Up Digital and Personal: How Heart Digital Twins Can Transform Heart Patient Care

Abstract

Precision medicine is the vision of healthcare where therapy is tailored to each patient. As part of this vision, digital twinning technology promises to deliver a digital representation of organs or even patients, using tools capable of simulating personal health conditions and predicting patient or disease trajectories based on relationships learned both from data and from biophysics knowledge. Such virtual replicas would update themselves with data from monitoring devices and medical tests and assessments, reflecting dynamically the changes in our health conditions and the responses to treatment. In precision cardiology, the concepts and initial applications of heart digital twins have slowly been gaining popularity and the trust of the clinical community. In this article, we review the advancement in heart digital twinning and its initial translation to the management of heart rhythm disorders.

Introduction

Precision medicine is envisioned to provide therapy tailored to each patient. The rapidly increasing ability to capture extensive patient data, coupled with machine learning, a powerful tool for processing massive amounts of data and identifying correlations in it, is a pathway to achieving this vision. A different pathway towards precision medicine is the increasing ability to encode known physics laws and physiology knowledge within mathematical equations and to adapt such mechanistic models to represent the behavior of a specific patient.

The expectation is that it will be highly beneficial to have a digital representation of ourselves. A digital doppelgänger that is tailored to represent our own unique physiology, structure, biophysical processes and even diseases could allow healthcare professionals to simulate our personal medical history and health conditions using relationships learned both from data and from biophysics knowledge. That virtual replica of ourselves would integrate data-driven learning and multiscale physics-based modeling and will update itself, either continuously with data from monitoring devices, or intermittently with data from healthcare provider visits and tests. It will thus reflect the changes in our health conditions due to interactions with the environment, changes in lifestyle, and changes in response to medical interventions These digital twins (DTs) would forecast the trajectory of a patient’s disease, estimate risk of adverse events, and predict treatment response so that the potential outcome would inform treatment decision.

A DT is a virtual replica of a physical object, person, or process that can be used to simulate its behavior. This dynamic model represents both the components of a system and their ongoing interaction. Digital twinning is not a new concept – digital twins have been used to replicate many real-world entities, from equipment lifecycles through to entire manufacturing processes¹, as it allows one to oversee the performance of an asset, identify potential faults, and make well-informed decisions about maintenance and global performance. In health-care, the DT represents the vision of a virtual tool that integrates dynamically clinical and/or tracked data acquired over time for an individual and predicts behavior using comprehensive multi-scale mechanistic simulations based on physiology knowledge and physics laws.

In precision cardiology, over the last decade, the concepts and initial applications of heart DTs have slowly been gaining popularity and the trust of the clinical community. Pathways forward have been charted^2,3, and current developments^4–7 have been assessed. Fig. 1 presents a flowchart of the clinical workflow using heart DTs, and their envisioned applications for clinical decision support in diagnosis and prognostication of the patient’s disease trajectory, as well as for guiding treatment that is optimized by considering the patient’s response to the potential therapy.

Figure 1:

Flowchart of the clinical workflow using heart DTs.

One of the most advanced heart DT applications is in the management of heart rhythm disorders, with major developments in the field reviewed in this article. The present article is the review of the body of work of each year’s recipient of the Gordon Moe lectureship presented by the Cardiac Electrophysiology Society. Dr. Trayanova was this year’s recipient, and this review article summarizes her work and that of her team at Johns Hopkins University on the development and application of heart DTs for the management of atrial and ventricular arrhythmias.

Digital Twins for Atrial Arrhythmia Management

Atrial fibrillation (AF) is the most common heart rhythm disorder, having a prevalence of 1–2% worldwide. It is associated with embolic stroke, heart failure, and cardiovascular hospitalization and death⁸. AF patients have increased rates of cognitive impairment and a diminished quality of life⁹. The growing burden of AF, and the high rate of healthcare expenditures associated with its management have led to a large body of basic and clinical research aimed at uncovering both a permanent cure of AF and an improved AF management strategy. Computational modeling of the atria and personalized atrial DT technologies have been and continue to be an integral part of these developments.

An important focal point in these efforts, including ours, has been the representation, in personalized DTs, of the fibrotic remodeling that takes place in the atria of (ageing) patients, and particularly in those experiencing the persistent form of AF (PsAF). In these patients, the mechanisms giving rise to AF shift from electrical abnormality in the pulmonary veins (PVs) to re-circulating electrical waves (reentries) perpetuated by the fibrotic substrate. Personalized DT technologies have provided understanding of how fibrotic remodeling results in the turbulent propagation associated with AF and have suggested management strategies, some of which have been tested in prospective studies. Prior to this review, we and others have reviewed different aspects of computational modeling of the human atria^6,10–12.

In developing a personalized atrial DT that represents the patient-specific fibrosis remodeling, the process commences with the assessment of the patient’s clinical contrast enhanced (late gadolinium enhancement, LGE)-MRI scan, typically a 3D scan of higher resolution to resolve the thin atrial walls^13,14. In addition to providing information about atrial shape, MRI imparts excellent heart tissue characterization, as gadolinium-based contrast agents accumulate in scar and fibrotic tissue¹⁵. Areas on LGE-MRI correspond to areas of scar and fibrotic remodeling, with the high image intensity corresponding to deep scar. Reconstruction of atrial geometry requires segmenting the chambers and the remodeled tissue from the LGE-MRI; the study of McDowell et al¹⁶ was the first to reconstruct a personalized DT incorporating fibrotic remodeling. The threshold for fibrosis segmentation is not well established and remains controversial. We have used a version of the image intensity ratio (IIR) since it uses ratiometric values instead of raw voxel intensities. Our team also reconstructed, for the first time, the personalized fibrotic remodeling in the right atria (RA) of patients; we adapted the IIR metric for RA fibrosis¹⁷. Fig.2A presents a number of reconstructed bi-atrial models of PsAF patients with fibrosis. Additionally, atrial DTs incorporate fiber orientations to ensure realistic conduction patterns. Reconstruction of personalized atrial DTs in our translational research has involved the use of atlas human fiber orientations¹⁸, acquired from explanted human hearts using diffusion-tensor MRI (Fig.2B), which visualizes the fiber tracts in the myocardium. These are then assigned in the patient-specific geometric model with the use of a universal atrial coordinate system¹⁹.

Figure 2:

Developing personalized atrial DTs. A. Reconstructed bi-atrial geometries from 6 PsAF patients. Fibrosis distribution is in green. B. Human fiber orientation acquired from explanted human hearts and used as fiber atlas for DT construction. Image modified with permission from⁶⁸.

In our translational work aimed at improving AF management, personalization of atrial DTs was done predominantly based on the patient-specific disease remodeling, where distinct electrophysiological (EP) properties are assigned in different regions based on image intensity. This has been a deliberate choice, as our intention has been strategically, down the road, to develop non-invasive technologies for the prognostication and treatment of AF. Several research groups have instead chosen to personalize the EP properties from invasive intra-procedural measurements^20,21. Instead, we developed, based on clinical measurements in AF patients, a set of baseline EP properties²² which could be modified, when needed, to explore different arrhythmogenic mechanisms. These included, but were not limited to, exploration of the effect of potential myocyte-fibroblast interactions^23,24, and arrhythmogenesis related to calcium-driven alternans in AF^25,26. For instance, in the latter study, we found that elevated Ca2+ alternans propensity due to decreased ryanodine receptor inactivation, and development of repolarization alternans at slower heart rates, resulted in increased ectopy-induced arrhythmia vulnerability, complexity, and persistence due to increased repolarization heterogeneity and wavebreak.

Our atrial modeling studies strongly support the notion that that the extent and distribution of atrial fibrosis are critical determinants of AF initiation, maintenance, and re-entrant driver dynamics during AF. Although presence of a certain amount of fibrosis is sufficient for initiation of AF in simulations, we found that patient-specific fibrosis distribution determines re-entrant driver dynamics²⁷. The study of Zahid et al.²⁸ took this understanding further, demonstrating, in a cohort of 20 patients, that reentrant drivers induced in the fibrotic substrate by rapid pacing persist only in areas with highly specific spatial patterns of fibrosis. The study was the first to construct atrial digital twins of both LA and RA. Fibrotic spatial patterns were characterized by calculating, from the 3D LGE-MRI images, maps of fibrosis density and fibrosis entropy. Local fibrosis density indicates the proportion of fibrotic elements among all elements surrounding the given location, while local fibrotic entropy quantifies the degree of disorganization between fibrotic and non-fibrotic elements in the local neighborhood. All the reentrant drivers that could be induced in each remodeled substrate persisted in fibrotic boundaries zone characterized by high fibrotic density and fibrotic entropy. Fibrotic patterns with such specific regional fibrosis metrics correspond to atrial areas with a high degree of intermingling between fibrotic and non-fibrotic tissue. The findings of this atrial DT study were subsequently validated²⁹ with clinical data (electrocardiographic imaging, ECGI).

Potentially the most transformative clinical application of atrial digital twinning is the development of personalized PsAF ablation strategies, tailored to each patient’s unique (fibrotic) atrial substrate. While pulmonary vein (PV) isolation (PVI), the electrical isolation of PV arrhythmia triggers, is the standard of care for patients with symptomatic AF^30,31, in PsAF patients with atrial fibrosis, AF recurrence rates after PVI are high^32,33, resulting in freedom from AF of only 40–50% one-year post-procedure. The presence of regions of fibrosis that extend beyond the traditional wide-area PVI and have arrhythmogenic propensity could explain PVI’s ineffectiveness in patients with atrial fibrosis. Establishing non-invasively whether PVI will result in subsequent AF recurrence in a patient with atrial fibrosis, and if so, determining prior to the ablation procedure the custom-tailored extra-PVI ablation targets that will deliver long-term freedom from AF could result in dramatically improved treatment efficacy and reduce the need for repeated ablations. In this, atrial digital twinning has found it calling.

McDowell et al.²⁷ was the first study to demonstrate, as a proof-of-concept, that LA digital twins reconstructed from the patient’s LGE-MRI scans (n=4) can be used to predict ablation targets in the fibrotic substrate. The targets were the locations in the fibrotic substrate where reentrant drivers (rotors) form following rapid pacing from locations in the substrate. Executing this virtual ablation strategy in the LA digital twins rendered the atrial model non-inducible for reentry. Having demonstrated the potential utility of atrial DTs in predicting ablation targets, we also assessed whether the predicted targets would be affected by different baseline cellular EP properties. In a sensitivity analysis³⁴, we varied atrial action potential duration (APD) or conduction velocity to address this question. These changes resulted in different likelihoods that a location in the fibrotic substrate would sustain a re-entrant driver. However, Hakim et al.³⁵ demonstrated that this uncertainty was mitigated by first executing re-entrant driver ablations followed by repeat inducibility tests to evaluate the occurrence of any emergent re-entrant drivers post-ablation. In other words, depending on the baseline EP properties in the patient’s atria, locations in the fibrotic substrate will give rise to rotors either on the first inducibility test, or on the subsequent inducibility tests probing for emergent rotors post-initial ablation, with the sequence determined by the patient’s EP properties. Thus, as long as both initial and emergent drivers were captured, the digital twin with the baseline AF EP properties would be a useful clinical tool to provide personalized guidance in AF ablation.

Boyle et al. pioneered a prospective ablation study for patients with persistent AF and fibrosis entirely guided by personalized atrial DTs¹⁵. In this landmark study termed OPTIMA (OPtimal Target Identification via Modelling of Arrhythmogenesis), 10 patients were enrolled. The locations of reentrant drivers were determined following a reentrant driver inducibility test. These locations in the DT atrial substrate were then ablated virtually. In essence, the OPTIMA DT approach for guiding AF ablation is a targeted substrate ablation approach, where locations in the fibrotic substrate capable of sustaining reentrant drivers are discerned and ablated to eliminate that capability. Following ablation, the DT inducibility test was repeated, and if new locations arouse capable of sustaining reentrant activity in the new (fibrosis plus initial ablation) substrate, then these emergent activities were also targeted with ablation, until a set of optimal ablation targets was found that results in a complete arrhythmia non-inducibility of the substrate. The proposed ablation targets were then used to steer patient treatment during the procedure, eliminating not only the clinically manifested AF but also any potential emergent AF drivers. Fig.3 shows the OPTIMA flowchart, illustrated with one of the patients in the prospective clinical study. While very successful, the efficacy of this study is now being tested in an FDA-approved randomized controlled clinical trial (ClinicalTrials.gov ID NCT04101539). Atrial DTs have also been used to determine how to ablate patients with atypical flutter³⁶. The use of atrial DT-driven ablation has also been compared to other approaches, such as focal impulse and rotor mapping (FIRM) or ECGI^37–39.

Figure 3:

Flowchart of the OPTIMA approach using personalized heart DTs. The individual steps are illustrated with the DTs of one of the participants in the study. LGE-MRI scans were used to construct the patient’s bi-atrial geometry and fibrosis distribution. Following a baseline inducibility test, one location at the left atrial anterior septal wall was determined to have a high likelihood of sustaining a rotor (bottom left). Following virtual ablations targeting the detected location, a repeat inducibility test identified an emergent rotor location at the right atrial posterior region (bottom middle), which was then ablated. The final optimal ablation targets resulted in a complete arrhythmia non-inducibility of the substrate. The proposed targets were imported to the CARTO system to guide ablation procedure.

Inadequate modification of the atrial rotor-sustaining fibrotic substrate may explain AF recurrence following failed PVI. In a retrospective longitudinal study of 12 AF patients with pre- and post-ablation LGE-MRIs, the study by Ali et al.⁴⁰ aimed to evaluate, using LA atrial digital twins, the post-ablation changes in the arrhythmogenic substrate, and to establish whether failure of AF ablation resulted from inadequate termination of pre-ablation rotors or emergence of new rotors post-ablation. The research demonstrated that recurrent AF after PVI in the fibrotic atria may be attributable to both the existence of locations in the substrate capable of sustaining rotors that were not modified/eliminated by ablation, and the emergence of new rotor-sustaining locations following ablation. The same levels of fibrosis entropy and density that underlie propensity to rotor formation in the pre-ablation substrate hold true for the post-ablation substrate as well, providing a uniform framework to understand fibrosis-induced arrhythmogenesis. These conclusions led to the development of a strategy to predict, pre-procedure before PVI, which patients are most likely to experience AF recurrence after PVI. To achieve that, Shade at al.⁴¹ combined LGE-based atrial digital twinning with machine learning in a proof-of-concept study of 32 patients. The algorithm used an input results of rotor induction simulations in the fibrotic substrate with imaging features derived from pre-PVI LGEs (Fig.4). The ML classifier to predict the probability of AF recurrence post-PVI achieved average validation sensitivity and specificity of 82% and 89%, respectively, and a validation area under the curve of 0.82. The study presented a highly generalizable AF recurrence predictor, despite the small training data set.

Figure 4:

Predicting AF recurrence post-PVI by using the results of simulations with atrial DTs, and training a machine learning classifier to predict, pre-procedurally, clinical outcome. Features in DT simulation results can be extracted in 2 ways, deductively and inductively. Modified with permission from⁴¹.

With the further development of the DT technology, the hope is that outstanding questions pertaining to AF management will also be addressed. These include why some patients with fibrosis never develop AF, whether PVI will be the required strategy if ablation lesions in the fibrotic substrate eliminate its ability to sustain reentrant activity, or what are the important characteristics of the substrate that render it arrhythmogenic in patients with AF.

Digital Twins for Ventricular Arrhythmia Management

Personalized ventricular DTs have also made significant contributions towards being part of the clinical management of ventricular arrhythmias (VAs). Such DTs have been constructed and utilized for a number of cardiomyopathies, both ischemic and non-ischemic. Construction of ventricular DTs follows the general steps as outlined for atrial DTs, with the difference that in addition to using LGE-MRI for model construction, other imaging modalities have been also used, often developing personalized DT by fusing the different clinical images to represent different types of structural remodeling and functional remodeling (such as for instance, inflammation). Fig. 5A shows ventricular DTs reconstructed from various imaging modalities. Frequently, before being used to suggest clinically-relevant decisions, the personalized DTs have been validated with clinical data (Fig. 5B). Below we review two major clinical applications of computational modeling, the prediction of SCD due to arrhythmias in various diseases, and the use of computational modeling to advance and ultimately provide guidance in arrhythmia treatment by catheter ablation.

Figure 5:

A. Personalized ventricular DTs with the corresponding predicted VTs. Shown here are the DTs of patients with ischemic cardiomyopathy (upper box) reconstructed from LGE-MRI (top)⁶¹ and CT (bottom)⁶³ scans. The lower box shows the DT of a patient with sarcoidosis which incorporates the region with inflammation detected from PET scan (top). Modified with permission from⁶⁹. Post-contrast T1 mapping scans were used to personalize the detection of diffuse and dense fibrosis in the DT reconstruction of a patient with hypertrophic cardiomyopathy (bottom of lower box). Modified from⁵⁴ (permission exempt per https://elifesciences.org/terms). B. Validation of the personalized DT of a patient with sarcoidosis with clinical ablation data. Two in silico VTs were induced in the DT (white arrows). The red spheres show the location of the clinical ablation lesions recorded in the patient electro-anatomical mapping (EAM) data during the procedure. The co-registered EAM data to the DT show the correspondence between the predicted VTs and the clinical ablation lesions. Modified from⁵⁶ (permission exempt per https://creativecommons.org/licenses/by-nc/4.0/).

Personalized ventricular DTs studies have been compelled to address the issue of predicting risk of sudden cardiac death (SCD). It is an important clinical issue, as worldwide, the prevalence of SCD already high, is on the rise; it results predominantly from VAs, particularly among patients with prior heart disease. Accurate SCD risk assessment is thus crucial, to enable primary prevention of SCD via the deployment of implantable cardioverter-defibrillators (ICD)⁴². Currently, the decision is based on a single clinical metric, which is not sensitive⁴³. Thus, many patients receive ICDs without deriving any health benefit⁴⁴ (90–95% of implanted ICDs are never utilized), whereas others are not protected, dying suddenly in the prime of their life. The conventional approach to SCD risk stratification has been to search for bio-markers that correlate with increased SCD risk. However, thus far, no biomarkers have enabled an accurate SCD risk assessment. Thus, inadequate SCD risk assessment poses a large public health and socioeconomic burden, and remains a major unmet clinical need.

The study by Arevalo et al.⁴⁵ demonstrated the first utilization of DTs of a cohort of ischemic cardiomyopathy patients (n=41) to determine the patients’ propensity to develop infarct-related ventricular arrhythmias and SCD. All patients were with reduced left ventricular (LV) ejection fraction (LVEF<35%), this deemed of high SCD risk, and all underwent ICD deployment. The patient’s risk was assessed based on whether arrhythmia was inducible from any of the numerous pacing sites tested in the DT; if it did, the patient was deemed at high risk. The comparison of the predictive capabilities of this DT approach with those of other clinical risk assessment metrics, including LVEF and other imaging variables, revealed that only the outcome of the heart DT was significantly associated with arrhythmic risk in this patient cohort. The study demonstrated that DTs of patients with prior infarction can be used to determine which patients should have a prophylactic ICD implantation for primary prevention. Additionally, in a small proof-of-concept study, SCD risk was investigated in a small cohort of myocardial infarction patients with preserved ejection fraction with the DT results matching clinical outcome⁴⁶. A more complex approach to the development of DTs of patients with prior infarcts was recently presented⁴⁷, where the patient heart DT also incorporated the distribution of penetrating adipose tissue (fat), which develops in infarcts >3 years old. This was a two-center prospective clinical-computational study, where enrolled patients underwent both LGE-MRI and CT scans (n=24). The penetrating adipose tissue (inFAT) was reconstructed in the DT from the patient computed tomography (CT) scans. The hybrid CT-MRI heart DTs, combined with EAM data, revealed that for these infarcts, inFAT exhibits greater pro-arrhythmic electrophysiological abnormalities than scar, and that it is the primary driver of substrate arrhythmogenic propensity. Subsequent clinical studies confirmed these DT predictions^48–50. Finally, in addition to further developing heart DTs that are based on mechanistic considerations, recently, new deep learning on multi-modality data has been proposed⁵¹. The deep learning analysis (termed SSCAR) combined learning on raw patient LGE-MRI scans and clinical covariates of patients with ischemic cardiomyopathy and was imbedded in survival analysis to predict time to SCD over a period of 10 years; it performed well on the external validation dataset demonstrating generalizability. Algorithms like SSCAR are paving the way for multi-modal prediction of patient outcome, but they will become particularly powerful when combined with DT for interpretability and mechanistic insight.

For SCD risk prediction in patients with non-ischemic cardiomyopathies, several heart DT studies have demonstrated the clinical utility of the approach. The first two applications of personalized DTs in non-ischemic cardiomyopathy were in pediatric patients. The first assessed VA risk in acute pediatric myocarditis⁵². In the second study, DTs were constructed from patients with repaired Tetralogy of Fallot⁵³ the childhood surgical intervention in these patients led to potential arrhythmogenic scarring in the heart. DT risk assessments in rToF predicted high risk, later validated by clinical outcome, in those patients where an ECG-based clinical algorithm predicted low risk. In hypertrophic cardiomyopathy (HCM), a common genetic disease characterized by thickening of heart muscle, high SCD risk arises from the proliferation of fibrosis in the heart. The study by O’Hara et al.⁵⁴ used DT technology to analyze how disease-specific remodeling promotes arrhythmogenesis and to develop a personalized strategy to forecast risk of arrhythmias in these patients (n=26). The authors combined LGE-MRI and T1 mapping data to construct fusion DTs that represented the patient-specific distribution not only of dense scar but also of diffuse fibrosis, the latter reconstructed from the T1 maps.⁵⁵ Analysis demonstrated that the presence of diffuse fibrosis, which is rarely assessed in these patients, increases VA propensity. In forecasting future arrhythmic events in these patients, the DT approach significantly outperformed current clinical risk predictors; both the ACCF/AHA and the ESC risk models offered prognoses that were inferior in accuracy, sensitivity, and specificity than the LGE-T1 DT prognosis. Another non-ischemic cardiomyopathy associated with high SCD risk and difficult risk prediction is cardiac sarcoidosis. Shade et al.⁵⁶ developed a two-step prediction approach, combining digital twinning with machine learning in a study of 45 patients. The patient’s arrhythmogenic propensity was assessed using a novel hybrid DT, reconstructed from the fusion of LGE-MRI and positron-emission tomography (PET) scans. The results from DT simulation were fed, together with a set of clinical biomarkers, into a supervised classifier – and the technology well outperformed current clinical decision making.

Finally, a genotype-specific heart DT (Geno-DT) approach was recently developed to investigate the role of pathophysiological remodeling in sustaining arrhythmia and to predict the arrhythmia circuits in arrhythmogenic right ventricular cardiomyopathy (ARVC) patients of different genotypes⁵⁷. This approach integrated the patient’s disease-induced structural remodeling and genotype-specific cellular electrophysiology properties and revealed that the underlying arrhythmia mechanisms differ among ARVC genotypes. In a retrospective study of 16 ARVC patients with two genotypes: plakophilin-2 (PKP2, n = 8) and gene-elusive (GE, n = 8), Zhang et al found that Geno-DT accurately and non-invasively predicted the ventricular tachycardia (VT) circuit locations for both genotypes with very high accuracy, sensitivity and specificity in both the for GE and PKP2 patient groups, when compared to VT circuit locations identified during clinical EP studies. Importantly, the results revealed that the underlying VT mechanisms differ among ARVC genotypes: in GE patients, fibrotic remodeling is the primary contributor to VT circuits, while in PKP2 patients, slowed conduction velocity and altered restitution properties of cardiac tissue, in addition to the structural substrate, are directly responsible for the formation of VT circuits. Fig. 6 shows an interesting result: when the genotype in the DTs was mismatched, the VT circuits could no longer be predicted correctly. With its incorporation of genetic EP information, Geno-DT is the latest development in heart DT applications. The Geno-DT approach demonstrated the potential to augment therapeutic precision in the clinical setting and lead to more personalized treatment strategies in ARVC.

Figure 6:

VT prediction using genotype-specific heart DT (Geno-DT) of ARVC patients. Reconstructed DTs from 2 patients of Gene Elusive and PKP2 genotypes along with the simulated VTs with A, genotype-matched condition, and B, genotype-mismatched condition. The mismatched condition led to incorrect prediction of the VT circuits. Modified from⁵⁷ (permission exempt per https://elifesciences.org/terms).

Like the management of AF, catheter ablation plays a major role in the contemporary management of VAs. Eliminating VA with ablation has achieved, however, modest success, 50–88%^58,59. Similar to AF ablation, a number of patients, for whom the initial procedure fails, are repeatedly ablated, further extending the adverse structural remodeling in the ventricles. Discovering new strategies that result in accurate identification of the optimal ablation targets for VA in patients with different heart diseases, and which also deliver long-term freedom from VA, is a quest of paramount clinical significance. Personalized DT technology has made major strides in improving ablation precision by providing noninvasive localization of ablation targets. A study employing DTs from 13 post-infarct patients who underwent ablation showed that ablation targets from DTs were consistent with the targets executed in the clinic⁶⁰. The landmark study by Prakosa et al.⁶¹ was demonstrated for the first time the clinical utility of personalized DTs in determining noninvasively the optimal (i.e., minimum lesion size) VT ablation targets and guiding the clinical procedure of VT ablation. The capability of the approach to was first assessed in a retrospective study (n = 21), where predicted targets we compared to the clinical. This included patients where image construction was carried out from clinical images with device artifacts. Furthermore, the feasibility of using DTs to guide clinical VT ablation was demonstrated in a proof-of-concept prospective study (n = 5), in two clinical centers (Fig.7A). This work highlighted the potential for DT technology to impact the clinical management of VAs. The sensitivity of DT ablation targets to EP parameter variability has also been assessed⁶². Using DTs based on the CT scans of patients has also been shown, in a retrospective cohort of 29 post-infarct patients, to be able to provide guidance in ventricular ablation⁶³. The DTs predicted not only the targets on index ablation, but also the ablation targets on a redo procedure several years later (Fig.7B). Overall, the predicted ablation targets by the heart DTs consistently encompassed much less lesion volumes. Since CT is accessible across a broad range of clinical centers, DTs could be readily deployed prospectively to improve ventricular ablation.

Figure 7:

DTs guiding ventricular ablation. A. Ablation in a prospective patient with ischemic cardiomyopathy guided by MRI-based DT. Shown are the two predicted VT circuits, and intra-procedural EAM (CARTO) with the predicted (purple) and actual (red) lesions. Modified from⁶¹ (Permission exempt per https://www.nature.com/nature-research/reprints-and-permissions/permissions-requests). B. Using a CT-based DT in a post-infarction patient with infiltrating fat to predict VT circuits (left; the inset presents detail of activation; white arrows denote reentrant pathway; zigzag arrows denote conduction channels) and ablation targets retrospectively, where predicted ablation targets and clinical ablation lesions co-localize in a patient who underwent redo ablation ≈4 y after the index procedure. Modified with permission from⁶³.

DTs have also been used to improve ablation by better understanding VT circuit morphology. Sung et al.⁶⁴ demonstrated that inclusion of repolarization gradients, both transmural and apico-basal, altered VT circuit morphologies, with minimal change of the ablation targets. DT simulations of VT circuits have also been combined with automated ECG-based localization algorithm to predict VT exit sites⁶⁵, highlighting a potential synergy between the two methodologies. A recent study further developed and validated a technique called reentry vulnerability index, demonstrating that the technique allows localization of ablation targets⁶⁶. In the clinic, multiple wavefront pacing (MWP) and decremental pacing (DP) are two EAM strategies that have emerged to characterize^58,59 the VT substrate and determine ablation targets. A recent DT study on 48 patients assessed how well MWP, DP, as well as other techniques used in clinical studies improve identification of electrophysiological abnormalities at critical VT sites⁶⁷. The study found that EAM with MWP is more advantageous for characterization of substrate for ablation in hearts with less remodeling.

Concluding Remarks

The developments and the examples of applications of heart DTs in arrhythmia management by our team presented in this review highlight the significant advancements that cardiac computational modeling has made in bringing such tools closer to the patient point of care. Heart DTs could potentially become a disruptive approach, fully embodying the expectations of precision medicine in cardiology, as these virtual tools leverage robust physics and physiology - based mechanistic insights, are capable of encoding patho-physiological complexity across multiple spatial scales, and can be continuously updated with the individual patient’s clinical and lifestyle data.

The pathway to accelerate the clinical impact of heart DTs is to continuously work on increasing the trust in the technology among researchers, clinicians, and healthcare professionals, to emphasize its benefits to patients, and to educate the society at large. An important aspect in this endeavor is to always recognize and account for the limitations of the technology. DTs of organs and patients will likely never represent all aspects of physiological reality. Thus focus should be steadily on DT performance and the resulting patient outcome, both becoming increasingly important. To cite George Box, “all models are wrong, some are useful” – it is this usefulness that we will strive to enhance in the years to come.

List of Abbreviations:

3D

three-dimensional

ACCF

American College of Cardiology Foundation

AF

atrial fibrillation

AHA

American Heart Association

APD

action potential duration

ARVC

arrhythmogenic right ventricular cardiomyopathy

CT

computed tomography

DP

decremental pacing

DT

digital twin

EAM

electroanatomic map

ECGI

electrocardiographic imaging

EP

electrophysiology

ESC

European Society of Cardiology

FDA

Food and Drug Administration

FIRM

focal impulse and rotor mapping

Geno-DT

genotype-specific heart digital twin

GE

gene-elusive

HCM

hypertrophic cardiomyopathy

ICD

implantable cardioverter-defibrillator

inFAT

penetrating adipose tissue

IIR

image intensity ratio

LA

left atria

LGE

late gadolinium enhancement

LVEF

left ventricular ejection fraction

MRI

magnetic resonance imaging

MWP

multiple wavefront pacing

OPTIMA

optimal target identification via modelling of arrhythmogenesis

PET

positron-emission tomography

PKP2

plakophilin-2

PsAF

persistent form of atrial fibrillation

PV

pulmonary vein

PVI

pulmonary vein isolation

RA

right atria

rToF

repaired tetralogy of fallot

SCD

sudden cardiac death

SSCAR

survival study of cardiac arrhythmia risk

VA

ventricular arrhythmia

VT

ventricular tachycardia