Patient-Specific Computational Analysis of Transvenous Defibrillation: A Comparison to Clinical Metrics in Humans

Transcription

1 Annals of Biomedical Engineering, Vol. 32, No. 6, June 2004 ( 2004) pp Patient-Specific Computational Analysis of Transvenous Defibrillation: A Comparison to Clinical Metrics in Humans DANIEL MOCANU, 1 JOACHIM KETTENBACH, 2 MICHAEL O. SWEENEY, 3 RON KIKINIS, 2 BRUCE H. KENKNIGHT, 4 and SOLOMON R. EISENBERG 1 1 Department of Biomedical Engineering, Boston University, 44 Cummington Street, Boston, MA; 2 Surgical Planning Laboratory, Brigham and Women s Hospital, 75 Francis Street, Boston, MA; 3 Cardiac Pacing and Implantable Device Therapies, Brigham and Women s Hospital, 75 Francis Street, Boston, MA; and 4 Heart Failure Research, Guidant Corporation, St. Paul, MN (Received 3 March 2003; accepted 29 January 2004) Abstract The goal of this study is to assess the predictive capacity of computational models of transvenous defibrillation by comparing the results of patient-specific simulations to clinical defibrillation thresholds (DFT). Nine patient-specific models of the thorax and in situ electrodes were created from segmented CT images taken after implantation of the cardioverter-defibrillator. The defibrillation field distribution was computed using the finite volume method. The DFTs were extracted from the calculated field distribution using the 95% critical mass criterion. The comparison between simulated and clinical DFT energy resulted in a rms difference of 12.4 J and a 0.05 correlation coefficient (cc). The model-predicted DFTs were well matched to the clinical values in four patients (rms = 1.5 J;cc=0.84). For the remaining five patients the rms difference was 18.4 J with a cc = These results suggest that computational models based soley on the critical mass criterion and a single value of the inexcitability threshold are not able to consistently predict DFTs for individual patients. However, inspection of the weak potential gradient field in all nine patients revealed a relationship between the degree of dispersion of the weak field and the clinical DFT, which may help identify high DFT patients. Keywords Transvenous defibrillation, Critical mass, Patientspecific, Modeling, Finite volume method. INTRODUCTION Ventricular fibrillation (VF) is a severe heart arrhythmia that can lead to sudden cardiac death if not treated promptly. The only effective clinical intervention to extinguish VF is electrical defibrillation. The implantable cardioverter defibrillator (ICD) is an electronic device designed to detect the onset of VF and to shock the heart back into normal sinus rhythm. ICDs have been shown to be very effective in protecting against sudden cardiac death and have become the treatment of choice for patients with drug-resistant arrhythmias. In the standard configuration, the ICD is surgi- Address correspondence to Solomon R. Eisenberg, Sc. D., Department of Biomedical Engineering, Boston University, Boston, MA Electronic mail: sre@bu.edu 775 cally implanted subcutaneously in the patient s chest, with two catheter electrodes inserted in the superior vena cava (SVC) and the right ventricle (RV) (Fig. 1). Typically, the shock energy is delivered via a dual-current pathway from the RV electrode to the SVC electrode and the metallic enclosure (CAN) of the ICD. Determining energy settings during ICD implantation requires repetitive sequences of induction and termination of VF to find the defibrillation threshold (DFT) energy. 15 Since each VF induction has some element of risk, including the possibility of nonconversion and death, the ICD implantation procedure would be substantially improved if an accurate estimate of the patient s DFT energy requirement was availabe prior to implant. Anatomically realistic computational models have been used previously to model and explore the current distribution produced by defibrillation shocks. 1,5,11,17 Jorgenson et al. 10 tested the accuracy of simulated defibrillation fields in six pig thoracic models and found a good correlation between model predicted voltages and the experimental data measured at 52 sites inside the thorax. Panescu et al. 16 also found a good correlation between measured voltages on the thorax surface and finite element model predictions. These results suggest that computational models with realistic geometries can closely simulate the field and current distributions produced by defibrillation shocks. The critical mass criterion 24 has been used in a number of modeling studies 6,13,14 to define successful defibrillation and to extract defibrillation metrics from the calculated field distribution. This criterion states that a minimum electric field E th is necessary throughout a critical mass of the ventricular myocardium to terminate fibrillation activation waves. These investigators have reported model DFT values that fall within one standard deviation of the mean DFTs reported in human clinical studies. The success of these computational studies suggested that similar computational models may be useful in the /04/ /1 C 2004 Biomedical Engineering Society

2 776 MOCANU et al. FIGURE 1. X-ray image showing the ICD with the catheter RV and SVC electrodes. presurgical planning of ICD implantation by providing an estimate of the DFT for individual patients. Thus, the goal of our research was to assess the predictive capacity of patientspecific computational models of transvenous defibrillation by comparing patient-specific simulated and clinical defibrillation metrics. In contrast with previous modeling studies, which have been based on a single average human thorax, we used a set of nine computational models built from segmented CT images of patients with an implanted ICD. The clinical defibrillation parameters were determined for each patient at the time of implantation and then used in the evaluation of the modeling results. We also present an analysis of the distribution of the weak field and its relationship to the DFT. METHODS Clinical DFT Testing Nine patients were recruited for implant (patient demographics shown in Table 1). Clinical DFT testing was performed with an active can emulator (CPI Ventak ECD, model 2185; 150 µf capacitance) and followed a standardized step-down clinical protocol. 19 VF was induced by applying a pulse of 60 Hz alternating current and the defibrillation shock was delivered 10 s later. The defibrillation waveform had a biphasic shape, with 60% tilt in the positive phase and 50% tilt in the negative phase (Fig. 2). The corresponding pulse width of the first and second phases was 60% and 40%, respectively, of the total pulse duration (10 15 ms). Typical trials started at 20 J (stored energy) and decremented until VF was no longer terminated using the following step-down scheme: 20 J 15 J 10 J 8 J 5 J. In cases where a 20 J shock could not defibrillate, the energy was gradually increased in 5 J increments until defibrillation was achieved. The ICD output was programmed at the minimum energy that successfully defibrillated on two occasions, single confirmation, plus a safety margin. The single confirmation could be consecutive, i.e. two consecutive at 5 J, or by way of a single reversal, i.e. 8 J success, 5 J failure, 8 J success. However, for the purposes of the modeling study, the minimum delivered energy that defibrillated on a single occasion was defined as the DFT and used in the comparison with model-predicted values. Finally, in all but one of the patients, the Endotak catheter lead system (Guidant/CPI) was implanted. A similar lead system (Medtronic) was used in the remaining patient. The ICDs were implanted in the left pectoral region with venous access via the subclavian vein (Fig. 1). Fluoroscopic imaging was used to verify the correct lead placement. CT Image Acquisition and Segmentation All patients were scanned post-implant on a spiral X-ray CT system (Somatom Plus 4, Siemens Medical Systems, NJ), with the ICD, SVC, and RV electrodes in place. The imaging protocol was approved by the Internal Review Board at Brigham and Women s Hospital, Boston, and each patient gave informed consent. The scanning TABLE 1. Patient demographics. Heart muscle Heart Patient # Gender Age volume a (cm 3 ) disease Heart arrhythmia Drugs 1 M CAD Primary prevention B-blocker 2 M CAD Spontaneous sustained VT Amiodarone Digitalis B-blocker 3 M CAD VF or cardiac arrest B-blocker 4 M CAD Primary prevention Amiodarone 5 M NDCP Spontaneous syncopal VT 6 M NDCP Spontaneous syncopal VT Digitalis 7 F CAD VF or cardiac arrest Digitalis B-blocker 8 M CAD Spontaneous sustained VT 9 M NDCP Spontaneous sustained VT Amiodarone B-blocker Note. CAD coronary artery disease; NDCP nonischemic dilated cardiomyopathy; VF ventricular fibrillation; VT ventricular tachycardia. a The heart muscle volume (ventricles and atria) was estimated based on the number and volume of the heart muscle voxels in the segmented CT images.

3 Patient-Specific Modeling of Defibrillation 777 FIGURE 2. CPI/Guidant biphasic defibrillation waveform with 60% tilt in the positive phase and 50% tilt in the negative phase. The pulse width of the first phase is 60% of the total pulse duration (typically ms). The DFT voltage is denoted with V th. procedure 12 was customized for high resolution image acquisition (120 kv, 200 ma, matrix size , mm/sec table feed). For better differentiation of soft tissues, a contrast agent was delivered to each patient prior to imaging (Omnipaque 300; Iohexol 300 g I/ml; Nycomed, Norway). To provide images of high contrast and a low noise level, an anisotropic filter was applied prior to the segmentation. Each segmented label was overlaid to the corresponding grayscale images to ensure accurate positioning. If the results of the semiautomated segmentation were not satisfying, manual fine-editing was done with MrX (General Electric Medical Systems, Milwaukee, WI) using a Sun workstation (UltraSPARC, Sun Microsystems Inc., Mountain View, CA). The following tissue types were identified: fat, blood, heart muscle, thoracic wall muscle, lung, bone (Fig. 3). The RV and SVC electrodes together with the ICD were also identified. Image-Based Computational Model For each patient, the 3-D computational model was constructed directly from the segmented images ( FIGURE 4. Voxel-based finite volume mesh. For clarity, only the bone structure and the lungs are shown. pixels) using a structured meshing algorithm in which the segmented voxels in the image data sets were defined as volume elements in the computational domain (Fig. 4). The size of each volume element was roughly 3 3 6mm 3, with slight variations depending on individual patient geometrical features. The total number of elements in the models varied between 350,000 and 450,000. Electrical conductivities 8,20 were assigned to the six tissue types according to Table 2. Conduction Boundary Value Problem In the quasistatic approximation of the electromagnetic field, 9 the electric potential distribution during a defibrillation shock is governed by Laplace s equation subject to Dirichlet boundary conditions on the electrodes (RV, SVC, and CAN) and Neumann homogeneous boundary condition (no flux) on the thoracic surface. The equation was solved numerically by the finite volume method using I-DEAS software package (Integrated Design and Engineering Analysis Software, Structural Dynamics Research Corporation, Milford, OH, USA). Calculation of Defibrillation Metrics Four defibrillation metrics were calculated to interpret and evaluate the solutions: interelectrode impedance Z, TABLE 2. Tissue conductivities. Tissue type Electrical conductivity σ (ms/cm) FIGURE 3. Segmented CT image. Heart muscle 2.5 Thoracic wall muscle 2.5 Blood 8 Lung 0.7 Bone 0.1 Fat 0.5

4 778 MOCANU et al. the DFT energy, the DFT current I th, and the DFT voltage V th. For each patient-specific simulation, the critical mass hypothesis 24 was used to define successful defibrillation with minimum delivered energy. According to this, a successful shock must expose a critical mass of the ventricular myocardium to electric fields equal to or greater than the inexcitability threshold E th. Zhou et al. 23 reported that a minimum electric field of 2.7 V/cm is required for biphasic shocks to successfully defibrillate approximately 77% of the time, and 5.4 V/cm to defibrillate approximately 87% of the time. We used an inexcitability threshold of 3.5 V/cm in our study to slightly exceed an 80% probability of success. In the simulations, a unit voltage was applied between the RV cathode and SVC and CAN anodes. The resulting electric field magnitudes were then scaled so that 95% of the ventricular myocardium was exposed to a minimum electric field magnitude of E th = 3.5 V/cm. The simulated DFT voltage and the DFT current are the corresponding scaled values of the applied voltage and the resulting delivered current. The DFT energy was calculated based on features of the Guidant/CPI biphasic waveform (Fig. 2) by integrating the power over the pulse duration: 1.6τ Vth 2 DFT = 0 Z e 2t/τ dt = 0.48CVth 2 (1) where Z is the patient-specific interelectrode impedance, V th is the DFT voltage, C is the capacitance (150 µf), and τ = ZC is the time constant. Statistical Analysis Correspondence between the clinical and model predicted defibrillation metrics was assessed using the correlation coefficient (cc) relating the predicted to measured values and the root mean square difference (rms) between the predicted and measured values. The paired t test was used to compare clinical and predicted DFT energy, and a value of p < 0.05 was used to establish statistically significant differences between the clinical and model-predicted DFTs. RESULTS Figure 5 shows the clinical DFT testing at the time of ICD implantation for all patients. Successful trials are shown with empty circles (Yes), while defibrillation failures are shown with solid circles (No). A demographic profile of the patient population used in this study is shown in Table 1. The comparison between the model-predicted and clinical interelectrode impedances resulted in a root mean squared difference (rms) of 8.2 and a correlation coefficient (cc) of 0.37 with a p-value = 0.09 (Fig. 6). In all patients but one (patient # 1), the model impedances overestimate the measured impedances with a cc = 0.7. Patient-specific clinical and model-predicted DFT energy values are compared in Fig. 7, where the clinical values correspond to the lowest energy shock that defibrillated on a single occasion. The rms difference for all nine patients was 12.4 J and the cc = Model-predicted DFTs were good estimates of the clinically determined thresholds in patients #1 4, with rms = 1.5 J and cc = A paired t test for patients #1 4 yielded a p-value of 0.41, indicating that there was no statistically significant difference between the clinical and model-predicted DFTs. For the second group of patients (# 5 9), the simulations underestimated the clinical DFT energy with rms = 18.4 J, cc = For patients # 5 9, the paired t test gave p = 0.01, indicating a statistically significant difference between the clinical and predicted DFTs. Similar trends were observed for the DFT voltages and the DFT currents (Figs. 8 and 9). We examined the electric field magnitude distribution associated with the clinical defibrillation threshold for each patient by applying the clinically measured DFT voltage to the corresponding model. The resulting cumulative histograms of the ventricular field magnitudes are shown in Fig. 10. We also examined the distribution of the weak field in each patient (Fig ), where the weak field is defined as the electric field magnitude in the ventricular myocardium less than 3.5 V/cm. By definition, for the critical mass criterion applied in this study, only 5% of the ventricular volume is included in the weak field region. Volume elements in the weak field region are indicated by a black cone (fixed volume) placed in the center of each volume element. The size of the largest cluster (percentage of total weak field) is also shown for each patient. DISCUSSION Our study investigates the degree to which patientspecific computational models can predict defibrillation threshold for individual patients. To our knowledge, this is the first study to directly compare clinical and modelpredicted DFTs on a patient-specific basis. The first step in our simulations was to compute the electric field distribution produced by a defibrillation shock using the finite volume method. We believe that this computational technique, like many other numerical methods (e.g. finite element method, finite difference method), is able to closely simulate the electric field distribution in the heart and thorax provided the model geometry closely approximates the patient s anatomy, and accurate tissue conductivity values are used. The second step of the simulation was to extract the defibrillation thresholds from the calculated cardiac electric fields. To do this, we used the critical mass criterion, a simple phenomenological model of defibrillation. Although debated, 3,18 the critical mass hypothesis for defibrillation is supported by experimental evidence 7,22,24

5 Patient-Specific Modeling of Defibrillation 779 FIGURE 5. Clinical DFT testing for all nine patients. = successful defibrillation attempt; = failed defibrillation attempt. For the modeling purposes, the lowest energy that defibrillated on a single occasion was defined as the DFT energy. and provides a simple means to relate the shock field at the continuum scale to the cellular-based events through which defibrillation is achieved. The interelectrode impedance is independent of the defibrillation assumptions used and is determined by the geometry and tissue properties. In all patients but one (patient FIGURE 6. Comparison between clinical and model-predicted impedances. FIGURE 7. Comparison between clinical and model-predicted DFT energies.

6 780 MOCANU et al. FIGURE 8. Comparison between clinical and model-predicted DFT voltages. #1) the model-predicted impedance overestimates the measured impedance (to a lesser extent for patient #5) (Fig. 6). Overestimates of the measured impedance were also obtained by Jorgenson et al. 10 for their porcine subject-specific finite element models based on similar mesh resolution and tissue conductivity values. This indicates the presence of a systematic error, which may orginate in the tissue conductivity values used or in the reconstruction of the anatomical structure. Model-predicted DFTs (energy, voltage, and current) were well matched to the clinically determined DFTs in four of the nine patients examined (first group: patients #1 4, Fig. 7). It is important to note that the respective clinical DFT energy values for these patients spanned an approximately 2-fold range. This suggests that the goodness of the achieved match was not due to the similarity of the patients examined but is a reflection of the ability of the modeling approach to capture the inherent anatomical variations in this group of patients. The clinical DFTs of the remaining five patients (second group: patients #5 9) FIGURE 9. Comparison between clinical and model-predicted DFT currents. FIGURE 10. Cumulative histograms for the computed electric field distribution. The applied voltage and all electric field magnitudes were scaled to correspond with the clinical DFT voltage. Vertical lines represent the mean field magnitudes (3.4 V/cm and 7.4 V/cm) associated with the two clusters. were not well matched by the simulated DFTs. We could not identify a consistent set of patient characteristics (e.g. heart volume, heart disease, or drug regimen) that were able to differentiate between these two groups (Table 1). The cumulative histograms shown in Fig. 10 were obtained by scaling the myocardial field distribution obtained for each patient model by the clinically measured DFT voltage corresponding to each patient. When viewed relative to the 95% critical mass perspective, these cumulative histograms (Fig. 10) reveal two distinct clusters of patients. The first cluster includes patients #1 4, with 95% of the ventricular myocardium exposed to field magnitudes greater than or equal to 3.4 ± 0.5 V/cm (mean ± standard deviation). The second cluster includes patients #5, 7, 8, and 9, with 95% of the ventricular myocardium exposed to field magnitudes greater than or equal to 7.4 ± 0.4 V/cm. Patient #6 belongs to neither cluster. One interpretation of these results is that a single inexcitability threshold is not able to describe all patients. While the 3.5 V/cm threshold does a very adequate job of characterizing patients #1 4, the data suggest that patients #5, 7, 8, and 9 require a substantially higher inexcitability threshold of 7.4 V/cm. It is unclear what causes the observed behavior. It has been shown that after unsuccessful shocks VF is reinitiated from the regions where the electric field is the weakest (weak field regions). 2,4 For this reason, we examined the spatial distribution of the weak field regions in each patient. In all of our simulations, the weak field was located in the posterolateral left ventricle (Figs ), consistent with what has been shown experimentally. 21 Our results show interesting differences in the patterns of the weak field distribution. Specifically, we found the weak field regions for patients that exhibited the highest clinical DFTs (Fig. 11) to be the most focal and compact (patients #7 9),

7 Patient-Specific Modeling of Defibrillation 781 FIGURE 11. Weak field distribution for patients with high clinical DFT. Patient #7: DFT = 19 J (the ICD pulse generator and the catheter are not shown for this patient); patient #8: DFT = 27 J; patient #9: DFT = 30 J. consisting of a single contiguous region that included 100% of the weak field region. Patients with more moderate clinical DFTs (#4, 6) had weak field regions that were still rather focal, but aggregated into two separate clusters (Fig. 12), with the largest being 90% of the weak field region. Finally, patients with the lowest clinical DFTs (#1 3, 5) exhibited the most scattered distribution of the weak field regions (Fig. 13). For these patients, the largest cluster contained less than 90% of the weak field region, with the exception of patient #3. These results suggest that large and compact weak field regions may favor the reinitiation of fibrillation, and that patients with such weak field distributions require a higher shock energy to defibrillate. The correspondence between the compactness of the weak field distribution and the clinical DFT may provide a predictive means of prospectively identifying high DFT patients prior to implant. Limitations The modeling approach used in this study uses a simple phenomenological model to relate the computed field distributions to defibrillation threshold values. Factors relating to the underlying cellular electrophysiology enter the model only through the inexcitability threshold E th and the critical mass criterion (expressed as a percentage of the ventricular mass). To the extent that the empirically-based model captures normative behavior, the model cannot be expected to predict defibrillation parameters for patients with cellular electrophysiology that differs substantially from the norm with a single E th. Similarly, we did not account for the presence of infarct regions, myocardial ischemia, or the effect of patient drug regimens in our models. The modeling approach is also inherently deterministic, and ignores the known probabilistic nature of defibrillation. Additionally, we have treated the lowest energy that defibrillates for each patient as the corresponding DFT, without taking into account the discrete nature of the clinical search protocol used to find the DFT. The truly lowest defibrillation energy could be somewhat lower than that found by the search protocol. CONCLUSIONS FIGURE 12. Weak field distribution for patients with moderate clinical DFT. Patient #4: DFT = 9.8 J; patient #6: DFT = 14 J. This paper presents comparisons between modelpredicted and clinical defibrillation metrics determined for individual patients with implanted ICDs Model-predicted parameters were extracted from the calculated cardiac electric field distribution using the 95% critical mass criterion and an inexcitability threshold of 3.5 V/cm for biphasic waveforms. The patient-specific simulations produced good estimates of the clinical DFTs in four of the nine patients investigated. Clinical DFTs and model predictions were well correlated for both well-matched and poorly-matched

8 782 MOCANU et al. FIGURE 13. Weak field distribution for patients with low clinical DFT. Patient #1: DFT=5J;patient #2: DFT = 4.8 J; patient #3: DFT = 7.4 J; patient #5: DFT = 7.7 J. groups of patients. Our results thus indicate that computational models based solely on the critical mass criterion and a single value of the inexcitability threshold are not able to consistently predict defibrillation thresholds for individual patients. However, inspection of the weak field distribution in all nine patient-specific models revealed that patients with the most compact weak field region were the most difficult to defibrillate and exhibited the highest clinical DFTs. This correspondence indicates that the weak field analysis may provide a promising approach to identify high DFT patients prior to implant. ACKNOWLEDGMENTS This study was supported in part by a grant from Guidant Corporation., St. Paul, MN, and the Trustees of Boston University. The authors would like to thank Dr. Michael Benser from the Cardiac Rhythm Management Laboratory (Guidant Corporation) for his valuable suggestions and comments. The excellent support of the Scientific Computing and Visualization Group at Boston University is also acknowledged. REFERENCES 1 Camacho, M. A., J. L. Lehr, and S. R. Eisenberg. A threedimensional finite element model of human transthoracic defibrillation: Paddle placement and size. IEEE Trans. Biomed. Eng. 42: , Chattipakorn, N., B. H. KenKnight, J. M. Rogers, R. G. Walker, G. P. Walcott, D. L. Rollins, W. M. Smith, and R. E. Ideker. Locally propagated activation immediately after internal defibrillation. Circulation 97: , Chen, P. S., N. Shibata, E. Dixon, P. D. Wolf, N. D. Danieley, M. B. Sweeney, W. M. Smith, and R. E. Ideker. Activation during ventricular defibrillation in open-chest dogs: Evidence of

9 Patient-Specific Modeling of Defibrillation 783 complete cessation and regeneration of ventricular fibrillation after unsuccessful shocks. J. Clin. Invest. 77: , Chen, P. S., P. D. Wolf, F. J. Claydon, E. Dixon, H. J. Vidaillet, N. D. Danieley, T. C. Pilkington, and R. E. Ideker. The potential gradient field created by epicardial defibrillation electrodes in dogs. Circulation 74: , Claydon, F. J., T. C. Pilkington, A. S. L. Tang, N. M. Morrow, and R. E. Ideker. A volume conductor model of thorax for the study of defibrillation fields. IEEE Trans. Biomed. Eng. 35: , De Jongh, A. L., E. G. Entcheva, J. A. Repogle, B. S. Booker, B. H. KenKnight, and F. J. Claydon. Defibrillation efficacy of different electrode placement in a human thorax model. PACE 22: , Garrey, W. E. The nature of fibrillary contraction of the heart: Its relations to tissue mass and form. Am. J. Physiol. 33: , Geddes, L. A., and L. E. Baker. The specific resistance of biological material: A compendium of data for the biomedical engineer and physiologist. Med. Biol. Eng. 5: , Haus, H. A., and J. Melcher. Electromagnetic Field and Energy Englewood Cliffs NJ: Prentice-Hall, 1988, p Jorgenson, D. B., P. H. Schimpf, I. Shen, G. Johnson, G. H. Bardy, D. R. Haynor, and Y. Kim. Predicting cardiothoracic voltages during high energy shocks: Methodology and comparison of experimental to finite element model data. IEEE Trans. Biomed. Eng. 42: , Karlon, W. J., S. R. Eisenberg, and J. L. Lehr. Effects of paddle placement and size on defibrillation current distribution: A three dimensional finite element model. IEEE Trans. Biomed. Eng. 40: , Kettenbach, J., A. G. Schreyer, S. Okuda, M. O. Sweeney, S. R. Eisenberg, C. F. Westin, F. Jacobson, R. Kikinis, B. H. KenKnight, and F. A. Jolesz. 3-D Modeling of the chest in patients with implanted cardiac defibrillator for further bioelectrical simulation. In: Proceedings of the C.A.R., edited by M. W. Vannier and K. Inmura. New York: Elsevier Science, 1998, pp Kinst, T. F., M. O. Sweeney, J. L. Lehr, and S. R. Eisenberg. Simulated internal defibrillation in humans using an anatomically realistic three-dimensional finite element model of the thorax. J. Cardiovasc. Electrophysiol. 8: , Min, X., and R. Mehra. Finite element analysis of defibrillation fields in a human torso model for ventricular defibrillation. Prog. Biophys. Mol. Biol. 69: , Naccarelli, G. V., and E. P. Veltri. Implantable cardioverterdefibrillators: What s in the future? In: Implantable Cardioverter- Defibrillators, edited by G. V. Naccarelli and G. V. Veltri. Boston: Blackwell Scientific, 1993 pp Panescu, D., J. G. Webster, W. J. Tompkins, and R. A. Stratbucker. Optimization of cardiac defibrillation by threedimensional finite element of the human thorax. IEEE Trans. Biomed. Eng. 42: , Sepulveda, W. J. P., and D. S. Echt. Finite element analysis of cardiac defibrillation current distribution. IEEE Trans. Biomed. Eng. 37: , Shibata, N., P. S. Chen, E. G. Dixon, P. D. Wolf, N. D. Danieley, W. M. Smith, and R. E. Ideker. Epicardial activation following unsuccessful defibrillation shocks in dogs. Am. J. Physiol. 255 H902-H904, Singer, I., and D. Lang. Defibrillation threshold: Clinical utility and therapeutic implications. PACE 15:932, Tacker, W. A., J. Mercer, P. Foley, and S. Cuppy. Resistivity of skeletal muscle, skin, and lung to difibrillation shocks. Paper presented at the AAMI 19th Annual Meeting, Washington, DC, 1984, p Tang, A. S. L., P. D. Wolf, Y. Afework, W. M. Smith, and R. E. Ideker. Three-dimensional potential gradient fields generated by intracardiac catheter and cutaneous patch electrodes. Circulation 85: , Witkowski, F. X., P. A. Penkoske, and R. Plonsey. Mechanism of cardiac defibrillation in open-chest dogs with unipolar DCcoupled simultaneous activation and shock potential recordings. Circulation 82: , Zhou, X., P. J. Daubert, P. D. Wolf, W. M. Smith, and R. E. Ideker. Epicardial mapping of ventricular defibrillation with monophasic and biphasic shocks in dogs. Circ. Res. 72: , Zipes, D. P., J. Fischer, R. M. King, A. Nicoll, and W. W. Jolly. Termination of ventricular fibrillation in dogs by depolarizing a critical amount of myocardium. Am. J. Cardiol. 36:37 44, 1975.