Arrhythmogenic vulnerability of re-entrant pathways in post-infarct ventricular tachycardia assessed by advanced computational modelling

Abstract

Aims

Substrate assessment of scar-mediated ventricular tachycardia (VT) is frequently performed using late gadolinium enhancement (LGE) images. Although this provides structural information about critical pathways through the scar, assessing the vulnerability of these pathways for sustaining VT is not possible with imaging alone.

This study evaluated the performance of a novel automated re-entrant pathway finding algorithm to non-invasively predict VT circuit and inducibility.

Methods

Twenty post-infarct VT-ablation patients were included for retrospective analysis. Commercially available software (ADAS3D left ventricular) was used to generate scar maps from 2D-LGE images using the default 40–60 pixel-signal-intensity (PSI) threshold. In addition, algorithm sensitivity for altered thresholds was explored using PSI 45–55, 35–65, and 30–70. Simulations were performed on the Virtual Induction and Treatment of Arrhythmias (VITA) framework to identify potential sites of block and assess their vulnerability depending on the automatically computed round-trip-time (RTT). Metrics, indicative of substrate complexity, were correlated with VT-recurrence during follow-up.

Results

Total VTs (85 ± 43 vs. 42 ± 27) and unique VTs (9 ± 4 vs. 5 ± 4) were significantly higher in patients with- compared to patients without recurrence, and were predictive of recurrence with area under the curve of 0.820 and 0.770, respectively. VITA was robust to scar threshold variations with no significant impact on total and unique VTs, and mean RTT between the four models. Simulation metrics derived from PSI 45–55 model had the highest number of parameters predictive for post-ablation VT-recurrence.

Conclusion

Advanced computational metrics can non-invasively and robustly assess VT substrate complexity, which may aid personalized clinical planning and decision-making in the treatment of post-infarction VT.

Graphical Abstract

Open in new tabDownload slide

Introduction

Clinical work-up of scar-mediated ventricular tachycardia (VT) is frequently performed using left ventricular (LV) late gadolinium enhancement cardiac magnetic resonance (LGE-CMR) images.¹ Novel post-processing algorithms have facilitated comprehensive non-invasive assessment of the underlying arrhythmogenic substrate.^2,3 Recent studies have demonstrated the clinical utility of an LGE-based assessment in determining the risk of post-ablation VT recurrence.^4,5 Although LGE-CMR-based strategies provide novel insights into substrate features,⁶ characterizing the potential electrical instability (i.e. that may cause the initiation and sustenance of re-entrant arrhythmias) of the identified substrate is currently not possible based on imaging alone.

The recent evolution of computational algorithms and hardware has introduced a new field of in-silico cardiology.^7,8 Modelling and simulations have shown promise in proof-of-concept studies towards improved characterization of arrhythmogenic substrate using a patient-specific virtual-heart approach both in ablation target prediction^9–11 and prediction of sudden cardiac death risk.^12,13 It seems evident that anatomically personalized image-based modelling can be used to improve the substrate complexity estimation provided by LGE-CMR,¹⁴ as it intrinsically integrates known structural information (from imaging) with the simulation of functional electrophysiological dynamics.

However, there are concerns regarding the computational inefficiency of current approaches.⁹ The significant computational resources required to compute the simulations are incompatible for incorporation into a practical clinical workflow, with current approaches requiring days to simulate virtual induction protocols on external high-performance computing facilities. Recently, a novel reaction-Eikonal-based tool for Virtual Induction and Treatment of Arrhythmias (VITA) has demonstrated the possibility of automated, near real-time identification of potential sites of block and corresponding re-entrant circuit vulnerability.¹⁵ VITA utilizes patient-specific anatomical models to assess the electrical vulnerability of potential re-entrant channels and provides simulation metrics such as the number of inducible VTs and their corresponding round-trip-time (RTT) that may be compared with patient follow-up.

An integrated approach, combining the structural information from LGE-CMR augmented by additional functional information provided by VITA, may provide the clinician with a more useful characterization of the underlying arrhythmogenic substrate complexity. Such information may be vital in pre-procedural planning especially in patients where VT is non-inducible or not hemodynamically tolerated. In addition, these insights can improve the identification of patients who are more likely to suffer from VT recurrence following catheter ablation whose treatments may then be personalized appropriately.

The objective of this study was to systematically evaluate the clinical value of VITA-derived simulation metrics in the non-invasive assessment of VT substrate complexity and predicting post-ablation VT recurrence. Due to the importance of scar delineation in the construction of the LGE-CMR models, along with the lack of consensus over LGE segmentation thresholds, the influence of scar thresholds on computed metrics was also investigated to evaluate the robustness of VITA to slight variations in image pre-processing.

Methods

Patient population

Patients with an ischaemic cardiomyopathy having undergone VT-ablation with a pre-procedural LGE-CMR were identified using a retrospective search in the Amsterdam University Medical Centers catheter ablation database. All ablations performed in the period between 2016 and 2020, with high-quality electro-anatomical mapping (EAM) data and pre-procedural LGE-MRI without any ischaemic events in the interim, were included. Study procedures were in accordance with the Declaration of Helsinki and the study received approval from the institutional scientific board. Study subjects provided written informed consent to utilize data consisting of clinical characteristics such as follow-up and arrhythmia recurrence (defined as a sustained VT requiring implantable cardioverter-defibrillator (ICD) therapy or hospitalization, and therapies adjudicated by 1 physiologist and 1 device cardiologist). In addition, approval was provided to utilize anonymized LGE-CMR imaging data.

Image acquisition and post-processing

The imaging workflow has been described previously.⁴ Briefly, 2D LGE-CMR images were acquired 10–15 min after contrast agent injection on a 1.5-Tesla (T) MRI scanner (Avanto, Siemens Healthcare, Erlangen, Germany). These images were semi-automatically segmented using ADAS 3D LV to create LV endo- and epicardial contours (ADAS3D Medical, Barcelona, Spain) (Figure 1, panel A).² To evaluate the sensitivity of computational modelling to the selection of pixel signal intensity (PSI) thresholds, we conducted a sensitivity analysis by using a range of different PSI thresholds based on our previous study.⁴ Colour coded pixel signal intensity maps (PSI) were generated using a full width at half maximum algorithm for four different threshold configurations: 40–60% (default), where <40% is healthy tissue, 40–60% border-zone (BZ) and >60% is scar core, as well as for 35–65%, 30–70%, and 45–55%. The resulting four models, containing PSI-derived tissue characteristics, were post-processed using a combination of custom-written Python scripts together with an open-source meshing programme (Meshtool) to create volumetric meshes (Figure 1, panel B).¹⁶ Subsequently, fibre orientations were assigned within the myocardium using an established rule-based approach.¹⁷

Simulation workflow

Simulations of VT induction were performed using a previously developed, custom-developed tool for VITA.¹⁵ In summary, VITA represents a rapid tool for automatically identifying all potential anatomical VT circuits within an image-based ventricular model in near-real time. Initially developed for identifying targets for VT ablation, VITA has been compared and demonstrated to be non-inferior to other conventional reaction-diffusion (monodomain) simulation approaches.¹⁸

Briefly, the VITA protocol involves sequentially pacing the ventricle at 17 sites (1 per American Heart Association segment) using a fast Eikonal model. Isolated isthmuses through the scar are automatically identified as places where the paced wavefront ‘splits’ before recombining (Figure 1, panel C, top). Corresponding re-entrant pathways involving the identified isthmus are computed by simulating uni-directional block from the isthmus exit/entrance site (Figure 1, panel C, middle). The RTT (or cycle-length) of the electrical re-entry is then obtained for each identified circuit (Figure 1, panel C, bottom). A cut-off of 50 ms was used with only circuits above this threshold being deemed clinically viable. This provides the total number of VTs that are inducible in the model that are taken forward for further analysis. Finally, correlation coefficients were calculated on cyclically aligned activation time maps to assess uniqueness among very similar VTs and filter out duplicate VTs being induced from different pacing sites thereby providing the number of unique VTs. This resulted in the following four simulation metrics derived from each simulation (for all 17 simulation locations): total number of VTs induced from all pacing sites, number of unique VTs, mean RTT of all unique VTs (meanRTT), and maximum round-trip-time (maxRTT) of the unique VTs.

Ventricular ablation

As described previously,⁴ all antiarrhythmic drugs, except amiodarone, were discontinued before catheter ablation. During the ablation procedure, implantable cardioverter-defibrillators therapy was inactivated. Substrate mapping was performed using the CARTO electroanatomical mapping system (Biosense Webster) with a multielectrode catheter (Pentaray, Biosense Webster). VT induction was attempted using programmed ventricular stimulation at a basic drive cycle length of 500 ms with up to three extra stimuli. Ablation was performed specifically targeting sites with long stimulus to QRS and pace-mapping correlation (>11/12 correlation with clinical VT) mostly in conjunction with scar homogenization using irrigated catheters (Thermocool SmartTouch DF, Biosense Webster), the example provided in Figure 2. Non-inducibility post-ablation was examined using programmed ventricular stimulation. The total number of applied ablation lesions was quantified in every EAM. In addition, the total ablated area was computed by assuming a surface lesion diameter of 4 mm (corresponding to the catheter tip size). Any lesions that were within 75% of each other were considered to be fully overlapping and were therefore disregarded. Lesions that were partially overlapping, either by 25% or 50%, were taken into account by including the respective percentage of the lesion area.

Statistical analysis

Statistical analysis was performed using IBM SPSS Statistics (Version 26). Continuous variables were expressed as mean ± standard deviation. Baseline variables, cardiac MRI-derived late-enhancement characteristics for PSI 40–60, simulation metrics, and EAM ablation lesions were compared in patients with- and without recurrence using an independent T-test. Differences between the four LGE-CMR models for the four simulation metrics were quantified using repeated measures analysis of variance (ANOVA). Post-hoc analysis was performed with a Bonferroni correction. Pearson’s correlation was calculated to study the relationship between the quantified parameters for the 4 PSI configurations and arrhythmia recurrence. Receiver operator characteristic (ROC) analyses were performed to evaluate the sensitivity and specificity of all parameters. Optimal cut-off values with the highest sensitivity were derived using a J-statistic in parameters with an area under the curve (AUC) of at least >0.7. A P < 0.05 was considered to be statistically significant.

Results

Twenty patients were included in the study. Baseline patient characteristics are shown in Table 1. All patients were male with a mean age of 71.5 ± 9.8 years. Ten patients (50%) had recurrence during the follow-up period of 2.6 ± 1.6 following the ablation procedure with one or multiple VT recurrences: one patient (10%) had a self-limiting tachycardia and in the other cases ATP (n = 4, 40%) or shocks (n = 5, 50%) was required for termination. Patients with recurrence had a significantly lower ejection fraction compared to patients that remained free of arrhythmias (27.5 ± 9.7 vs. 35.2 ± 5.6, P < 0.05). Both end-diastolic volume and end-systolic volume were significantly higher in the group with recurrence (Table 1). Scar core did not show a significant difference between the two patient groups (14.4 ± 7.2 vs. 9 ± 6.8, P = 0.11). However, patients who experienced VT recurrence had a significantly higher borderzone weight (23.5 ± 12.4 vs. 14 ± 6.5, P = 0.04) and corridor weight (6.3 ± 4.3 vs. 2.9 ± 2.3, P = 0.04) compared to patients without recurrence. Further analysis using ROC revealed a good predictive value for post-ablation VT recurrence, with an AUC of 0.800 (P = 0.023) for BZ and an AUC of 0.780 (P = 0.034) for corridor weight.

Simulation-based recurrence prediction

VT was successfully induced in 19 cases (95%). In one case (5%) VITA was unable to induce VT in any of the four models i.e. PSI 45–55, 40–60, 35–65, and 30–70.

Significant differences were observed between the recurrence and non-recurrence groups in VITA metrics for the default PSI 40–60 model. Patients with VT-recurrence had a higher number of total VTs (85 ± 43 vs. 42 ± 27, P = 0.01) and unique VTs (9 ± 4 vs. 5 ± 4, P = 0.04) compared to patients free of recurrence after ablation (Figure 3). Mean RTT showed a slight (non-significant) increase for patients with recurrence (173 ± 37 ms) compared to patients without (155 ± 103 ms). Maximum RTT demonstrated a trend towards significance of 293 ± 90 ms vs. 200 ± 114 ms (P = 0.06) for recurrence and non-recurrence, respectively. Total VTs had the highest predictive capability, analysed using ROC, demonstrating an AUC of.820 (P = 0.02). Unique VTs were also predictive of recurrence with an AUC of 0.770 (P = 0.04). An example of these metrics is provided in Figure 4 which illustrates the differences in two patients, one with and another without recurrence.

Impact of thresholds on simulation metrics

The number of VTs, unique VTs and mean RTT were relatively consistent for all different PSI models. ANOVA testing did not show any differences for these three metrics between the four different models (Figure 5). Maximum RTT (F = 4.502, P = 0.007) did show a significant difference between the four models. However, after post-hoc correction using Bonferroni adjustment, only a trend towards significance remained for maximum RTT (slightly longer re-entrant pathways) for the PSI 30–70 model.

Across the different threshold models, total and unique VTs were significantly different between the recurrence and non-recurrence groups, except for PSI 30–70 where the difference between both VT measures did not reach significance (Figure 3). Mean RTT could only differentiate between recurrence and non-recurrence in the PSI 30–70 category (216 ± 45 ms vs. 155 ± 68 ms, P = 0.03). The other PSI models demonstrated comparable mean RTT values between the two groups. A significantly longer maximum RTT was observed in the PSI 45–55 model for the recurrence group compared to patients that were free from recurrence (300 ± 103 ms vs. 191 ± 114 ms, P = 0.04). For the other two models (PSI 35–65 and PSI 30–70) minor differences were observed between max RTT that did not reach statistical significance.

Simulation metrics had varying reliability (J-statistic) in predicting recurrence. Total and unique VTs were predictive of recurrence for two of the additional three scar thresholds i.e. PSI 45–55 and PSI 35–65 (Figure 3). The accuracy for total VTs in models with PSI 45–55 and 35–65 (AUC = 0.805, P = 0.02) was only slightly less reliable compared to the default PSI 40–60. The number of unique VTs also remained a strong predictor of recurrence with an AUC of 0.765 and 0.790 for PSI 45–55 and 35–65, respectively. Overall, models with threshold PSI 45–55 could utilize the largest number of metrics to predict post-ablation recurrence (3 out of 4) with total VT, unique VT, and maximum RTT demonstrating a significant relation with recurrence.

EAM derived lesions

A total of 16 cases had usable EAM data. Overall, an average of 41 ± 22 lesions was applied during the ablation procedure with an average ablated area of 477 ± 243 mm². More extensive ablation, signified by a larger number of ablation lesions, was observed in patients with post-ablation recurrence (n = 9, 50 ± 22) vs. non-recurrence (n = 7, 29 ± 15) (P = 0.053). A strong positive correlation was observed for total ablation lesions (r = 0.500, P = 0.048) with unique VTs for PSI 40–60. This correlation was stronger for wider PSI ranges and non-significant for PSI 45–55. In this group, only maximum RTT demonstrated a strong correlation with the number of ablation lesions (r = 0.531, P = 0.034) and ablated area (r = 0.526, P = 0.037). In patients with post-ablation recurrence, the ablated area was significantly higher compared to those without recurrence (586 ± 241 vs. 337 ± 171 mm², P = 0.037). Furthermore, a strong correlation was observed between unique VTs and the ablated area for PSI 35–65 (r = 0.695, P = 0.003) and PSI 30–70 (r = 0.595, P = 0.015). Additionally, the maximum RTT metric (PSI 45–55) demonstrated a strong correlation with the ablated area (r = 0.547, P = 0.028), similar to the observation for the total number of lesions.

Discussion

This is the first study to use a clinically applicable, computational path-finding tool to predict substrate complexity in relation to VT recurrence and to demonstrate the impact of imaging thresholds. The main findings are (i) VITA metrics, borderzone, and corridor weight enables prediction of post-ablation arrhythmia recurrence with comparable accuracy, (ii) the overall number of total and unique VTs are not susceptible to changes in scar thresholds, (iii) the number of unique VTs is strongly correlated to the complexity of substrate (derived from total ablation lesions), and (iv) simulation metrics derived from models with scar threshold PSI 45–55 have the highest overall predictive value for VT recurrence following catheter ablation.

Complexity of substrate

Computational modelling is increasingly being applied towards improving risk prediction of malignant ventricular arrhythmias and sudden cardiac death^10,12,19,20 as well as proof of concept studies demonstrating the potential of simulations to guide catheter ablation.^9,21 However, despite their proposed benefit, current computational approaches can take several days to compute and require costly high-performance computing facilities. Due to the high computational cost, many of these in-silico approaches have not been translated into clinical application. VITA is a more efficient method which can perform near real-time simulations on a regular desktop computer and can therefore be easily integrated and applied as part of a routine clinical workflow. In addition, VITA metrics add a new dimension to the utility of modelling by investigating the relation between scar-complexity, derived from a number of automatically computed VITA simulation metrics, and post-ablation VT recurrence.

Previous studies have demonstrated that patients presenting with multiple VT morphologies are likely to suffer from VT recurrence following catheter ablation.²² While the VT recurrence observed in our patient cohort could potentially be associated with ablation outcome, i.e. VT inducibility following catheter ablation, it is important to note that the majority of patients in both the recurrence and non-recurrence groups had no VTs inducible after ablation. However, patients with VT recurrence did have a more complex substrate, as evidenced by a significantly higher borderzone (23.5 ± 12.4 vs. 14 ± 6.5, P = 0.04) and corridor weight (6.3 ± 4.3 vs. 2.9 ± 2.3, P = 0.04). Therefore, these findings are likely due to a higher complexity of the underlying substrate rather than differences in ablation outcome. These findings are consistent with our previous study, which evaluated imaging metrics in a similar dataset.⁴ Collectively, these findings further support the notion that the complexity of the underlying substrate is a key determinant of VT recurrence.

Building on these insights, this study also examined the functional impact of an imaging-derived substrate and discovered significant differences in simulation metrics between patients with and without VT recurrence. Depending upon the threshold utilized, patients with recurrence had a significantly higher number of total and unique VTs (Figures 3 and 4). In addition, the induced VTs in this patient group were shown to have a larger RTT which was considered to be an indication of the size of the arrhythmia circuit. A larger circuit could also be more susceptible to different possible wavelengths i.e. more susceptible to action potential duration and conduction velocity changes (autonomic nervous system or anti-arrhythmic drug related). These findings are likely attributed to a more ‘complex’ substrate in patients with post-ablation VT recurrence, as confirmed by EAM-derived invasive data revealing a larger number of total ablation lesions applied in these patients.

In the last decade, attention has been focused towards predicting arrhythmia recurrence following atrial fibrillation and VT ablation.^23–25 Multiple studies have demonstrated the utility of VT inducibility following catheter ablation as a parameter to identify patients at risk for VT-recurrence.^23,25 Recently it has been shown that despite negative electrophysiological study (EPS) a large group of patients can develop post-ablation arrhythmia recurrence, with almost 80% of arrhythmia recurrence caused due to new VTs.²⁶ This may be caused by re-entrant circuits arising from areas of substrate that were insufficiently modified, e.g. transient conduction block due to ablation-induced oedema, or new circuits, formed due to shortcomings in the ablation strategy.

Although imaging-derived metrics and simulation metrics demonstrated comparable accuracy in this study, this work serves as an important proof-of-concept. It is expected that the application of VITA in larger cohorts will provide a more comprehensive understanding of the functional nature of arrhythmogenic substrate complexity, and augment findings from imaging-derived metrics such as BZ weight. VITA provides a non-invasive solution for both complexity assessment as well as recurrence prediction problems by thoroughly identifying and evaluating all potential re-entrant circuits in the patient-specific model thereby performing an in-silico EPS. As demonstrated by the current results, VITA metrics can be utilized to identify patients with a complex substrate (high risk of recurrence) and subsequently tailor the ablation strategy. For instance, an endocardial ablation may be sufficient for patients with relatively simple substrate, whereas an epicardial/stereotactic radiotherapy approach may be necessary for more complex cases. In addition, a combined workup using LGE-based scarmaps and VITA can be used to predict ICD therapy and could facilitate early identification of high-risk patients in addition to left ventricular ejection fraction and has the potential to improve risk-stratification for ICD implantation. These findings support the incorporation of VITA-derived simulation metrics in the work-up of VT(-ablation) to perform a non-invasive assessment of substrate complexity.

Impact of thresholds on simulations

A recent study from our group demonstrated that scar threshold selection, during the segmentation of LGE-CMR images and the creation of patient-specific ventricular models, has a significant impact on the quantification of BZ and scar core.⁴ As computational modelling extracts the functional implications of the anatomical structure from such personalized segmentation as input, it can be hypothesized that simulation metrics will be influenced by threshold variations. To date, due to the heavy computational effort required, only one study has investigated this aspect of image post-processing.²⁷ All subsequent studies have used a fixed threshold to segment and characterize scarred myocardium.^3,6,10,18,28 This study evaluated a cohort of ischaemic cardiomyopathy patients and demonstrated that the overall impact of thresholds on the number of virtually generated VTs is limited. As expected, there is a slight, but significant increase in the RTT for PSI 30–70 which can be explained by the presence of a larger area of borderzone and the resulting slower conduction in this region. Although PSI 45–55 demonstrated the highest accuracy in predicting recurrence, overall, VITA metrics performed well despite variations in the applied threshold. Therefore, it can be advocated that VITA is robust against varying LGE-CMR thresholds and can reliably perform arrhythmic risk stratification and scar complexity analysis for varying scar thresholds.

Limitations

Future studies will aim to integrate VITA as part of the clinical ablation workflow, enabling the important direct comparison between re-entrant circuits computed using VITA and clinically identified ablation targets. One of the most important parameters in circuit computation using VITA is the RTT as that defines the smallest circuit identified as a clinically-viable isthmus. The value chosen in this study (50 ms) could be considered too low to sustain VT, which normally has an upper-limit of the cycle-length around 200 ms. However, the spectrum of circuits identified between 50 and 200 ms could enable an existing arrhythmic focus such as an ectopic premature ventricular contraction or non-sustained VT to deteriorate into ventricular fibrillation. In addition, the study aim was to identify overall complexity of the substrate i.e. identify all possible channels even those that might not be clinically relevant at this stage of the disease but potentially become arrhythmogenic following catheter ablation of the ‘main’ pathways. The prospective application of VITA for procedural guidance will enable further insights into the therapeutic implications of identified circuits and be useful in determining an appropriate RTT cut-off range. An additional limitation is the retrospective nature of this study, which introduces variability in the follow-up duration. This variation can influence the detection of VT recurrence and potentially result in an overestimation of recurrence events. Future (prospective) studies will utilize a standardized follow-up period to better control for the potential impact of follow-up duration on VT recurrence.

Conclusion

In-silico VT simulation metrics can be used to non-invasively assess VT substrate complexity and predict post-ablation arrhythmia recurrence. Variations in scar thresholds during post-processing of LGE have a limited functional impact, and the overall number of total and unique VTs is not susceptible to changes in scar thresholds. Metrics simulated in meshes using PSI range 45–55 have the highest predictive value for post-ablation VT recurrence. Prospective studies should aim to include VITA as part of the clinical VT work-up to aid decision-making and guide catheter ablation.

Funding

This work was supported by an EACVI Research Grant to Dr. Bhagirath. The research was also supported by an Academy Van Leersum grant of the Academy Medical Sciences Fund (Royal Netherlands Academy of Arts & Sciences), Netherlands Heart Institute Fellowship, CVON PREDICT2 Young Talent Program, NIHR Biomedical Research Centre and CRF at Guy’s and St Thomas’ NHS Foundation Trust and King’s College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health. This work was also supported by the Wellcome Trust, Wellcome EPSRC Centre for Medical Engineering at King’s College London (WT 203148/Z/16/Z), and a Wellcome Trust Innovator Award to Dr. Bishop (213342/Z/18/Z) and British Heart Foundation Project Grant (PG/18/74/34077). For the purpose of open access, the author has applied a CC BY public copyright license to any Author Accepted Manuscript version arising from this submission.

Data availability

The data that support the findings of this study are available from the corresponding author, PB, upon reasonable request.

References

Author notes