Ventricular Arrhythmia Risk Prediction in Repaired Tetralogy of Fallot using Personalized Computational Cardiac Models

Abstract

Background:

Adults with repaired Tetralogy of Fallot (rTOF) are at increased risk of ventricular tachycardia (VT) due to fibrotic remodeling of the myocardium. However, the current clinical guidelines for VT risk stratification and subsequent ICD deployment for primary prevention of sudden cardiac death in rTOF remain inadequate.

Objective:

To determine the feasibility of using a rTOF-specific virtual-heart approach to identify patients stratified incorrectly as being at low VT risk by current clinical criteria.

Methods:

This multi-center retrospective pilot study included seven adult rTOF patients who were considered low-risk for VT based on clinical criteria. Patient-specific computational heart models were generated from the LGE-MRI, incorporating the individual distribution of rTOF fibrotic remodeling in both ventricles. Simulations of rapid pacing determined VT inducibility. Model creation and simulations were performed by operators blinded to clinical outcome.

Results:

Two patients in the study experienced clinical VT. The virtual hearts constructed from LGE-MRI scans of seven rTOF patients predicted correctly reentrant VT in the models from VT-positive patients and no arrhythmia in those from VT-negative patients. There were no statistically significant differences in clinical criteria commonly used to assess VT risk, including QRS duration and age, between patients who did and did not experience clinical VT.

Conclusions:

This study demonstrates the feasibility of image-based virtual-heart modeling in patients with congenital heart disease and structurally abnormal hearts. It highlights the potential of the methodology to improve VT risk stratification in patients with rTOF.

Introduction

Tetralogy of Fallot (TOF) is the most common cyanotic congenital heart defect, occurring in about 3 in 10000 births.¹ As surgical repair has evolved since the landmark surgical intervention for TOF in 1955¹, 90% of patients now survive at least 30 years after successful repair at a young age.^2,3 Adults with repaired TOF (rTOF) are at risk of developing ventricular tachycardia (VT) which can lead to sudden cardiac death (SCD).^4,5 The risk of SCD increases with age and time since repair.⁶ Implantable cardioverter defibrillators (ICD) successfully mitigate the risk of SCD in patients with rTOF at high risk for VT.⁷ However, the rate of ICD-related complications for adult patients with rTOF is three times higher than in the post-myocardial infarct population.^7–9

Current guidelines for ICD deployment in rTOF remain insufficient. Many clinical criteria have been proposed for VT risk stratification, including age at repair, need for transannular patch or ventriculotomy, and various MRI parameters.^4,10 A widely used clinical criterion for VT risk in rTOF is prolonged QRS duration, QRSd ≥180 ms,¹¹ however, in their study of rTOF patients undergoing VT ablation, Laredo et al. state that 71% of patients who experienced clinical VT had QRSd <180 ms.¹² As a result of inaccurate VT stratification, some rTOF patients undergo unnecessary ICD implantation, while others, who would benefit from an ICD, do not receive one.⁷ There is a need for improved VT risk stratification methodologies to guide clinical decision making and prevent SCD in the growing population of adult patients with rTOF.

Personalized late gadolinium enhanced magnetic resonance imaging (LGE-MRI)-based virtual-heart modeling has recently emerged a superior methodology for VT risk stratification in patients with ischemic cardiomyopathy (ICM).^13,14 These models allow for patient specific, non-invasive assessment of the electrophysiological effects of myocardial fibrosis, a known arrhythmogenic substrate for VT and nearly universal pathophysiology in rTOF.^15,16 We hypothesize that the LGE-MRI based virtual-heart approach could help clinical decision making regarding ICD implantation in patients with rTOF.

In this study, we present an rTOF-specific version of the personalized virtual heart approach as applied to VT risk prediction in this population. In this first use of the virtual-heart methodology for rhythm disorders in congenital heart disease, we focus specifically on using the methodology to identify rTOF patients deemed incorrectly at low VT risk by current clinical criteria, as this is the area where the virtual-heart approach could have maximum impact, potentially reducing mortality rates in patients with this complex heart disease.

Methods

Patient Population

This was an institutional review board-approved multi-center retrospective pilot study. Institutional cardiac MRI databases were queried for patients with rTOF who had undergone LGE-MRI. Inclusion criteria were as follows: age greater than 18 years, history of TOF with pulmonary stenosis, an EKG to assess QRS duration, and VT status recorded following the MRI and preceding and intervention such as valve replacement or electrophysiology procedure. Patients were excluded from the study if their LGE-MRI was not suitable for our virtual-heart reconstruction process or if they had pulmonary atresia. Chart review was conducted for an endpoint of multiple episodes of non-sustained VT or any episodes of sustained VT at the time of model construction. Non-sustained VT was defined as a wide QRS tachycardia at a rate greater than 120 beats per minute and lasting less than 30 seconds.

Imaging Protocol

Each patient in the cohort had undergone a standard clinical contrast enhanced cardiac MRI (Siemens Aera, 1.5T). Each diastolic short axis stack consisted of 8–12 slices with 7–10mm slice thickness. Imaging parameters were as follows: repetition time 2.5–9.1 ms, echo time 1.09–1.22 ms, average in-plane spatial resolution 1.6 mm, and inversion time adjusted to null the signal of normal myocardium. At the first center, LGE images were acquired 15–30 minutes after a total injection of 0.1 mmol/kg of gadolinium contrast agent with an inversion recovery fast gradient-echo pulse sequence. At the second center, LGE images were acquired similarly, but with 0.15 mmol/kg of gadolinium contrast and a single shot, phase-sensitive inversion recovery sequence which averages 8 images to create a single LGE image. All imaging acquisition parameters were standard clinical sequences designed to minimize artifact.^13,17

Overview of Computational Approach

Patient-specific computational heart models were generated for the seven low VT risk patients from the existing clinical LGE-MRI data, incorporating the individual distribution of fibrotic remodeling. Simulations of rapid pacing of the fibrotic substrate were executed to determine VT inducibility. A patient was considered at high risk for VT if their virtual electrophysiologic study demonstrated induction of VT. Model creation and simulations were performed by operators blinded to clinical outcome. Once simulations were completed, model predictions were compared to clinical outcomes.

Virtual-Heart Model Construction

Model construction, illustrated in Figure 1, was performed as previously described.¹³ Briefly, the ventricular myocardium was segmented from the LGE-MRI using a previously validated semi-automatic method (Fig. 1A).¹³ Each voxel in the myocardium was classified as normal, border zone, or core scar based on intensity using the signal threshold to reference mean method (Fig. 1B).²⁰ Unlike patients in our previous virtual-heart studies,^13,14,21,22 patients in this cohort had right ventricular (RV) fibrosis detectable on LGE-MRI. The model reconstruction pipeline was thus extended to accommodate this. Otsu thresholding was used to binarize the myocardium into regions of high and low intensity. Separate thresholds were computed for the RV and left ventricle (LV) to account for differing intensity of non-fibrotic myocardium. The mean of the low intensity region was chosen as the reference mean of normal myocardium. The standard deviation (SD) of the low intensity region was then used for thresholding of border zone and core scar. An intensity of ≥5 SD above the reference mean was used to classify voxels in the core scar region. Voxels between 3 and 5 SD above the reference mean were classified as border zone. All other voxels were classified as non-fibrotic myocardium. These thresholds have been used successfully in previous virtual-heart modeling studies.²² A 3D mesh was then created from the myocardial segmentation and myocardial fiber orientations were assigned in the mesh using a previously validated approach (Fig. 1C and 1D).^18,19

Figure 1:

Model creation workflow. Contouring of epicardium and endocardium on short-axis LGE-MRI (A). Intensity-based thresholding to identify regions of normal tissue, border zone, and core scar (B). Reconstruction of 3-dimensional ventricular geometry (C). Myocardial fiber orientation (D). Action potential traces for normal tissue and border zone (E). Pacing sites marked with (*) (F).

Assigning Electrophysiological Properties

Regions of core scar were modeled as non-conducting, while regions of border zone and normal myocardium were assigned human ventricular myocyte action potential dynamics (Fig. 1E). In regions of normal myocardium, we used the human ventricular myocyte model by ten Tusscher et al.,²³ with added representation of I_NaL.²⁴ In border zone areas, those typically found surrounding core scar, ion channel conductances were based on experimental recordings in human myocytes from hypertrophic cardiomyopathy (HCM) patients, as reported by Coppini et al.²⁵ As interstitial myocardial fibrosis is a common pathophysiology in both HCM and rTOF, a cell model that incorporates these experimental observations more accurately reflects the diffuse pathophysiologic changes in rTOF than the cell model for the ischemic cardiomyopathy border zone used in previous modeling studies.^25–27 In the absence of measurements of ionic current remodeling in areas of diffuse fibrosis in rTOF patients, the similarities in fibrosis histology between HCM and rTOF justified this modeling choice. Specifically, analogous to pathophysiological findings in rTOF,²⁶ replacement-type fibrosis and interstitial perimyocyte fibrosis were often found in the LV of patients with HCM.²⁷

Changes in the ten Tusscher model based on these experimental recordings included 107% increase of I_NaL maximal conductance, 19% increase of I_CaL maximal conductance, 34% decrease of I_Kr maximal conductance, 27% decrease of I_Ks maximal conductance, 85% decrease of I_to maximal conductance, 15% decrease of I_K1 maximal conductance, 34% increase of NCX activity, and 43% reduction of SERCA activity. The net results of the changes to the cell model in areas of border zone included increased action potential duration and abatement of the notch after depolarization (Fig. 1E).

Conduction velocities in the non-fibrotic myocardium and border zone were those used in our previous studies.^13,14,21,22 In non-fibrotic myocardium, the tissue conductivities were 0.136 and 0.0536 S/m in the longitudinal and transverse directions, respectively. This resulted in conduction velocities of 54.4 and 33.5 cm/s. Longitudinal and transverse conductivities in the border zone were altered as in previous modeling studies to reflect experimentally-observed connexin-43 remodeling in the border zone.²⁸ The longitudinal and transverse border zone conductivities were 0.0925 and 0.0209 S/m, resulting in conduction velocities of 43.2 and 17.9 cm/s.

Simulation Protocol

Simulations were performed using the software package CARP (https://carp.medunigraz.at/). Programmed electrical stimulation, similar to clinical protocols and to those used in previous simulation studies, was performed to examine the models’ propensity to VT induction.^13,14,21,22 Briefly, all models were paced from eight endocardial sites, including seven sites on the LV and one near the right ventricular outflow tract (RVOT), for six beats (S1) at a cycle length of 450 ms followed by a premature stimulus (S2) initially given at 300 ms (Fig. 1F). The timing between S1 and S2 was progressively shortened until VT was induced. If VT was not induced, an additional premature stimulus (S3) was delivered 250 ms after the previous stimulus. If VT was still not induced after S3, the model was considered non-inducible from that pacing site.

Descriptive statistics were calculated for all variables of interest and included count, median, mean, and standard deviation when appropriate. The independent samples t-test or Fisher’s exact test were used to compare differences between patients with and without clinical VT (considered statistically significant when p value ≤0.05).

Results

Chart review identified a total of 19 potential patients from two centers, 10 of which had sufficient image quality for model construction. Of these, seven patients were considered of low VT risk by the clinical criterion of QRS duration <180 ms. Model reconstruction and simulation-based VT risk assessment were performed for these seven patients. Five of the seven patients had experienced the primary endpoint at the time of model reconstruction. All seven patients included in the virtual-heart approach were adults (median age 21 years) who underwent successful TOF repair at a young age (median age at repair 8 months). Patient characteristics are summarized in Table 1 and details of repair history can be found in Table 2. Repairs were performed between 1972 and 1999. There was no significant difference (p=0.33) in the median year of repair between patients who experienced clinical VT (1981) and patients who did not experience clinical VT (1995). There was also no significant difference in age at the time of cardiac MRI (p=0.40) or at the time of TOF repair (p=0.50). RV ejection fraction was significantly higher in patients who experienced clinical VT (p=0.002). There were no statistically significant differences in any other common clinical metric used to assess heart function and risk of SCD.

Table 1:

Baseline characteristics of the low VT risk rTOF cohort. All patients have QRS duration < 180ms. Values expressed as mean ± standard deviation or mean (range) as appropriate. Abbreviations: VA (ventricular arrhythmia), CMR (cardiac magnetic resonance), TOF (tetralogy of Fallot), RV (right ventricle), EDVi (end diastolic volume index), ESVi (end systolic volume index), EF (ejection fraction), LV (left ventricle), LGE (late gadolinium enhancement). P-values calculated using t-test or Fisher’s exact test as appropriate.

	No Endpoint (n=5)	Primary Endpoint (VA) (n=2)	p
Age at CMR	21 (17–27)	44 (27–61)	0.40
QRS duration	148±27	125±21	0.34
TOF Repair History
Age at complete repair (months)	9 (4–17)	124 (8–240)	0.50
Year of repair	1995 (1989–1999)	1981 (1972–1989)	0.33
Transannular patch	4 (80%)	2 (100%)	1
CMR Characteristics
RV EDVi	138±50	141±90	0.98
RV ESVi	72±27	107±96	0.69
RV EF	46±1	49±0	0.002
LV EDVi	87±21	83±19	0.86
LV ESVi	40±10	32±4	0.21
LV EF	54±7	61±4	0.15
LGE present	2 (40%)	1 (50%)	0.52
Border zone volume (% of total volume)	10±2.6	6.3±3.5	0.35
Core scar volume (% of total volume)	4.0±2.0	2.2±0.1	0.17

Table 2:

Repair history for the low VT risk rTOF cohort. Abbreviations: VT (ventricular tachycardia), BT (Blalock-Taussig), LPA (left pulmonary artery), RPA (right pulmonary artery), RV (right ventricle)

ID	VT	QRS (ms)	Initial Palliation	Age at Complete Repair (m)	Year of Complete Repair	Transannular Patch	Extensive RV Ventriculotomy	Pulmonary Homograft
1	Yes	110	Potts shunt (aorta to LPA)	240	1972	Yes	Unknown	No
2	Yes	140	No	8	1989	Yes	Unknown	No
3	No	102	No	6	1989	Yes	No	No
4	No	148	No	17	1996	Yes	No	No
5	No	166	No	10	1995	Yes	No	No
6	No	172	No	4	1999	Yes	Yes	No
7	No	150	Right modified BT shunt	8	1997	No	No	Yes

The virtual-heart approach correctly identified the two patients who experienced clinical VT but were incorrectly deemed low risk by the clinical criterion. Further, the simulations correctly predicted VT noninducibility in the remaining five patients who did not experience clinical VT. There was no significant difference in QRSd between the two patients who experienced clinical VT and the five who did not (p=0.34). Virtual-heart models for all patients demonstrate the presence of and the patient-specific nature of myocardial fibrosis (Fig. 2). All patients had RV fibrosis, reflected in the respective models, most often near the RVOT; some also had LV fibrosis. While the ventricular geometry and fibrosis distribution were unique for each patient, mean volumes of border zone (8.9±3.1%) and core scar (3.5±1.9%) were not statistically significantly different between patients who did and did not experience clinical VT (p=0.35 and p=0.17, respectively).

Figure 2:

Reconstructed models for 7 patients with QRS duration <180 ms, 2 with clinical VT (A) and 5 without clinical VT (B).

For both patients who experienced clinical VT, simulated reentrant VT was induced by RV pacing, while no reentry was induced by LV pacing. Activation maps in Figure 3 show the patterns of reentrant VT observed in these patient-specific models; in both cases, a single VT morphology was induced by the pacing protocol. For both patients, simulated reentries occurred around an area of RV fibrosis near the RVOT. Movies of the VT reentrant activity in these two models are provided (Supplementary Video 1 and Supplementary Video 2). No reentrant VT was induced in the models of the five patients who did not have clinical VT.

Figure 3:

Activation maps showing simulated reentry morphologies induced from the pacing sites marked with (*) for the two patients with clinical ventricular tachycardia and QRS duration <180 ms. Isochrone line spacing is 18.75 ms.

Discussion

The goal of this proof-of-concept study was to assess computationally VT inducibility in rTOF patients deemed at low VT risk by clinical criteria. Clinical criteria for VT risk stratification in rTOF patients remain uncertain; we focused on the low-risk population as these patients might not receive ICD therapy as primary prevention of SCD. We hypothesized that in this patient cohort, the personalized virtual-heart approach could identify patients stratified improperly; it is for these rTOF patients where predictive simulations could make potentially a difference in patient management.

Models were constructed from LGE-MRI scans of seven rTOF patients deemed clinically at low VT risk. The personalized virtual-heart methodology predicted reentrant VT in the models for both VT-positive patients. In contrast, no arrhythmia was induced in any of the models from the five VT-negative patients. This study demonstrates the feasibility of image-based virtual-heart modeling in patients with congenital heart disease and the potential of the methodology to improve VT risk stratification in patients with rTOF.

In rTOF myocardial fibrosis is nearly universal in both ventricles and serves as an arrhythmogenic substrate.^15,16 Definitive surgical repair undoubtedly improves survival but often leads to formation of scar tissue near the RVOT and potentially elsewhere in the RV if a transventricular approach was used.^2,3,5,29,30 Fibrotic remodeling and dilation of the RV can also occur as a result of chronic pulmonary regurgitation.^16,30 Despite the known role of fibrosis in arrhythmia development, it is unclear whether fibrosis volume and arrhythmia burden are correlated in rTOF.^16,31,32 In our cohort, the total volume of core scar and of border zone did not correlate with arrhythmia burden, indicating that the amount of fibrosis in the ventricles could not be used to stratify VT risk. This highlights the concept that the spatial distribution of scar and grey zone rather than the total volume of remodeled tissue is important in the development of arrhythmia.^33,34 Our computational methodology allows for unique representation of this relationship by incorporating patient-specific myocardial geometry and tissue structural and electrophysiological heterogeneity.

This is the first use of complex computational cardiac models in patients with congenital heart disease and structurally abnormal hearts. Our modeling pipeline notably accounted for the disease-specific fibrotic remodeling in rTOF, including fibrosis in both ventricles, as the role of RV remodeling in arrhythmogenesis in rTOF is well documented.^{5,15,16,29,30} Our patient-specific ventricular models also incorporated experimentally-observed ionic changes in cardiomyocytes from patients with interstitial fibrosis.²⁵ Simulated reentrant VT morphologies in the two patients with documented VT were located near the RVOT, a common location for clinical reentrant VT in rTOF patients,^29,32 validating the rTOF-specific aspects of the virtual-heart modeling platform. While the virtual-heart modeling platform is capable of identifying multiple VT morphologies, which are commonly observed in the LV of older patients with rTOF, we did not observe multiple morphologies in any patients in this study.^21,35

In addition to QRSd <180 ms,⁴ many other criteria have been proposed for VT risk stratification of adult patients with rTOF, including transannular patch use, age, year of repair, and age at repair. Prior studies have shown that the risk of VT increases with increasing age and time post-surgery.⁴ Notably, surgical techniques have continuously improved since their advent in 1955, so patients repaired with more modern techniques have lower VT risk.^1,2,6 In our study, there were no statistically significant differences in transannular patch use, age at MRI, year of repair, or age at repair between patients who did and did not experience clinical VT. This highlights the complexity of risk stratification in this cohort and the potential of our virtual-heart approach in providing personalized, mechanistically-guided VT risk assessment.

Our study was limited by the presence of imaging artifact, preventing us from creating models for all slow VT risk patients in the cohort, and by small sample size. Resultant artifact from the presence of surgical sternal wires was the most common reason for exclusion; however, sternal wires are not an automatic cause for exclusion if they do not result in artifact which obscures the myocardium or areas of fibrosis. In this first attempt at modeling structurally abnormal hearts, we strove to be conservative in our patient exclusion process in order to avoid introducing error in the first steps of our modeling approach. In our current thresholding approach, artifact might be construed as fibrosis and could potentially result in spurious VT induction in the model.

While the models incorporated patient specific geometry and fibrosis distribution, the electrophysiology was, by design, not personalized for each patient, as our virtual-heart approach is intended for non-invasive risk assessment. This grants our approach the advantage of not requiring any additional tests beyond the standard clinical LGE-MRI sequence acquired as part of contemporary clinical assessment of patients with rTOF. Additionally, electrophysiological changes in regions of fibrosis in the model were based on experimental recordings of myocytes from patients with HCM. While the histologic changes in rTOF and HCM are similar, the etiologies of the changes are different. Future work is needed to determine whether non-invasively including personalized, rTOF-specific regional variations in our models for patients with rTOF is possible and whether doing so would enhance risk stratification.

Conclusions

This proof-of-concept study used a virtual-heart computational approach to assess VT inducibility in rTOF patients deemed at low VT risk by clinical criteria. The virtual hearts constructed from LGE-MRI scans of seven rTOF patients predicted correctly, blind to the clinical outcome, reentrant VT in the models from VT-positive patients and no arrhythmia in those from VT-negative patients. The study demonstrated the potential value of non-invasive computational heart modeling for outcome prediction in patients with congenital heart disease. It is the first study to implement complex, organ-scale modeling of potential arrhythmogenic behavior in patients with structurally abnormal hearts following operative repair. The modeling pipeline uniquely incorporated fibrotic changes known to occur in both ventricles as well as the ion level alterations associated with diffuse fibrosis. We propose including virtual-heart risk stratification in the clinical decision-making process to identify at-risk patients who could be incorrectly stratified by the current clinical criteria. The predictive capability of the methodology will need to be prospectively validated in a larger clinical cohort.