Virtual Electrophysiologic Study in a Three-dimensional Cardiac MRI Model of Porcine Myocardial Infarction

Abstract

Objective

This study sought to test the hypothesis that “virtual” electrophysiologic studies (EPS) on an anatomic platform generated by 3D MRI reconstruction of the left ventricle (LV) can reproduce the reentrant circuits of induced ventricular tachycardia (VT) in a porcine model of myocardial infarction (MI).

Background

Delayed-enhancement MRI has been used to characterize MI and “gray zones”, which are thought to reflect heterogeneous regions of viable and non-viable myocytes.

Methods

MI by coronary artery occlusion was induced in eight pigs. After a recovery period, 3D cardiac MRIs were obtained from each pig in-vivo. Normal areas, gray zones, and infarct cores were classified based on voxel intensity. In the computer model, gray zones were assigned slower conduction and longer action potential durations than those for normal myocardium. Virtual EPS was performed and was compared to results of actual in vivo programmed stimulation and non-contact mapping.

Results

The LV volumes ranged from 97.8 to 166.2 cm³ with 4.9 to 17.5% of voxels classified as infarct zones. Six of the seven pigs that developed VT during actual EPS were also inducible with virtual EPS. Four of the six pigs that had simulated VT had reentrant circuits that approximated the circuits seen with non-contact mapping, while the remaining two had similar circuits but propagating in opposite directions.

Conclusions

This initial study demonstrates the feasibility of applying a mathematical model to MRI reconstructions of the LV to predict VT circuits. Virtual EPS may be helpful to plan catheter ablation strategies or to identify patients who are at risk for future episodes of VT.

INTRODUCTION

Sudden cardiac death (SCD) is a major health issue faced in the United States affecting up to 400,000 people a year (1). Most SCDs are due to arrhythmias, namely ventricular tachycardia (VT) or ventricular fibrillation (VF). The ability to identify at risk patients prior to having a cardiac arrest is critical, as prophylactic implantable cardioverter defibrillators can significantly reduce mortality. Although current techniques and algorithms identify populations that are at higher risk for SCD, they lack the ability to truly discriminate the low risk group from the high risk group (2).

Electrical mapping and pacing during animal (3,4) and clinical (5–7) studies have demonstrated that VTs following myocardial infarction (MI) are often macroreentrant circuits supported by an anatomic substrate resulting from infarct scarring. These circuits can be complex, containing areas of slow conduction and multiple pathways of reentry. Stevenson et al (8) showed that some of the pathways may be critical while others may simply be bystander areas which when interrupted through ablation will not terminate the arrhythmia.

It has been shown that magnetic resonance imaging (MRI) with contrast can be used to detect scarring after MI. It has also been shown that infarct size is a better predictor of inducible VT than the left ventricular ejection fraction (9), highlighting the importance of the infarct substrate in the generation of VT. Infarct size measure by MRI has also been shown to be an independent predictor of mortality (10) and spontaneous VT (11). The surrounding border zones, also known as “gray zones”, are thought to be a heterogeneous mix of viable and non-viable myocardium and have been shown to correlate with post-MI mortality (12), clinical VT (13) and VT inducibility (14). Infarct transmurality measured from the MRI has also been shown as a predictor of ventricular arrhythmias (15). Because of the strong correlation observed between the infarct morphologies observed on MRI and pathologic analysis (16), we hypothesized that “virtual electrophysiologic study” (VEPS) consisting of the combination of 1) anatomic characterization of infarct morphology by 3D MRI in vivo and 2) the computer simulation of cardiac electrophysiology could be used to predict the characteristics of induced VT during actual electrophysiologic study (EPS). This hypothesis was tested using a porcine model of chronic MI.

METHODS

Animal Model of Myocardial Infarction

Eight pigs weighing from 55 to 75 kg were used for this study. The closed chest coronary occlusion protocol to induce MI has been previously described (17) and presented in more detail in the data supplement. Briefly, 300 mL of agarose gel beads diluted in 1.5 mL saline were injected by a balloon catheter into either the left circumflex artery or the left anterior descending coronary artery just distal to the second marginal or diagonal branch to cause infarction. The pigs were then allowed 4 to 8 weeks to recover. The experimental protocol was approved by the Animal Care and Use Committee of Northwestern University.

Contrast-enhanced 3D MRI

Following the recovery period, cardiac MRIs were obtained from the pigs under general anesthesia (1–2.5% isoflurane) using a Siemens 3.0 Tesla Trio MRI scanner (Siemens Medical Solutions, Erlangen, Germany) in the Center for Advanced MRI at Northwestern University. A free-breathing 3D phase sensitive inversion-recovery (PSIR) turbo FLASH pulse sequence was used for acquisition (18). Average 3D PSIR scan time was 9.8±3.4 minutes. PSIR reconstruction was utilized to eliminate the need for precise setting of the inversion time and parallel imaging was employed to improve the acquisition speed. Image data were collected during free breathing by synchronizing the acquisition to the respiratory cycle using a crossed slice navigator. This technique yields near isotropic spatial resolution with voxel sizes of 1.8×1.9×1.8mm. Images were acquired approximately 15 to 20 minutes after an intravenous injection of contrast (0.2 mmol/kg of gadopentetate dimeglumine, Magnevist, Bayer HealthCare).

Non-contact Mapping of Ventricular Tachycardia

EPS was performed after the four-to-eight week recovery period within three days of the MRI. Non-contact mapping (19) was performed using a commercial system (EnSite 3000, Endocardial Solutions, Inc., St. Paul, MN, USA) which recorded signals from a 64-electrode array mounted on a 9Fr catheter positioned in the left ventricle (LV) via a retrograde transaortic approach. The system creates a three-dimensional geometry on which sequential isopotential maps constructed from over 3,000 virtual unipolar electrograms are displayed. Induction of VT was attempted by programmed ventricular stimulation in the right and left ventricle with up to 3 extrastimuli after a drive train of 8 paced beats at 2 basic cycle lengths. The virtual unipolar electrograms of any induced tachycardias (either by programmed ventricular stimulation or due to catheter manipulation) were saved for offline determination of arrhythmia characteristics and scar exit sites. Episodes greater than 30 seconds were considered sustained VT.

MR Image processing

All image processing of MRI data was performed using custom designed software. The regions corresponding to the LV were then manually segmented. The segmented data were then linearly interpolated for a resulting resolution of 0.45×0.475×0.45 mm.

Each voxel of the LV was classified as either normal, gray zone, or non-viable infarct core using a modification of a previously published algorithm by Schmidt et al (14). First, the enhanced area of the MRI corresponding to the infarct was manually approximated. The selected area was overestimated to include normal regions at the boundaries of the scar. The non-selected area was therefore classified as normal myocardium. The classification algorithm was not applied to the non-selected area to avoid false detection of scar due to noise or artifacts. Within the selected scar region, all voxels with intensity values less than the mean plus three standard deviations of the intensities of the selected normal myocardium were classified as completely viable. A second threshold was used to determine whether the remaining voxels were classified as part of the gray zone or part of the infarct core. Threshold values between 10 to 50% of the range bounded by the mean and three standard deviations of the normal myocardium intensities on the lower end and the maximum intensity of the identified scar area on the upper end were used at 5% increments. Figure 1 shows how the distribution of gray zone as seen on the epicardium increases relative to the non-viable scar as the threshold value is increased from 10% to 50%.

Figure 1. Myocardium/scar Classification.

The scar profiles of a pig left ventricle with gray zone thresholds increasing from 10% to 50% are shown. As the viability thresholds increase the infarct core area decreases while the gray zone area increases.

Mathematical Modeling of Cardiac Electrophysiology

Three-dimensional isotropic computer models were implemented using the three-dimensional geometry of the LV constructed from the MRIs. The Fenton-Karma three-current ventricular action potential model (20) was used for this study. The tissue-conduction model was created by the coupling of neighboring cells in a 3D lattice with current diffusion controlled by a diffusion constant D. A scaled membrane voltage variable u is dependent on a fast inward current J_fi, a slow inward current J_si, a slow outward current J_so, and a stimulus current J_stim as described by the following partial differential equation:

$$\frac{\partial u}{\partial t}=\mathrm{\nabla }·D\mathrm{\nabla }u-(J_{\mathit{\text{fi}}}+J_{\mathit{\text{si}}}+J_{\mathit{\text{so}}}-J_{\mathit{\text{stim}}})$$

The ordinary differential equations describing the Fenton-Karma gating variables were integrated with the Rush and Larsen method (21) with adaptive time steps of 0.01 ms to 0.1 ms. The partial differential equation was solved with the Euler forward method. The no-flux boundary condition was used (20).

The parameters of the Fenton-Karma model were altered using the procedure recommended by Oliver and Krassowka (22) to obtain the restitution curve characteristics shown in Figure 2. The gray zone areas were assigned parameters to mimic the longer action potential duration observed in healed infarction relative to normal myocardium (a roughly 30 to 40 ms difference) (23). The diffusion constants were chosen to obtain conduction velocity values of 0.33 m/s and 0.17 m/s for the normal and gray zone areas, respectively. Simulations were performed on workstations equipped with a dual-processor motherboard and two Intel Xeon Quad Core processors with clock speeds of 3.16 GHz with parallelization across the eight total processing cores.

Figure 2. Model Restitution Curves.

The restitution curves of (A) action potential duration and (B) conduction velocity assigned to the normal and gray zone myocardium are displayed. The action potential durations at 90% repolarization (APD90) of the gray zone areas (shown in red) are longer action for a given diastolic interval than those of the normal restitution curve (shown in blue). Normal areas are assigned conduction velocities that are higher in the those of the gray zones at any given diastolic interval.

Virtual Electrophysiologic Study

VEPS for each LV model at each of the gray zone/infarct core thresholds (10 to 50% at 5% increments) was performed. The following protocol was performed while being blinded to the actual inducibility of the eight pigs and the characteristics of any induced VT. Programmed stimulation consisted of three beats at times 0, 400 ms, and 210 ms with pulse width of 2 ms. Stimulation was performed in the LV at one basal site, one apical site, and two sites midway between base and apex at either side of the scar for a total of 36 simulations per pig (4 sites × 9 threshold values). An arrhythmia was considered sustained if it lasted at least 10 seconds. Simulations were automatically terminated if no sustained activation was detected after stimulation. Animations of the simulated VT were viewed at 50% transmurality of the LV in order to evaluate the endocardial activation sequences. Comparison of any VTs induced by virtual EPS with actual VTs captured by non-contact mapping was performed by two reviewers. Simulations were repeated with the assumption of 1) no gray zone and 2) no gray zone or infarct core to confirm that the combined infarct and gray zone characteristics were primarily responsible for sustaining VT.

Statistics

Student’s t-test was used to compare the actual versus simulated VT cycle lengths. Pearson’s correlation coefficients were used to correlate VT cycle lengths with LV and infarct volumes. P-values less than 0.05 were considered statistically significant.

RESULTS

MRI Characteristics

The LV myocardium as determined by MRI had volumes ranging from 97.8 to 166.2 cm³ (mean 134±22 cm³). The left circumflex artery was occluded in four pigs and the left anterior diagonal descending artery was occluded in four pigs. The infarct volumes when considering both infarct cores and gray zones comprised 4.9 to 17.5% (mean 10±4%) of the total LV myocardium volume.

Electrophysiologic Study

Sustained monomorphic VT could be induced at least twice in 7 of the 8 pigs during EPS. Six pigs had VTs induced from the LV while one pig was induced by from the right ventricle. Two of the pigs had two different VT circuits. The VTs had a mean cycle length of 280±51 ms.

Virtual Electrophysiologic Study

A set of 36 simulations for each pig required on average 14±2 hours to complete. Table 1 shows the results of VEPS at each of the gray zone/infarct core threshold. Each of the thresholds labeled “x” correspond to having VT lasting at least 10 seconds. Pig 1 had inducible VT with virtual EPS but not with actual EPS while Pig 7 had inducible VT with actual EPS but not virtual EPS. There was no significant difference between the mean cycle lengths of the actual VT versus the virtual VT (280±51 vs. 263±61 ms, p=0.58). The comparison of cycle lengths for each pig is shown in Table 2. Repeating the VEPS on the eight heart models with the assumption of no gray zone (all infarct core) and no gray zone or infarct core resulted in no VT.

Table 1.

Inducibility with EPS and VEPS at Different Gray Zone Threshold Levels.

Pig	Inducible w/EPS?	Gray Zone/Infarct Core Threshold
		10%	15%	20%	25%	30%	35%	40%	45%	50%
1	NO					x	x	x	x
2	YES				x
3	YES	x
4	YES	x
5	YES				x	x
6	YES						x	x
7	YES
8	YES							x

x – indicates inducibility with VEPS

Table 2.

Comparison Between Actual and Virtual VT Cycle Lengths

Pig	VT Cycle Lengths (ms)
	Actual	Virtual
1	None	234
2	306	214
3	370	390
4	270	226
5	237	236
6	291	255
7	270	None
8	213	286

Simulated VT was induced via a common mechanism for all episodes as illustrated in the 3D membrane potential maps shown in Figure 3. The 3D maps show the simulated membrane potentials of the LV epicardial surface with blue areas representing resting tissue and red areas representing depolarized tissue. The gray regions are scar. The first paced beat, located outside but near the infarct area, at time 0 propagates through both the normal and gray zone regions of the left ventricle. The second beat at time 400 ms also propagates through both the normal and gray zone areas. The second beat in normal areas have action potential durations of 197 ms while the second beat in the gray zones have action potential durations of 236. The third beat at 610 ms propagates through the normal region but blocks at a scar isthmus due to the longer refractory period of the gray zone in the isthmus. The wavefront then propagates around the infarct area to the other side at which point the gray zone is repolarized and the wavefront is able to re-enter. VT then follows.

Figure 3. Illustration of Arrhythmia Induction.

The membrane potential maps illustrates simulated arrhythmia induction. The first beat propagates through both healthy and scar regions (Panels A and B). The second beat 400 ms later also propagates through both healthy and scar regions with the scar region take longer to repolarize (Panels C and D). The third beat 210 ms later propagates through the normal regions buts blocks at the initial contact with the scar (Panels E and F). Reentry around the scar is now initiated (Panels G and H).

VEPS vs. EPS

Six pigs had VT induced both with VEPS and actual EPS. Figure 4 shows the comparison of the endocardial membrane potential maps obtained by VEPS and the corresponding non-contact unipolar voltage maps obtained during EPS for the four pigs where good agreement in the scar exit sites were seen. Complete agreement of the VT circuits and scar exit sites was seen in Pigs 3, 5, and 6. In Pig 8, two exits sites were seen in the simulated VT circuit, one of which appears to match the exit site of the actual VT. Figure 5 shows the results for two pigs where the VT exit sites did not match. However, examination of the VT circuits reveal that activation propagated in opposite directions in the simulated versus actual VTs with reversed entrance and exits sites; this suggests that a similar isthmus may be supporting both the simulated and actual VT.

Figure 4. Simulated and Actual VT Circuits with Good Agreement.

The membrane potential maps of simulated VT from 4 pigs are shown to have good agreement in infarct exit sites and activation directions was seen when compared with non-contact maps of the actual VT.

Figure 5. Simulated and Actual VT Circuits with Reversed Directions.

The membrane potential maps of simulated VT from two pigs are shown to not have good agreement in infarct exit sites and activation directions was seen when compared with non-contact maps of the actual VT. However, the reversal of the VT entrance and exit sites suggests that a similar isthmus may be supporting reentry for both the simulated and actual VT circuits.

Correlation of VT cycle lengths with LV and infarct sizes

Table 3 shows the correlation coefficients of actual and simulated VT cycles lengths versus LV, infarct core, gray zone, and combined infarct core and gray zone volumes. The actual VT cycle lengths were not significantly correlated with any of the volume measurements. However, the simulated VT cycle lengths were significantly correlated with LV volume (r=0.79, p=0.04) and infarct core volume (r=0.90, p=0.007).

Table 3.

Correlation between VT Cycle Lengths and MRI Characteristics

Volume	Actual VT Cycle Length		Simulated VT Cycle Length
	r	p-value	r	p-value
LV	0.25	0.58	0.79	0.04
Infarct core	0.52	0.28	0.90	0.007
Gray zone	0.15	0.78	0.03	0.95
Infarct core + gray zone	0.2	0.67	0.56	0.19

DISCUSSION

This study demonstrated the feasibility of applying a mathematical model of the action potential to simulate conduction and predict VT circuits in an individualized model of the left ventricle generated in vivo using contrast enhanced MRI for the purpose of scar characterization. Virtual EPS results were similar to those obtained by actual EPS in 6 of 8 pigs. Virtual EPS may be useful for evaluating VT circuits to better plan approaches to catheter ablation and could be evaluated for its ability to predict future VT events in patients with MI and no history of VT.

MRI to predict ventricular tachycardia

A combination of structural (24–28) and electrophysiologic (23,29) remodeling resulting from myocardial infarction contribute to susceptibility to VT. In order for structural remodeling to provide a sufficient substrate for VT, a combination of infarct core and peri-infarct regions are needed. The infarct core provides anatomic boundary needed for sustained reentry. Infarct size/mass quantification by MRI has been shown to predict EPS-induced (9) and clinical VT (10,11,15). The peri-infarct zones are damaged but viable regions of myocardium which contribute to VT through slowing of conduction and uni-directional block due to altered refractory periods. The peri-infarct zones are characterized by “gray zones” in the MRI, intermediate intensities between the darker normal myocardium and the enhanced infarct core. Quantification of MRI gray zones have been used to predict both EPS-induced (14) and clinical VT (12,13).

In addition to the volume of infarct core and gray zones, there are morphological factors of both that appear to play a significant role in the inducibility of VT (30). Channel isthmuses within the infarct are known to facilitate VT and have been targeted for ablation. David-Perez et al showed these isthmuses can be detected by MRI and that patients with sustained monophorphic VT were more likely to have MRI-detected isthmuses than patients with similar ejection fractions without VT (31). The transmurality of the infarct is another important feature, as VT circuits likely require both an endocardial and epicardial boundaries to support reentry. Boyé et al showed that relative infarct transmurality measured from the MRI was an independent predictor of clinical VT events (15).

Although the size and morphology characteristics of the infarct mentioned likely all play a critical role in VT, their relative contributions are not easily quantified. For example, a larger infarct volume with a shorter isthmus length may be equally arrhythmogenic as a smaller infarct with a longer channel. The proposed VEPS system takes all of these factors into account in the prediction of VT circuits. Although electrophysiologic information cannot be obtained from the MRI, it is known that refractory period gradients across the normal myocardium and peri-infarct zones are needed for the VT induction process. Thus, our model assumes higher refractory periods in the MRI gray zones to allow realistic induction of VT with simulated programmed stimulation.

Other computer models of VT

Vigmond et al demonstrated simulated VT in a computer model created using high resolution MRI of a canine ventricle imaged ex vivo four weeks after MI (32). This model incorporated fiber orientation with diffusion tensor imaging and a simulated His-Purkinje network, while also characterizing non-viable infarct areas and the viable border zones. Pop et al showed in a pig with induced MI that computer simulation based on ex-vivo diffusion tensor MRI could predict the VT circuit obtained with actual EPS (33). In another study, Sermesant et al (34) demonstrated the feasibility of an electromechanical model of the infarct ventricle registered to fit clinical data from MR and X-ray, as well as actual cardiac electrical activity.

The present study is the first to demonstrate the possibility of computer simulation of VT with in vivo MRI using clinically available sequences. In contrast to the previously published study using ex vivo MRI, certain details such as fiber orientation cannot be obtained due to lower spatial resolution. In the development of our model, we therefore focused the model on the key elements of VT which can be obtained by clinical MRIs, namely the infarct core and gray zone characteristics. Even with the omission of the right ventricle, fiber orientation, and His-Purkinje system, we demonstrated VT circuits can predicted in our VEPS system. The simple model also makes simulation practical on commercial workstations rather than requiring a sophisticated computer cluster. Further study is needed to test whether more realistic ventricular modeling is necessary for the prediction of VT induction.

Limitations

A limitation of the study is the small sample size. Spatial resolution of the current images was approximately 2 mm in all three dimensions. This may be inadequate to delineate some VT circuits. However, image resolution is likely to improve with time and advances in MR technology. Defining the gray zone is challenging and there are no well-accepted criteria for this. Our study found that there was not a consistent threshold value that allowed inducible VT with simulation for all the pigs. Differences in signal-to-noise ratio may in part account for the differences in threshold. Better delineation of this region will also likely improve the performance of VEPS. Further work to develop and evaluate an algorithm for accurate tissue classification from MRI data is an important future study that will require large numbers of subjects. Clearly, further development and testing will be required to establish the utility of VEPS in humans.

Potential Clinical implications

Computer modeling of the heart has generally been confined to research applications. This pilot study suggests that contrast enhanced MRI for VEPS could have several potential applications – to help identify VT circuits prior to catheter ablation and potentially as a platform for risk stratification for sudden cardiac death. The technique is noninvasive, avoiding the risks and time required for the sedation and catheterization needed for standard electrophysiology testing. In addition, VEPS could allow detailed characterization of the induced arrhythmia that would only otherwise be possible with high resolution mapping. Ablation strategies could be planned with this information. Further evaluation and development of this technology will be necessary to assess its utility in the clinical setting.

Abbreviations

EPS

Electrophysiologic study

LV

left ventricle

MI

myocardial infarction

MRI

magnetic resonance imaging

PSIR

phase sensitive inversion-recovery

SCD

sudden cardiac death

VT

ventricular tachycardia

VF

ventricular fibrillation