infarct is also captured noninvasively by ECGI. Potential maps measured during repolarization (200 ms) show the expected pattern of an elliptical positive region (dark gray) in about the same position as the initial negative elliptical region associated with early activation (30 ms, white). This polarity reversal is characteristic of excitation during ventricular pacing. Repolarization potentials from the same time frame in the infarcted heart show a pattern similar to the control heart, but with reduced magnitudes associated with the presence of the infarct. The noninvasively reconstructed epicardial potentials for both the control and infarcted hearts during repolarization clearly show similar patterns to the directly measured epicardial potentials, including magnitude reduction in the presence of the infarct. Errors for the 250 reconstructed time frames in the control and infarcted hearts are presented in the Table⇓. Consistent with previous results from our laboratory,11 the average error observed throughout depolarization is similar to the error observed throughout repolarization. This raises the possibility of using ECGI to investigate repolarization properties such as activation recovery intervals.
Epicardial EGMs and Isochrones During LV Pacing
Epicardial EGMs at 4 selected sites are presented in Figure 4⇓. Site A is located on the right ventricle (RV), sites B and C are over the infarcted region of the LV, and site D is over a necrosis free region of the basal LV. Directly measured and noninvasively reconstructed EGMs are shown for the control (left panels) and infarcted (right panels) heart. EGMs at sites A and D show little change after infarct formation because they lie outside the region perfused by the LAD. The EGM from site B (located over the region of infarction) shows a 58% reduction in negative peak magnitude after infarction. The LV EGM at site C, also located directly over the infarcted tissue, has an RS morphology in the control heart that is altered to a Q-wave morphology by the presence of the infarct. All EGM morphologies are reconstructed correctly in both the control and infarcted heart. The Table⇑ summarizes average errors for the 490 reconstructed EGMs in the control and infarcted hearts. In general, all EGMs around the heart are accurately reconstructed, with anterior EGMs having slightly higher CCs than posterior EGMs.
Epicardial isochrones from the LV paced control heart (Figure 5A⇓, top) show the elliptical activation pattern aligned with the epicardial fibers that is typical of epicardially paced beats. Isochrones from the infarcted heart (Figure 5A⇓, bottom) have the same elliptical pattern and orientation as in the control heart, only with a lower density of isochrones, indicating an increased propagation velocity away from the pacing site. The faster conduction could be due to reduced electric load from intramural myocardium on the epicardial wave front in the infarcted heart. Isochrones constructed from noninvasively reconstructed epicardial potentials, in both the control and infarcted hearts, closely match the directly measured isochrones. The general activation patterns and the increased conduction velocity observed in the infarcted heart are accurately reproduced.
Epicardial and Torso Potential Maps During RA Pacing
The ability to noninvasively detect, locate, and characterize an abnormal EP substrate during sinus rhythm is of important clinical significance because the information can be used to identify patients at risk of developing life-threatening arrhythmias and to guide strategies for treatment. Results during RA pacing are presented in detail below to demonstrate the ability of ECGI to noninvasively reconstruct abnormal infarct-related EP properties during simulated sinus rhythm. Figure 6⇓ shows potential maps from 48 ms (left panels) and 65 ms (right panels) after pacing. The 48-ms map shows the localized minimum (white) on the RV associated with RVBT. RVBT occurs in both the control and infarcted hearts because LAD occlusion leaves the tissue and activation sequences of the RV unaltered. Comparison of the LV epicardial potentials from the control heart with those measured from the infarcted heart at 48 ms shows the formation of a large region of negative potentials over the infarcted LV (light gray). These negative potentials are a result of the underlying electrically inactive necrotic tissue, which cannot support normal spread of activation from endocardium to epicardium. Activation fronts associated with such spread generate positive potentials on the overlying epicardium.15 In their absence, a major positive component is subtracted from the potentials over the infarct, shifting the balance of potentials in the negative direction. The contribution of negative potentials is probably from more remote (“far-field”) activation fronts propagating away from the infarct. The large negative area over the infarct is reflected in the corresponding torso potentials as an anterior region of negativity. Notice how the torso surface RVBT minimum is completely hidden in this negative region despite its presence on the RV epicardium (48 ms, infarct, measured). Despite this lack of spatial resolution on the torso surface, ECGI successfully reconstructs from the torso potentials both the extensive negative potential region over the LV infarct and the RVBT epicardial minimum (48 ms, infarct, reconstructed). The zero-potential line that delineates the LV negative and positive regions closely resembles the directly measured maps. The RVBT epicardial minimum is also successfully reconstructed for the control heart, in the absence of infarction (48 ms, control, reconstructed).
At 65 ms, the LV epicardial potential map of the control heart shows an island of large positive potentials, probably reflecting late activation of this LV free wall region. In contrast, the region of positive potentials does not form over the infarcted LV, which remains negative. Note that the torso potentials reflect the negative epicardial potentials associated with the infarcted heart (65 ms, infarct), whereas in the control maps, they contain a positive pseudopod that reflects the positive epicardial region (arrow in torso, control). However, the torso potentials are smooth and low in resolution, they do not reflect details of the epicardial potentials, nor do they provide information regarding the location of potential features on the heart. ECGI accurately reconstructs the positive region over the control heart LV and the negative region over the infarcted heart LV. The Table⇑ summarizes errors from RA-paced epicardial potential map reconstructions.
Epicardial Electrograms During RA Pacing
Figure 7⇓ illustrates EGMs from 4 representative sites. Site A is located on the RV, sites B and C are located over the infarcted region of the LV, and site D is over the necrosis-free region of the anterior basal LV. EGMs from the control heart (column 1) show the expected RS morphology associated with a sinus beat in a normal heart and contain a sharp intrinsic deflection (down slope, arrow in A), which indicates local activation at the electrode site. These EGMs are noninvasively reconstructed with the same RS morphologies and sharp intrinsic deflections as with noninvasive ECGI (column 2). The −dV/dtmax of the intrinsic deflection at site C is −6.5 mV/ms (measured) and −6.0 mV/ms (noninvasively reconstructed). The EGMs recorded from the infarcted heart (column 3) show changes from control EGMs in regions overlying the infarct. The EGM recorded at site B is drastically altered by the presence of the infarct, with the disappearance of the R wave and appearance of a large Q wave with a superimposed sharp RS wave. EGMs with large Q waves are known to be associated with recordings over necrotic tissue and can generally be explained by the necrotic tissue interfering with the spread of the wave front toward the recording electrode (a process that generates the positive R-wave deflection), thereby leaving the electrode to record mostly far-field activity of wave fronts spreading away from the infarct. The superimposed RS wave reflects local activation of a small region of surviving myocardium within the infarct substrate, close to the recording electrode. Notice that the noninvasively reconstructed EGM in column 4 reproduces the measured Q-wave EGM morphology with the superimposed local RS deflection, demonstrating that ECGI can noninvasively detect local activation of surviving tissue within the infarcted substrate. The morphology of the EGM recorded at site C reflects its location at the border of the infarcted myocardium. The small initial R wave can be attributed to viable tissue bordering the infarct. After the small R wave is a large-magnitude, slow negative deflection (−dV/dtmax=−1.5 mV/ms), indicating reflection of far-field activity rather than of local activation. The small sharp downstroke observed near the negative peak may reflect local activation of local viable tissue within the infarct border zone. This EGM is noninvasively reconstructed with the use of ECGI, including the slow downward deflection (−dV/dtmax=−1.8 mV/ms) and the small sharp deflection caused by local activation. Sites A and D show no change in EGM morphology or amplitude as a result of the infarct development because they are situated over regions of the heart that were not affected by LAD occlusion and ethanol injection. These EGMs are noninvasively reconstructed (column 4) to have the same RS morphology and amplitude as the directly measured EGMs. The Table⇑ summarizes errors associated with reconstructed EGMs during RA pacing.
Noninvasive Localization of Electrophysiologically Abnormal Substrate
The ability of ECGI to noninvasively identify the region of electrically altered myocardium caused by infarction was evaluated during RA pacing (simulating sinus rhythm) through the use of several EP properties associated with infarcted hearts. Laxer et al2 proposed a method for determining infarct location that uses EGMs with Q-wave magnitudes >2 mV to indicate the presence of infarcted myocardium below the recording electrode. Application of this criterion and its extension to include EGMs containing a predominant slow negative deflection after a small R-wave yield the regions shown in Figure 8A⇓, marked for both measured (gray lines) and noninvasively reconstructed (black lines) data. Note how the regions estimated using the measured and noninvasively reconstructed data closely resemble each other. A second method for estimating the location and extent of abnormal substrate is based on the observation that during RA paced beats, negative epicardial potentials cover the infarcted region during most of the cycle. The regions outlined in Figure 8B⇓ show areas where the potential was mostly negative, with maximum positive potential <5 mV. Again, the measured and noninvasively reconstructed regions closely resemble each other. A third criterion for estimating the region of altered substrate is a large negative QRS integral (<−200 mV · ms) in EGMs overlying the infarct. The QRS integral (unlike the large Q-wave criterion) considers not only the Q wave but also the superimposed RS deflection caused by local activation when it is present. Figure 8C⇓ demonstrates close agreement between the noninvasively (black line) and invasively (gray line) estimated region. Importantly, comparison of the anatomically estimated region of infarction in Figure 1B⇑ to the regions of abnormal EP substrate estimated with the use of the 3 different criteria of Figure 8⇓, A through C, shows good correlation.
This study is our first demonstration of the application of ECGI in the presence of structural heart disease. Although the model used here is that of myocardial infarction, a similar approach can be applied in the more general context of altered EP substrate caused by EP remodeling. Myocardial infarction (and other remodeling processes) produces a substrate that can be highly arrhythmogenic.1 It is extremely important to identify the existence of such EP substrate in patients before they have a major arrhythmic event that can lead to sudden cardiac death. ECG late potentials,8 presumably reflecting slow conduction in such a substrate, have been suggested as a noninvasive identifier of patients at risk. Unfortunately, this measure found very limited application because of low sensitivity and specificity and difficulties in interpretation of such low level signals that are measured far away from the heart.
The results demonstrate the ability of ECGI to noninvasively reconstruct, from body surface potentials, characteristics of epicardial potentials, EGMs, and isochrones that are associated with abnormal EP substrates. Through these reconstructions, the existence of such a substrate can be identified and its location and extent can be determined (Figure 8⇑). The characteristics include (1) a large region of negative epicardial potentials over the infarct, (2) EGMs with large predominant negative deflections (eg, Q-wave EGMs),2 6 (3) Q-wave EGMs with superimposed RS deflections reflecting local activation of surviving myocardium,2 6 (4) reduced magnitude of the EGM maximum negative derivatives, and (5) large negative QRS integrals. The ability of ECGI to noninvasively provide detailed EP information on the epicardial surface could be extremely helpful to clinical electrocardiographers who currently rely on interpretation of the body surface ECG for noninvasive diagnosis. The quantitative inverse solution provided by ECGI will permit identification and localization of arrhythmogenic substrates in the heart.
The ability of ECGI to noninvasively reconstruct small RS deflections on Q-wave EGMs over the infarct during sinus rhythm (Figure 7⇑) has important clinical and mechanistic implications. The small, sharp RS deflection reflects local activation of an island of surviving tissue in the infarct border zone. Healing or healed infarcts tend to be patchy and nonuniform, with necrotic regions separating surviving myocardium. The structural nonuniformities create local mismatches of electrical properties, such as abrupt increases in electrical loading, where narrow myocardial strands merge into larger islands of surviving tissue. Such structures are highly susceptible to the development of unidirectional block and reentry. Page et al6 have shown that VT inducibility is correlated with the presence of islands with pure Q-wave EGMs bridged by regions of EGMs with late local deflections. These observations are consistent with the importance of surviving cardiac fibers traversing the infarct to the development of VT in the infarcted, Langendorff-perfused human heart.4 The demonstrated ability of ECGI to reconstruct Q-wave EGMs with superimposed local RS deflections and EGMs with small R-waves followed by predominant slow negative deflections (Figure 7⇑) suggests the possibility of noninvasive characterization and mapping of regions with different EP properties within the infarct border zone. The degree of EP heterogeneity and “patchiness” of the substrate could provide a measure of the vulnerability to arrhythmias and help stratify patients at risk. Other potential applications associated with the capability of ECGI to noninvasively map the abnormal substrate could include (1) optimization of implantable cardioverter-defibrillator lead placement, (2) identification of potential sites for ablation, and (3) evaluation of drug therapy through noninvasive examination of its effects on the EP substrate.
The experimental strategy in this study is to use epicardial potentials recorded in situ to compute torso potentials, which, after contamination by realistic measurement and geometrical errors, serve as the input data for ECGI reconstructions. This approach has advantages and limitations. In this preparation, the preinfarction and postinfarction epicardial potentials are measured from a working heart subject to autonomic innervation and is therefore physiologically realistic. The high-density, 490-electrode epicardial sock provides high spatial resolution for capturing spatial variations such as regions with or without local RS EGM deflections. The high-density measured epicardial potentials also serve as high-resolution data for a direct detailed evaluation of ECGI performance. The use of ethanol injection to create the infarction allows the heart to serve as its own control. This approach provides preinfarction and postinfarction potentials from the same heart during the same experiment, revealing the pathological EP changes associated with infarction. A limitation of the in situ approach is that the torso and epicardial potentials were not simultaneously recorded as was done in our previous torso tank studies.9 10 11 In fact, the epicardial potentials were recorded in an open-chest dog and then used to compute potentials in a human torso model. It should be pointed out, however, that this model is self-consistent. In other words, the computed torso potentials are the same as those that would have been generated in a human torso by a heart with the same epicardial potentials. The difference is that the computed torso potentials, unlike their measured counterpart, are not contaminated by noise and measurement error. The addition of signal and geometric gaussian noise to all torso points overcomes this limitation. It should also be added that the use of a homogenous torso volume (eg, no lungs) does not affect the conclusions of this study. Previous studies have demonstrated that torso inhomogeneities affect epicardial potential magnitudes but have a minimal effect on epicardial potential patterns, EGM waveforms, or isochrones.15 16
This study was supported by NIH-NHLBI grants R37-HL-33343, R01-HL-49054 (Y.R.), and R01-HL-43276 (B.T.). Additional support was provided by awards from the Whitaker Foundation (Y.R.), the Nora Eccles Treadwell Foundation, and the Richard A. and Nora Eccles Harrison Fund for Cardiovascular Research (B.T., R.S.M.).
- Received March 31, 1999.
- Revision received August 9, 1999.
- Accepted August 16, 1999.
- Copyright © 2000 by American Heart Association
Janse MJ, Wit AL. Electrophysiological mechanisms of ventricular arrhythmias resulting from myocardial ischemia and infarction. Physiol Rev. 1989;69:1049–1169.
Hatala R, Savard P, Tremblay G, Page P, Cardinal R, Molin F, Kus T, Nadeau R. Three distinct patterns of ventricular activation in infarcted human hearts: an intraoperative cardiac mapping study during sinus rhythm. Circulation. 1995;91:1480–1494.
Watabe S, Taccardi B, Lux RL, Ershler PR. Effect of nontransmural necrosis on epicardial potential fields: correlation with fiber direction. Circulation. 1990;82:2115–2127.
Klein H, Karp RB, Kouchoukos NT, Zorn GL Jr, James TN, Waldo AL. Intraoperative electrophysiologic mapping of the ventricles during sinus rhythm in patients with a previous myocardial infarction: identification of the electrophysiologic substrate of ventricular arrhythmias. Circulation. 1982;66:847–853.
Kuchar DL, Thorburn CW, Sammel NL. Late potentials detected after myocardial infarction: natural history and prognostic significance. Circulation. 1986;74:1280–1289.
Oster HS, Taccardi B, Lux RL, Ershler PR, Rudy Y. Noninvasive electrocardiographic imaging: reconstruction of epicardial potentials, electrograms, and isochrones and localization of single and multiple electrocardiac events. Circulation. 1997;96:1012–1024.
Oster HS, Taccardi B, Lux RL, Ershler PR, Rudy Y. Electrocardiographic imaging: noninvasive characterization of intramural myocardial activation from inverse-reconstructed epicardial potentials and electrograms. Circulation. 1998;97:1496–1507.
Messinger-Rapport BJ, Rudy Y. Noninvasive recovery of epicardial potentials in a realistic heart-torso geometry: normal sinus rhythm. Circ Res. 1990;66:1023–1039.
Dillon SM, Allessie MA, Ursell PC, Wit AL. Influences of anisotropic tissue structure on reentrant circuits in the epicardial border zone of subacute canine infarcts. Circ Res. 1988;63:182–206.
Taccardi B, Macchi E, Lux RL, Ershler PR, Spaggiari S, Baruffi S, Vyhmeister Y. Effect of myocardial fiber direction on epicardial potentials. Circulation. 1994;90:3076–3090.
Rudy Y. The electrocardiogram and its relationship to excitation of the heart. In: Sperelakis N, ed. Physiology and Pathophysiology of the Heart. Boston, Mass: Kluwer Academic Publishers; 1995:210–239.
Green LS, Taccardi B, Ershler PR, Lux RL. Epicardial potential mapping: effects of conducting media on isopotential and isochrone distributions. Circulation. 1991;84:2513–2521.