Construction of 3D MR image-based computer models of pathologic hearts, augmented with histology and optical fluorescence imaging to characterize action potential propagation

Abstract

Cardiac computer models can help us understand and predict the propagation of excitation waves (i.e., action potential, AP) in healthy and pathologic hearts. Our broad aim is to develop accurate 3D MR image-based computer models of electrophysiology in large hearts (translatable to clinical applications) and to validate them experimentally. The specific goals of this paper were to match models with maps of the propagation of optical AP on the epicardial surface using large porcine hearts with scars, estimating several parameters relevant to macroscopic reaction–diffusion electrophysiological models. We used voltage-sensitive dyes to image AP in large porcine hearts with scars (three specimens had chronic myocardial infarct, and three had radiofrequency RF acute scars). We first analyzed the main AP waves’ characteristics: duration (APD) and propagation under controlled pacing locations and frequencies as recorded from 2D optical images. We further built 3D MR image-based computer models that have information derived from the optical measures, as well as morphologic MRI data (i.e., myocardial anatomy, fiber directions and scar definition). The scar morphology from MR images was validated against corresponding whole-mount histology. We also compared the measured 3D isochronal maps of depolarization to simulated isochrones (the latter replicating precisely the experimental conditions), performing model customization and 3D volumetric adjustments of the local conductivity. Our results demonstrated that mean APD in the border zone (BZ) of the infarct scars was reduced by ∼13% (compared to ∼318 ms measured in normal zone, NZ), but APD did not change significantly in the thin BZ of the ablation scars. A generic value for velocity ratio (1:2.7) in healthy myocardial tissue was derived from measured values of transverse and longitudinal conduction velocities relative to fibers direction (22 cm/s and 60 cm/s, respectively). The model customization and 3D volumetric adjustment reduced the differences between measurements and simulations; for example, from one pacing location, the adjustment reduced the absolute error in local depolarization times by a factor of 5 (i.e., from 58 ms to 11 ms) in the infarcted heart, and by a factor of 6 (i.e., from 60 ms to 9 ms) in the heart with the RF scar. Moreover, the sensitivity of adjusted conductivity maps to different pacing locations was tested, and the errors in activation times were found to be of approximately 10–12 ms independent of pacing location used to adjust model parameters, suggesting that any location can be used for model predictions.

Introduction

Cardiac computer models can be utilized to study normal propagation of action potential (AP) and potential inducibility of arrhythmic events, with the most dangerous manifestation of arrhythmias being ventricular tachycardia (VT) and ventricular fibrillation (VF) (Janse and Wit, 1989, Kleber and Rudy, 2004). Theoretical predictions of AP propagation play a major role in electrophysiology since they can complement clinical observations (limited to surfacic measures) helping one understand inducibility of reentrant VT/VF, and they can give mechanistic insights into the generation of VF and defibrillation mechanisms. In particular, attention has been recently given to 3D MR image-based models built from high resolution scans of hearts with myocardial infarction (Vadakkumpadan et al., 2009, Vigmond et al., 2009), a major cause of sudden death in the industrialized world (Martinez-Rubio et al., 1999). These sophisticated in silico computer models of the heart could not only advance the basic cardiac research field, but also help clinicians tailor radiofrequency ablation (RFA) therapy, assist in the implantation of defibrillator devices and optimize cardiac resynchronization therapies (Trayanova, 2006, Trayanova, 2009). However, before integrating these mathematical models into realistic clinical platforms, they have to be validated experimentally with a sufficient level of detail and accuracy (Clayton, 2001, Hunter et al., 2008). In addition, regardless of the species modeled, such validations are often hampered by the degree of complexity specific to multi-scale models as well as the large number of variables involved in the mathematical equations.

Research is needed to enrich the current cardiac experimental databases and to perform model customization from electrophysiological data/measures, particularly for large animal hearts that are the most relevant to human clinical applications. Important tasks include: (1) the choice of appropriate imaging modality that provides sufficient detail of anatomy, fiber direction and morphology (scar) to construct the 3D heart model, including histological validation of scar heterogeneities; (2) the choice of appropriate mathematical model whose output parameters can be compared at a spatiotemporal scale replicating experiments performed under precisely controlled conditions; (3) the selection of a suitable experimental technique for electrophysiological measures; and (4) model customization and adjustment of the input parameters.

One imaging modality of choice is MRI, currently used to evaluate cardiac disease in preclinical studies and routine clinical examinations. For instance, late-enhancement MRI (using contrast agents) is known to be a powerful tool used to characterize infarct heterogeneities associated with reentrant VT in patients with myocardial infarction (Yan et al., 2006, Schmidt et al., 2007), but the current 5–8 mm slice thickness used can introduce partial volume effects. In addition, the right ventricle wall cannot be well visualized and segmented; therefore, the accuracy in reconstructing a 3D heart in vivo model could be severely affected. In an animal model of macro-reentrant VT, the electrophysiological properties of tissue were correlated with the contrast-enhanced MR signal (i.e., anatomic identification of dense scar and the border zone) obtained ex vivo, with sub-mm resolution (Ashikaga et al., 2007). Unfortunately, these studies lack the evaluation of fiber direction and the local quantification of anisotropy within the infarction area, which is crucial to the modelling of wave propagation in myocardial tissue (Kadish et al., 1988, Taccardi et al., 2008, Vetter et al., 2005). Alternatively, other ex vivo MR studies have explored non-contrast techniques like diffusion-tensor (DT) MRI, which are particularly useful in scar identification and reconstruction of fiber orientation (Wu et al., 2007). In order to accurately determine the extent of dense, collagenous scar in chronic infarcts, one can compute maps of fractional anisotropy (FA) and/or apparent diffusion coefficient (ADC), and, based on the MR signal heterogeneity, these maps can be further segmented into healthy tissue, border zone and core scar zone. Using such high-resolution maps, 3D cardiac computer models have been constructed so far for rabbit, dog, and porcine hearts (Bishop et al., 2009, Vadakkumpadan et al., 2010, Pop et al., 2009b) and used for different simulation purposes.

Cardiac multi-scale and multi-dimensional mathematical models give unprecedented detail as they integrate structural and functional information from sub-cellular levels, to slabs of myocardial tissue and whole organ (Gavaghan et al., 2006, Austin et al., 2006, Hunter and Nielsen, 2005, Hunter et al., 2008, Prassl et al., 2009, Niederer et al., 2011); however, one has to consider the level of detail of the model when considering implementation in a clinical setting where computational time is an issue. Moreover, certain applications might not justify the use of sophisticated equations and super-computers. The fastest and simplest numerical model is based on the Eikonal equations (Keener and Sneyd, 1998) which compute wavefront propagation; the result can be compared with clinically observed surfacic measures of depolarization isochrones; however, this model lacks refractory properties of the myocardium. At an intermediate level of complexity are the monodomain equations based on reaction–diffusion phenomena (Aliev and Panfilov, 1996b). In these models, the heart is modeled as a continuous medium (syncytium) and the solution for AP captures the main characteristics of the AP wave: duration, shape and upstroke. For simple applications, the monodomain model can be used to simulate normal and reentrant wave propagation in large hearts (Nash and Panfilov, 2004, Aliev and Panfilov, 1996a), requiring less than 1 h computational time (on an ordinary PC) to simulate 1 s of cardiac cycle (Sermesant et al., 2006). This choice is also attractive for some clinical (Sermesant et al., 2005), given that little difference (2%) in the activation times was found compared to the bidomain model (Potse et al., 2006), which requires 130 CPUs to simulate 1 s of heart cycle.

Finally, the experimental technique should be appropriately selected to enable comparison between measurements and the outputs of the theoretical model, at the same spatio-temporal scale. The comparisons can be used to customize the model; that is, several parameters of interest can be estimated from measurements and used to adjust the model variables (Sermesant et al., 2003). At the organ level, except for monophasic action potential catheter measurements (Kim et al., 2002, Hao et al., 2004), the clinical electrophysiological techniques are limited to measurements of depolarization times (Schmitt et al., 1999, Dukkipati et al., 2008); moreover, all provide surfacic measurements only, primarily from the endocardium. An alternative is to study electrophysiology at the tissue level; optical fluorescence imaging (which uses fast voltage-sensitive dyes) provides accurate measurements of AP waves in explanted perfused hearts prepared under physiological conditions approximating those seen in vivo (Bayly et al., 1998, Efimov et al., 2004, Hillman et al., 2007, Hyatt et al., 2005a, Hyatt et al., 2005b, Kay et al., 2004, Banville and Gray, 2002, Yang et al., 2007, Nanthakumar et al., 2007, Qin et al., 2003). Several optical studies demonstrated that the optical technique is also feasible for mapping action potential propagation in the presence of structural obstacles, which perturb normal propagation. For instance, epicardial propagation of AP (during pacing conditions or VT/VF) was mapped in rabbit and rat hearts with chronic infarcts (Li et al., 2003, Mills et al., 2005) and revealed changes in conduction velocity at the border zone of the infarct. Other studies mapped the AP propagation in the presence of macroscopic obstacles created by thermally damaging the tissue using cryoablation, high intensity focused ultrasound or laser energy (Qu et al., 2004, Pastore and Rosenbaum, 2000, Girouard and Rosenbaum, 2001). Using our experimental set-up (Pop et al., 2007) we recently reported results obtained by mapping optical AP waves in large, healthy porcine hearts. We also performed 3D model construction from MRI scans of the same hearts and adjusted the associated model based on optical measurements (Pop et al., 2009a).

In this paper, we present novel results obtained in experimental and theoretical studies for explanted porcine hearts with scars. In accordance with the diagram below (Fig. 1), we map the optical AP propagation in pathologic hearts and build 3D cardiac computer models allowing monodomain formalism, specifically, reaction–diffusion equations proposed by Aliev and Panfilov (1996b). We further estimate model parameters from the comparison between the measured and computed activation maps, and observe the model’s behavior in response to stimulation at different locations and with various pacing frequencies. Specifically, we built 3D MR-image-based models obtained from large porcine hearts with two different types of large scars: (1) large lesions generated by RF ablation and (2) chronic infarct scars. We hypothesized that the first type is an unexcitable scar that will not propagate action potential due to myocytes’ necrosis within the thermally coagulated scar; this corresponds to a simplified 3D heart model comprised of two zones: healthy (with normal conductivity) and scar (with zero conductivity). The second type of pathology (i.e., chronic infarct) was selected due to heterogeneities specific to the healing process, where dense collagenous scar and border zone (i.e., a mixture of necrotic and surviving myocytes with altered electrophysiological properties) give rise to a more complex cardiac electrical propagation.

The last goal was to demonstrate the potential for parameter estimation and model adjustment from experimental measures for a given stimulus pattern and subsequent prediction of response to different stimuli. To achieve the customization, we estimated several parameters (e.g., anisotropy ratio, AP duration and up-stroke) directly from 2D optical recordings; and also adjusted the 3D conductivity map (tuning the conduction velocity) by minimizing iteratively the error between simulated and measured depolarization times. Finally, to observe the influence of perturbation in fiber orientation, we generated synthetic fibers and studied their effect on the activation times compared with those obtained with the realistic directions calculated from DTI.